<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:itunes="http://www.itunes.com/dtds/podcast-1.0.dtd" xmlns:googleplay="http://www.google.com/schemas/play-podcasts/1.0"><channel><title><![CDATA[konstantin]]></title><description><![CDATA[CEO & Founder of Blockdaemon, the world's leading enabler of encrypted transactions for Financial Institutions, follow my journey on konstantin.ai]]></description><link>https://writing.konstantin.ai</link><image><url>https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png</url><title>konstantin</title><link>https://writing.konstantin.ai</link></image><generator>Substack</generator><lastBuildDate>Wed, 18 Nov 2026 19:23:12 GMT</lastBuildDate><atom:link href="https://writing.konstantin.ai/feed" rel="self" type="application/rss+xml"/><copyright><![CDATA[konstantin]]></copyright><language><![CDATA[en]]></language><webMaster><![CDATA[kkr1111@substack.com]]></webMaster><itunes:owner><itunes:email><![CDATA[kkr1111@substack.com]]></itunes:email><itunes:name><![CDATA[konstantin]]></itunes:name></itunes:owner><itunes:author><![CDATA[konstantin]]></itunes:author><googleplay:owner><![CDATA[kkr1111@substack.com]]></googleplay:owner><googleplay:email><![CDATA[kkr1111@substack.com]]></googleplay:email><googleplay:author><![CDATA[konstantin]]></googleplay:author><itunes:block><![CDATA[Yes]]></itunes:block><item><title><![CDATA[The Binance Hack and the God You Can’t Predict]]></title><description><![CDATA[$6 Billion in 20 Minutes: Faith Is Not a Security Model]]></description><link>https://writing.konstantin.ai/p/the-binance-hack-and-the-god-you</link><guid isPermaLink="false">https://writing.konstantin.ai/p/the-binance-hack-and-the-god-you</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Sat, 07 Nov 2026 18:00:30 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>On October 29, an attacker moved about $6.36 billion out of Binance wallets across ten blockchains, most of it in roughly 20 minutes. It&#8217;s the largest crypto theft on record, and TRM Labs attributes it, with high confidence, to North Korea&#8217;s TraderTraitor. <a href="https://www.trmlabs.com/resources/blog/%20north-korea-steals-usd-6-36-billion-from-binance-in-the-largest-hack-in-history">trmlabs</a></p><p>The number isn&#8217;t the lesson. The structure is.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>On Ethereum, the attacker used an EIP-7702 delegation to turn a hot wallet into a sweeper, with 50 pre-signed authorizations prepared in advance. On the other nine chains, Binance wallets simply sent funds through ordinary signed transfers. Different chains, key types and signing schemes, and valid signatures for the attacker on all of them within about half an hour. TRM&#8217;s read is that the signing infrastructure itself was compromised. Binance hadn&#8217;t published a root cause when I wrote this, so that&#8217;s an assessment, not a verdict. <a href="https://www.trmlabs.com/resources/blog/%20north-korea-steals-usd-6-36-billion-from-binance-in-the-largest-hack-in-history">trmlabstrmlabs</a></p><p>Read it as an engineer. A signing system that speaks for ten chains isn&#8217;t a quorum. It&#8217;s one very large node. And look at the 50 authorizations. The attacker couldn&#8217;t predict the wallet&#8217;s transaction count, so it didn&#8217;t try. It covered the whole range and let one land. That&#8217;s a human team already designing around unpredictability.</p><h3>Now add a swarm</h3><p>That was people, rehearsing for at least 36 hours. Picture the same operation run by agents. This is a projection, not a forecast. <a href="https://www.trmlabs.com/resources/blog/%20north-korea-steals-usd-6-36-billion-from-binance-in-the-largest-hack-in-history">trmlabs</a></p><p>Rehearsal becomes continuous and parallel across every chain and signing path. Probing gets cheap enough to run against every vendor, approval flow and employee at once. The defender&#8217;s window, about 37 minutes before the first big stablecoin freeze in this case, shrinks toward zero. And the attacker doesn&#8217;t need to be right, just to cover the space, the way those 50 authorizations did.</p><p>The defenders will have agents too. That&#8217;s where it gets uncomfortable.</p><h3>Gods are what we call systems we&#8217;ve given up predicting</h3><p>In 1936 Turing proved that no general program can inspect another program and certify what it will do. Rice&#8217;s theorem stretches that to almost any behavior you care about: will it leak, will it sign something it shouldn&#8217;t. The word &#8220;general&#8221; matters, because you can still verify specific, well-structured systems. What you can&#8217;t have is a universal inspector. Wolfram&#8217;s computational irreducibility adds the other half: for many systems, the only way to know what they&#8217;ll do is to run them.</p><p>A model is a huge, irreducible computation, and a swarm of them interacting is more so. You won&#8217;t predict what a swarm does, on offense or defense. Any security strategy that depends on spotting bad behavior in advance, whether by reading prompts, scanning code, or trusting a model to police a model, is betting on the inspector Turing ruled out.</p><p>We do this every time something gets too complex to predict. Storms had gods, markets have moods, and models now have intentions. &#8220;It decided.&#8221; &#8220;It&#8217;s lying.&#8221; Those are agent words for mechanisms we can&#8217;t forecast. The engineering response is the opposite of worship: stop betting on the component, and design the system so the component doesn&#8217;t matter.</p><p>The blockchain world already learned this. Satoshi published nine pages in 2008, the network went live in 2009, and by 2011 he was gone, leaving a system that runs without its creator. That isn&#8217;t a god you pray to. It&#8217;s a god you don&#8217;t need. Stellar gets there with overlapping quorum slices. Ethereum caps execution with gas because it can&#8217;t know whether a contract will finish. Gas is the halting problem&#8217;s workaround.</p><h3>Quantum is the same lesson on a slower clock</h3><p>Every key in this industry is an identity, and the math behind those keys is what a large enough quantum computer running Shor&#8217;s algorithm could break. That machine doesn&#8217;t exist today, and the target is signatures, not mining. Some of the oldest coins sit behind public keys that are already exposed, and their creator can&#8217;t sign the migration.</p><p>NIST finalized post-quantum signature standards in 2024. The tools exist. What takes years is moving an industry onto them, and nobody can tell you the date the threat becomes real. That&#8217;s the same problem again: you can&#8217;t predict the timeline, so you build for it not mattering. The answer is crypto-agility, keys and algorithms you can swap without rewriting the system. It&#8217;s boring work, and it lands on whoever controls signing.</p><h3>Why we moved from nodes to signatures</h3><p>Blockdaemon started by running as many nodes as possible. Our direction now is to sign as many transactions as possible.</p><p>That&#8217;s not a pivot away from infrastructure. It follows where the risk went. A node tells the network what&#8217;s true. A signature tells it what to do. Nodes make the chain trustworthy, and signing makes an institution&#8217;s authority trustworthy. Binance didn&#8217;t lose money because nodes lied. It lost it because something valid signed what it shouldn&#8217;t have.</p><p>Both pressures converge there. AI agents act through keys, so a swarm attacks signing. Quantum breaks signing, and the migration happens in signing. If the future is unpredictable components, the control point is who can move value, under what policy, with whose approval. Institutions will pay to get that right.</p><h3>What survives</h3><ul><li><p><strong>No single signer.</strong> A quorum of independent parties, not one system that speaks for everything. Our vault splits keys into shares on nodes the customer controls, and a transaction signs only when the required quorum approves.</p></li><li><p><strong>Independence that&#8217;s real.</strong> If your approvers run the same model, one exploit compromises all of them. Correlated failure is the biggest swarm risk, because it turns f=1 into f=n.</p></li><li><p><strong>Authority separated by chain and wallet tier.</strong> TRM&#8217;s own advice after the hack is to separate signing infrastructure and approval paths this way. <a href="https://www.trmlabs.com/resources/blog/%20north-korea-steals-usd-6-36-billion-from-binance-in-the-largest-hack-in-history">trmlabs</a></p></li><li><p><strong>Policy outside the model.</strong> The agent proposes, a deterministic rule set or a human approves. You can&#8217;t audit the mind. You can audit the record and constrain the permissions.</p></li><li><p><strong>A veto that isn&#8217;t the signer.</strong> Tether, Circle, Ripple and World Liberty Financial froze about $2.67 billion, because other parties with their own authority could say no after the signer said yes. The attacker still holds roughly $2.4 billion that no issuer can reach. <a href="https://www.trmlabs.com/resources/blog/%20north-korea-steals-usd-6-36-billion-from-binance-in-the-largest-hack-in-history">trmlabstrmlabs</a></p></li><li><p><strong>Revocation you&#8217;ve already tested,</strong> and keys you can migrate. Speed and algorithm change are both variables now.</p></li></ul><p>We&#8217;ve operated blockchain infrastructure since 2017, and this is its oldest lesson. Nodes fail, operators make mistakes, and something will do what nobody expected.</p><p>You don&#8217;t predict the node. You design the quorum.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Selling AI to the enterprise is SAP all over again]]></title><description><![CDATA[Pre-configs, harnesses, and engineers in the room: how to build an AI company enterprises actually buy from.]]></description><link>https://writing.konstantin.ai/p/selling-ai-to-the-enterprise-is-sap</link><guid isPermaLink="false">https://writing.konstantin.ai/p/selling-ai-to-the-enterprise-is-sap</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Tue, 20 Oct 2026 14:29:08 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>In the early 2000s, SAP didn&#8217;t win because the software was beautiful. It won because it shipped a pre-baked version of how a company should run (finance, procurement, HR) and then brought an army of implementers to bend it onto the customer&#8217;s actual mess. The product was the template. The money was in the deployment.</p><p>AI is heading to the same place, and fast. The model is the engine. Nobody buys an engine. They buy the thing that gets their invoices matched, their claims triaged, or their KYC queue cleared.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p><strong>The labs figured this out first.</strong> OpenAI and Anthropic both launched joint ventures for enterprise AI services in May, with OpenAI&#8217;s valued at $10 billion and Anthropic&#8217;s, built with Blackstone and Hellman &amp; Friedman, valued at $1.5 billion. Salesforce has publicly committed to building a team of 1,000 forward deployed engineers to drive Agentforce adoption. LinkedIn found forward deployed engineering roles grew 42-fold between 2023 and 2025. The companies that make the best models are building services arms. Take that seriously.</p><h2>Why the model alone doesn&#8217;t sell</h2><p>The demo is easy. Production is where deals die. Kyndryl found 57% of enterprises now have AI embedded in core processes, but only 11% have hit both of their top two AI objectives. Everyone has AI. Almost nobody has the result.</p><p>The gap is workflow, data access, permissions, and trust. A model doesn&#8217;t close that. A person sitting next to the customer&#8217;s ops team for six weeks does, if they have the right tooling in their bag.</p><h2>The company you&#8217;d build today</h2><p>Think of it as five layers.</p><p><strong>1. Models, rented.</strong> Don&#8217;t marry one. Vercel says the average enterprise on its AI Gateway routes across 17 models. Your customers will switch models every few months. Your product should make that a non-event.</p><p><strong>2. The harness.</strong> This is the real product. Agent runtime, tool connections, memory, evals, guardrails, identity, audit trail. Vercel is a good example of the substrate forming: its Agent Stack bundles the AI SDK, AI Gateway, Sandbox, Workflow SDK and Chat SDK, and its enterprise package puts internal agents behind the company&#8217;s own identity provider by default. Build or buy this, but own the experience.</p><p><strong>3. Pre-configs.</strong> This is your SAP best-practice template. A packaged agent for one workflow in one industry, with the connectors, evals, and policies already wired. &#8220;Reconcile settlements.&#8221; &#8220;Triage AML alerts.&#8221; &#8220;Review contracts against our playbook.&#8221; The more specific, the faster you close. Glean started with search and Agentforce started with support tickets. Narrow wedge first, platform later.</p><p><strong>4. Forward deployed engineers.</strong> They take the pre-config, plug it into the customer&#8217;s real systems, tune it on their data, and hand it off. AWS embeds them for about 45 days to build production-ready systems and make the customer self-sufficient. That&#8217;s the right shape: fixed scope, fast, and a clean handoff.</p><p><strong>5. Trust.</strong> SSO, audit logs, data residency, run-in-your-own-cloud. For regulated buyers this isn&#8217;t a feature, it&#8217;s the entry ticket. Boring work, and it wins deals.</p><h2>How to organize it</h2><ul><li><p><strong>Founder-led sales.</strong> You sell the outcome and the risk reduction, not the model.</p></li><li><p><strong>FDE pods</strong> that own a deployment end to end, with real authority. A solutions engineer relabeled as an FDE is just a relabel.</p></li><li><p><strong>A product team whose job is to eat the FDEs&#8217; work.</strong> Every deployment should leave behind a better pre-config. If it doesn&#8217;t, you&#8217;re a consultancy with a nice website.</p></li><li><p><strong>Security and compliance early.</strong> Not after the first bank asks.</p></li><li><p><strong>Pricing:</strong> paid, fixed-scope deployment to land, then usage or outcome-based to expand.</p></li></ul><h2>We&#8217;ve seen this movie at Blockdaemon</h2><p>None of this is new to us. We&#8217;ve lived a version of it in digital assets.</p><p>Institutions don&#8217;t want to run their critical infrastructure in someone else&#8217;s black box. A bank touching a blockchain wants it inside its own perimeter, under its own controls, wired into its own compliance stack. We learned that selling nodes, staking, and key infrastructure to regulated buyers. AI is hitting the same wall now.</p><p>Three things carried over directly:</p><p><strong>The control panel is the product.</strong> When infrastructure runs on-prem or in a customer&#8217;s own cloud, the buyer&#8217;s first question is &#8220;who can do what, and how do I see all of it?&#8221; For us that was a control panel: what&#8217;s running, who approved it, what the policy is. For AI it&#8217;s the same list: which agents exist, which models they call, what data they touch, what they cost, who signed off. Whoever owns that view owns the relationship.</p><p><strong>Compliance integration decides whether you ship.</strong> In digital assets, the software is never done until it plugs into the customer&#8217;s existing compliance and reporting stack. AI is identical. Audit logs, model risk review, data residency, human-in-the-loop approvals. Nobody in a regulated shop deploys an agent that can&#8217;t produce an audit trail. This is where deals stall or close.</p><p><strong>Pre-configured, then deployed by engineers.</strong> Same SAP pattern. A hardened, pre-configured stack, then engineers who fit it into the customer&#8217;s workflows and hand it off.</p><p>That&#8217;s why I describe Blockdaemon as the trust layer for institutional finance. The pattern is the same whether the thing being trusted is a validator or an agent.</p><h3>The AI box</h3><p>Here&#8217;s the opportunity. The buyers we already serve are the same buyers struggling to get AI past security review. They&#8217;ve approved us as a vendor, we know how to deploy inside their environment, and we speak compliance. So putting an AI box next to what we already sell is a natural extension.</p><p>We use Konstant at Blockdaemon for exactly this reason. Every institutional customer wants something specific, and that&#8217;s how you end up with a codebase per customer and a payroll to match. Konstant&#8217;s answer is that your company is ten to twenty parts in one box. It takes the software, the methods, and even one person&#8217;s judgment out of the stack you already run, and turns each into a named part with a defined input, output, and revision, so an agent can call it. Nothing gets rewritten or migrated. The parts run in a box you control, hosted by Konstant or in your own cloud.</p><p>Then you pick the number you want to move (cost per order, margin, tickets touched by hand) and the system tries parts in and out of the live work, measures whether each move got you closer, and rolls back what doesn&#8217;t, with a person judging every step.</p><p>If that sounds like the SAP playbook, it is. The named parts are the pre-configs, the box is the harness, and the engineers who fit it into the customer&#8217;s workflow are the FDEs. The difference is that it runs inside the customer&#8217;s perimeter, which is exactly where our buyers need it.</p><p>The market is moving the same direction. Vercel&#8217;s enterprise package lets internal apps and agents live behind a company&#8217;s identity provider by default, with the option to deploy inside a company&#8217;s own AWS account. When infrastructure players build &#8220;run it in my perimeter&#8221; into their pitch, the demand is clear.</p><h2>Who to study</h2><ul><li><p><strong>Palantir</strong> wrote the FDE playbook. Its AIP platform posted 133% year-over-year US commercial revenue growth in Q1 2026.</p></li><li><p><strong>Salesforce Agentforce</strong> is the incumbent doing the SAP move in real time: template plus FDEs plus an installed base.</p></li><li><p><strong>Sierra and Decagon</strong> show what a sharp wedge (customer service) looks like.</p></li><li><p><strong>Glean</strong> won with search before expanding to agents.</p></li><li><p><strong>Harvey</strong> is the vertical play: legal, deep, trusted.</p></li><li><p><strong>OpenAI&#8217;s Deployment Company and Anthropic&#8217;s Ode</strong> are the labs becoming the implementers.</p></li><li><p><strong>Vercel</strong> is the substrate for teams building and shipping their own agents inside the enterprise.</p></li></ul><h2>Where this goes wrong</h2><p>You become a body shop. Margins collapse, every customer is custom, nothing compounds. The fix is the loop in layer 3: deploy, extract, productize, repeat.</p><h2>The bet</h2><p>The labs are betting billions that deployment is the bottleneck. They&#8217;re probably right. If you&#8217;re building in this space, decide which workflow you&#8217;ll own, then ship the template that makes the next customer&#8217;s deployment half as long as the last one.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[What AI founders get wrong when they sell to enterprises]]></title><description><![CDATA[Most AI startups I meet have a great product and a bad plan for selling it to a large company.]]></description><link>https://writing.konstantin.ai/p/what-ai-founders-get-wrong-when-they</link><guid isPermaLink="false">https://writing.konstantin.ai/p/what-ai-founders-get-wrong-when-they</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Thu, 15 Oct 2026 15:44:35 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<h1>What AI founders get wrong when they sell to enterprises</h1><p>The product usually isn&#8217;t the problem. The mistakes are about how enterprises buy, and they&#8217;re predictable.</p><h2>1. You&#8217;re selling to the wrong person.</h2><p>The person who loves your demo is rarely the person who can sign. Innovation teams and &#8220;head of AI&#8221; titles are great at getting you a meeting and terrible at getting you a contract. Find out early who owns the budget, who owns risk, and who can say no. In regulated industries the last one is the one that matters. If you haven&#8217;t met them by month two, you don&#8217;t have a deal, you have a hobby.</p><h2>2. You treat security and compliance as a final step.</h2><p>Founders save the security questionnaire for the end. Big buyers start there. If you&#8217;re selling to a bank or an exchange, your data handling, audit trail, and vendor risk story are part of the product. Build them early. A mediocre model with a clean compliance story beats a brilliant one that makes the risk team nervous.</p><p>We rebuilt the Blockdaemon site around exactly this. The first thing a bank sees is on-prem deployment, key management, and the certifications that get you through vendor risk, not a feature list. That&#8217;s the order a risk team reads in, so that&#8217;s the order we sell in.</p><h2>3. You let the pilot run forever.</h2><p>Pilots are where AI deals go to die. No success criteria, no end date, no named person on the buyer side who owns the decision. Before you start, write down what success looks like, what happens if you hit it, and who signs. If the buyer won&#8217;t agree to that, the pilot is a free consulting project.</p><h2>4. You sell the technology instead of the outcome.</h2><p>Enterprise buyers don&#8217;t care that your model is better. They care that a process costs less, runs faster, or carries less risk. Every pitch should be one sentence about a number they already track. If you can&#8217;t name that number, you&#8217;re not ready for the meeting.</p><h2>5. You underestimate how slow it is, then run out of money.</h2><p>Large enterprise cycles take six to eighteen months, and AI adds extra steps because nobody has a clear policy yet. Plan your runway around the real cycle, not the one in your forecast. This is where investor management and sales collide: the founder who promised a fast ramp has to explain a slow one.</p><h2>6. You forget the investors are watching the same funnel.</h2><p>Your board sees &#8220;enterprise pipeline&#8221; and assumes revenue is coming. If you haven&#8217;t managed that expectation, a long cycle looks like failure. Tell them early what the sales process really looks like, which milestones matter (security review passed, pilot signed, champion identified), and when revenue shows up. Investors forgive slow. They don&#8217;t forgive surprised.</p><h2>What to do instead</h2><p>Pick one buyer type and go deep. Find the person who owns the risk and get them on your side early. Make the compliance story part of the product. Put an end date and a decision-maker on every pilot. And brief your investors on the real timeline before they ask.</p><p>None of this is complicated. It&#8217;s just not what you learn building a model.</p><p>If you want to see what selling to banks looks like from the inside, the new <a href="https://www.blockdaemon.com/institutional-stack">Blockdaemon site</a> is a good place to start. Look at what we lead with and what we leave for later.</p><p>If you&#8217;re an AI founder trying to sell into a large enterprise and something&#8217;s stuck, <a href="https://konstantin.ai/#advising">apply here</a>.</p>]]></content:encoded></item><item><title><![CDATA[The labs are slowing down. We should talk about what speeds up.]]></title><description><![CDATA[the binding constraint on adoption in regulated finance is governance, not capability]]></description><link>https://writing.konstantin.ai/p/the-labs-are-slowing-down-we-should</link><guid isPermaLink="false">https://writing.konstantin.ai/p/the-labs-are-slowing-down-we-should</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Mon, 14 Sep 2026 04:08:08 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>This weekend Dario Amodei published &#8220;We Must Pace the Frontier.&#8221; Within hours, Sam Altman agreed. Musk backed it. Hassabis called the direction correct. Whatever gets formalized this week, the signal is unambiguous: the people building the most capable AI systems in the world just told the market that capability has outrun verification.</p><p>I run infrastructure for regulated finance, and I&#8217;ve spent the past 18 months rewiring Blockdaemon itself to be AI-first. So I read the essay less as an AI story and more as a finance story. Here&#8217;s why.</p><p>The labs didn&#8217;t discover a new problem. They discovered the problem regulated finance has been organized around for a century: past a certain level of stakes, trust cannot be asserted. It has to be proven, continuously, by someone other than the party making the claim. That&#8217;s why clearinghouses exist. That&#8217;s why DTCC exists. The frontier labs just reinvented the audit function.</p><p>Look at the first concrete commitment. Not a compute cap. Embedded third-party evaluators with employee-level access, verifying that labs actually do what they say. Humans, checking. That&#8217;s the right move for model development. It does not scale to what&#8217;s coming next: agents acting at machine speed against real systems, and increasingly against real money. Treasury, payments, settlement, custody. No bank will let an autonomous agent touch value transfer because an evaluator reviewed the lab&#8217;s safety culture last quarter. They will require proof per action.</p><p>That layer is what we build. Every agent action running through Blockdaemon&#8217;s control plane gets evaluated against institution-defined policy, authorized or escalated, enforced through MPC threshold signing, and recorded as a signed, exportable attestation: which agent acted, under which policy, at what time. Not a log entry a vendor promises is accurate. A cryptographic artifact the institution itself can hand to a regulator. It sits on the same infrastructure that supports 400+ institutional customers and over $110B in digital assets, with zero key compromise events in six years. DTCC, NYSE, JPM, GS, Citi, Fidelity, RBC, SocGen and LSEG don&#8217;t work with us because we move fast. They work with us because everything we do is verifiable.</p><p>So my position on the slowdown: it isn&#8217;t a retreat. It&#8217;s the market repricing trust. For a decade the AI race rewarded velocity. From here, <strong>the binding constraint on adoption in regulated finance is governance, not capability</strong>. The models are already good enough for most institutional workflows. What&#8217;s missing is the execution layer that lets a risk committee, an auditor, and eventually an examiner say yes.</p><p>I&#8217;ve argued before that if AI hasn&#8217;t changed the workflow, it has changed nothing. The corollary now: if an agent&#8217;s action can&#8217;t be proven, it won&#8217;t be permitted. The labs are pacing the frontier so safety can catch up. Our job is to make sure that when institutions are ready to deploy, the trust layer is already there. It is.</p>]]></content:encoded></item><item><title><![CDATA[How Blockdaemon Is Rewiring the Company for AI ]]></title><description><![CDATA[How our operating system now looks.]]></description><link>https://writing.konstantin.ai/p/how-blockdaemon-is-rewiring-the-company</link><guid isPermaLink="false">https://writing.konstantin.ai/p/how-blockdaemon-is-rewiring-the-company</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Fri, 24 Jul 2026 09:13:22 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!xgzm!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><span>A year ago, most companies were still treating AI like a faster search box, a writing assistant, or a code copilot. Useful, yes. Transformational, not yet.</span></p><p><span>That phase is over.</span></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p><span>What matters now is not whether a model can produce an answer. What matters is whether a company can route intelligence into real work: into decisions, into execution, into systems with memory, permissions, accountability, and human judgment where it still matters.</span></p><p><span>That is the shift we have been making at Blockdaemon.</span></p><p><span>We have not approached AI as a novelty layer sitting on top of the company. We have approached it as infrastructure. The same way great companies once had to wire in cloud, payments, data, and security as foundational layers, we believe this generation will have to wire in intelligence the same way.</span></p><p><span>That sounds abstract until it becomes operational.</span></p><p><span>Inside Blockdaemon, AI is already changing how work moves. It is changing how context is surfaced, how workflows are designed, how decisions get made, and how execution happens across teams. The real gain is not that individuals can do isolated tasks faster. The real gain is that the company itself becomes more adaptable.</span></p><p><span>That is the prize.</span></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!xgzm!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!xgzm!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png 424w, https://substackcdn.com/image/fetch/$s_!xgzm!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png 848w, https://substackcdn.com/image/fetch/$s_!xgzm!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png 1272w, https://substackcdn.com/image/fetch/$s_!xgzm!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!xgzm!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png" width="1456" height="767" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/a6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:767,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!xgzm!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png 424w, https://substackcdn.com/image/fetch/$s_!xgzm!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png 848w, https://substackcdn.com/image/fetch/$s_!xgzm!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png 1272w, https://substackcdn.com/image/fetch/$s_!xgzm!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa6ba89a5-288d-4e34-b446-2fa029a7b73d_2048x1079.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p><em><span>&#8220;konstant.cloud is designed as an adaptive operating layer for people and AI.&#8221;</span></em></p><h2><strong><span>The mistake most companies are making</span></strong></h2><p><span>Most companies still think the AI question is: which model should we use?</span></p><p><span>That is a second-order question.</span></p><p><span>The first-order question is: where does intelligence belong in the operating system of the company?</span></p><p><span>Which workflows deserve the best models? Which tasks can run on cheaper intelligence? Which decisions need strong guardrails? Where should budget visibility sit? Which calls still require human judgment? How do you make sure the gain shows up in output rather than just in spend?</span></p><p><span>That is the real management problem now.</span></p><p><span>And once you see it that way, you stop talking about AI as a tool category and start talking about workflow design, control, memory, and leverage.</span></p><p><span>At Blockdaemon, we are remote-first by default. We document aggressively because async work demands it. Our teams already operate across complex systems, approvals, and high-consequence environments. Those are not side details. They are exactly the conditions under which well-wired AI works best.</span></p><p><span>That is why crypto-native companies, and infrastructure companies in particular, may have a structural advantage here. If your company already runs on strong documentation, clear systems, and technical depth, AI has something real to plug into.</span></p><p><span>If it does not, the model may still look impressive in a demo, but the company does not get transformed.</span></p><h2><strong><span>What has actually worked for us</span></strong></h2><p><span>The best AI results at Blockdaemon have not come from generic prompting.</span></p><p><span>They have come from well-wired intelligence.</span></p><p><span>That means models connected to real context. Real documentation. Real systems. Real permissions. Real workflows.</span></p><p><span>Publicly, one visible example of that philosophy is our support for the Model Context Protocol in Institutional Vault. Developers can connect tools like Claude Desktop, Cursor, and Windsurf directly to documentation, APIs, and live account context. That changes the quality of work immediately. The model is no longer reasoning in the dark. It is operating with access to the environment it needs to be useful.</span></p><p><span>That same principle scales internally.</span></p><p><span>The future is not one giant all-knowing chatbot. The future is intelligence routed into the right layer of work with the right controls around it.</span></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!aFMd!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!aFMd!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png 424w, https://substackcdn.com/image/fetch/$s_!aFMd!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png 848w, https://substackcdn.com/image/fetch/$s_!aFMd!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png 1272w, https://substackcdn.com/image/fetch/$s_!aFMd!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!aFMd!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png" width="1213" height="891" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/a2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:891,&quot;width&quot;:1213,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!aFMd!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png 424w, https://substackcdn.com/image/fetch/$s_!aFMd!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png 848w, https://substackcdn.com/image/fetch/$s_!aFMd!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png 1272w, https://substackcdn.com/image/fetch/$s_!aFMd!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2989ecb-93b9-4a02-8677-07abf3b7f8b8_1213x891.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p><em><span>&#8220;AI becomes materially more useful when it can operate against live documentation, APIs, and approved system context.&#8221;</span></em></p><h2><strong><span>Why we co-developed konstant.cloud</span></strong></h2><p><span>We co-developed konstant.cloud because we did not want another chat interface.</span></p><p><span>We wanted a coordination layer.</span></p><p><span>AI is speeding up individuals everywhere. But faster individuals inside a rigid company do not automatically create a more adaptive organization. In many cases, they just create more output, more noise, and more fragmentation.</span></p><p><span>The harder problem is coordination.</span></p><p><span>How does context move up the company without losing fidelity? How do decisions move down without requiring endless meetings? How does execution move sideways across org charts, calendars, documents, threads, and tools without being trapped in somebody&#8217;s head?</span></p><p><span>That is the problem konstant.cloud is built to solve.</span></p><p><span>Its public framing is direct: it is an adaptive operating layer for people and AI, connecting people, agents, decisions, and capacity in one coordination layer that adapts as things change, without requiring a reorg. That is exactly the right ambition.</span></p><p><span>What makes the platform interesting is that it does not start with prompts. It starts with organizational memory and execution.</span></p><p><span>Documents, meetings, messages, CRM records, and approvals become structured memory. Active work lives in workfields. Repeatable operating moves can be packaged and reused. Execution runs through approved paths. The output of each cycle returns to memory, so the next cycle starts from truth instead of starting over.</span></p><p><span>That is not &#8220;AI tooling&#8221; in the shallow sense. That is the beginnings of AI-native company infrastructure.</span></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!gXOM!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!gXOM!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png 424w, https://substackcdn.com/image/fetch/$s_!gXOM!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png 848w, https://substackcdn.com/image/fetch/$s_!gXOM!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png 1272w, https://substackcdn.com/image/fetch/$s_!gXOM!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!gXOM!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png" width="1456" height="923" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:923,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!gXOM!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png 424w, https://substackcdn.com/image/fetch/$s_!gXOM!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png 848w, https://substackcdn.com/image/fetch/$s_!gXOM!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png 1272w, https://substackcdn.com/image/fetch/$s_!gXOM!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3e8cae42-317e-48aa-aa3c-12124ee386a6_2048x1298.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p><em><span>&#8220;konstant.cloud is structured around organizational memory, active work, repeatable execution, and compounding proof.&#8221;</span></em></p><h2><strong><span>Gary, memory, and the shape of an AI-native company</span></strong></h2><p><span>One detail I particularly like is that Gary sits on top of this layer as a named operator.</span></p><p><span>That may sound cosmetic. It is not.</span></p><p><span>Systems get adopted when they become usable in day-to-day work. When they feel operational. When teams know where to go to ask, route, decide, or act. The interface matters because culture matters, and the most powerful infrastructure usually wins by becoming simple enough to use every day.</span></p><p><span>Underneath that interface, the deeper point is memory.</span></p><p><span>Companies waste extraordinary amounts of energy re-explaining context, re-litigating decisions, and rebuilding workflows that were already solved once. Most organizational drag is not caused by lack of effort. It is caused by broken memory and broken coordination.</span></p><p><span>If AI is going to matter at the company level, it has to fix that.</span></p><p><span>That is why systems like konstant.cloud matter more than a standalone assistant does. They give intelligence somewhere durable to live.</span></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!GFTq!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!GFTq!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png 424w, https://substackcdn.com/image/fetch/$s_!GFTq!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png 848w, https://substackcdn.com/image/fetch/$s_!GFTq!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png 1272w, https://substackcdn.com/image/fetch/$s_!GFTq!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!GFTq!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png" width="1456" height="904" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/dddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:904,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!GFTq!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png 424w, https://substackcdn.com/image/fetch/$s_!GFTq!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png 848w, https://substackcdn.com/image/fetch/$s_!GFTq!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png 1272w, https://substackcdn.com/image/fetch/$s_!GFTq!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdddf00e2-1117-48b1-825e-221d363fc627_2048x1272.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p><em><span>&#8220;Gary is the operating interface on top of the coordination layer.&#8221;</span></em></p><h2><strong><span>Trust is part of the product</span></strong></h2><p><span>There is another reason this matters, especially for enterprise customers and investors.</span></p><p><span>AI adoption does not stall because models are weak. It stalls because trust is weak.</span></p><p><span>Who can see the data? Where does the context come from? What gets retained? What gets executed? What gets approved? What gets audited? Does your data become somebody else&#8217;s training corpus?</span></p><p><span>Konstant&#8217;s public materials are clear on this point: customer data is used to provide the service to that organization, is not sold, and does not train public AI models. That is not a marketing footnote. It is part of the architecture required for enterprise adoption.</span></p><p><span>The same logic shows up in what we are building more broadly.</span></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!yK19!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!yK19!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png 424w, https://substackcdn.com/image/fetch/$s_!yK19!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png 848w, https://substackcdn.com/image/fetch/$s_!yK19!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png 1272w, https://substackcdn.com/image/fetch/$s_!yK19!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!yK19!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png" width="1456" height="838" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:838,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!yK19!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png 424w, https://substackcdn.com/image/fetch/$s_!yK19!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png 848w, https://substackcdn.com/image/fetch/$s_!yK19!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png 1272w, https://substackcdn.com/image/fetch/$s_!yK19!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0ab68425-545b-4416-9770-62e1ad870da8_1922x1106.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p><em><span>&#8220;Enterprise AI only scales when data handling, permissions, and trust are part of the system design.&#8221;</span></em></p><h2><strong><span>Why this connects directly to Blockdaemon AI</span></strong></h2><p><span>Our internal AI evolution and our external product evolution are not separate stories.</span></p><p><span>They are the same story told at two levels.</span></p><p><span>Internally, we are learning what it takes to run a company with better memory, better routing, stronger leverage, and human judgment held where it matters.</span></p><p><span>Externally, we are building Blockdaemon AI, our control plane for institutional AI.</span></p><p><span>The market is beginning to understand something important: model intelligence is advancing quickly, but regulated execution is a different problem entirely. An agent can reason. It can recommend. It can plan. But when real money, regulated workflows, approvals, and auditability enter the picture, intelligence alone is not enough.</span></p><p><span>You need policy enforcement. You need agent identity. You need secure execution. You need proof.</span></p><p><span>That is why we believe the next category will not be won by whoever has the most theatrical demo. It will be won by whoever can make AI operable inside real systems of trust.</span></p><p><span>For institutions, that means governance. For companies, that means coordination. In both cases, the lesson is the same: the future belongs to organizations that can connect intelligence to execution without losing control.</span></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!R0Oy!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!R0Oy!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png 424w, https://substackcdn.com/image/fetch/$s_!R0Oy!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png 848w, https://substackcdn.com/image/fetch/$s_!R0Oy!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png 1272w, https://substackcdn.com/image/fetch/$s_!R0Oy!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png 1456w" sizes="auto, 100vw"><img src="https://substackcdn.com/image/fetch/$s_!R0Oy!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png" width="1456" height="665" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/e4446537-692e-4e31-b329-385e14f05cf6_2048x935.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:665,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!R0Oy!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png 424w, https://substackcdn.com/image/fetch/$s_!R0Oy!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png 848w, https://substackcdn.com/image/fetch/$s_!R0Oy!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png 1272w, https://substackcdn.com/image/fetch/$s_!R0Oy!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4446537-692e-4e31-b329-385e14f05cf6_2048x935.png 1456w" sizes="auto, 100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p><em><span>&#8220;Blockdaemon AI extends the same philosophy outward: intelligence is only valuable when it can operate inside trusted execution.&#8221;</span></em></p><h2><strong><span>Where this is going</span></strong></h2><p><span>We are still early.</span></p><p><span>But one thing already feels obvious to me.</span></p><p><span>The companies that win in the AI era will not be the ones that simply adopt the best models. They will be the ones that redesign themselves around them. They will treat intelligence as infrastructure. They will build better memory. Better routing. Better controls. Better execution. Better proof.</span></p><p><span>That is the work we are doing at Blockdaemon.</span></p><p><span>Not because AI is fashionable.</span></p><p><span>Because it is changing the economics of how companies build, decide, and operate &#8212; and because the same shift is about to reshape how value moves across digital systems as well.</span></p><p><span>That is the deeper opportunity in front of us.</span></p><p><span>And we intend to build for it.</span></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Most AI Apps Are Just Workflow Taxes]]></title><description><![CDATA[Why the next enterprise winners will remove broken workflows instead of monetizing them.]]></description><link>https://writing.konstantin.ai/p/most-ai-apps-are-just-workflow-taxes</link><guid isPermaLink="false">https://writing.konstantin.ai/p/most-ai-apps-are-just-workflow-taxes</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Tue, 23 Jun 2026 02:44:29 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>The market is moving fast, but the pattern is already visible. A large share of enterprise AI apps are not rewriting work. They are charging companies to navigate the mess those companies never fixed in the first place.</p><p>That is the frame I keep coming back to when I look at the current crop of AI software. The demos are good. The interfaces are clean. The summaries are faster. Search feels smarter. But the deeper question is still the only one that matters: does this product change the economics of execution inside the company, or does it simply make a broken workflow feel more tolerable?</p><p>I do not think that distinction is academic. It is the difference between a durable enterprise platform and an expensive convenience layer. It is also the difference between real enterprise value and a very temporary AI premium.</p><h2>The wrong market question</h2><p>Most people still ask AI app questions the way the SaaS market trained them to ask software questions. How many seats can it sell? How fast is adoption? Does usage expand across departments? Does the product become part of the daily routine? Those are not irrelevant questions, but on their own they miss the real test.</p><p>Enterprise AI does not earn durable value because people use it a lot. It earns durable value when it reduces coordination cost, compresses cycle time, improves decision quality, lowers exception handling overhead, and turns hidden organizational work into machine-readable process. If an AI app gets heavily used but the company still runs on the same broken approvals, the same fragmented handoffs, the same private side channels, and the same memory-based operating model, the company has not upgraded its operating system. It has added a nicer interface to organizational debt.</p><p>That is why I am increasingly skeptical of categories that get priced like platforms but behave like overlays. In a lot of cases, the product is not fixing the underlying operating model. It is charging rent on top of it.</p><h2>Why search is useful and still not enough</h2><p>Take enterprise search. I understand why it is attractive. Knowledge inside companies is fragmented, badly tagged, politically distributed, and usually trapped across too many tools. If an AI layer helps people find information faster, that is real utility. Nobody serious would dismiss that.</p><p>But utility is not the same thing as transformation. Search is often what a company buys when it has accepted that the workflow upstream is already too messy to fix quickly. Instead of reducing the number of places where truth can hide, it builds a more elegant way to search across the hiding places.</p><p>That is why I think Glean is a useful negative example. Not because it is a bad product. It is clearly not. It is a useful product in a real category. The issue is what the product reveals about the state of the market. If the center of gravity in enterprise AI becomes premium-priced retrieval across corporate sprawl, then the market is still rewarding navigation more than redesign.</p><p>That is a tell. It means companies are paying to search the maze instead of removing walls from the maze. It means the AI layer is monetizing fragmentation rather than eliminating it. It means the budget is flowing first to software that helps people ask, &#8220;Where is the answer?&#8221; rather than to software that makes the answer appear where work is actually being done.</p><p>That is a meaningful distinction. A search product can be sticky, useful, and commercially successful. It can also still be the wrong end state.</p><h2>The better test: does the app sit inside the workflow?</h2><p>The higher-value category is not &#8220;AI that helps you find things.&#8221; It is AI that helps the company operate with less friction because it sits inside the workflow itself. That means the software is not merely observing work after the fact. It is participating in routing, drafting, checking, enforcing policy, escalating exceptions, preserving state, and preparing decisions while the work is moving.</p><p>This is where my view gets sharper. I am much less interested in whether an AI app has a clever chat experience than whether it is wired into the actual decision path. Does it touch the transaction? Does it understand permissions? Can it operate under policy constraints? Does it know when to stop, when to escalate, and when a human sign-off is required? Can it learn from workflow exhaust instead of forcing users to restate context every time?</p><p>If the answer is no, then in many cases the company has not bought an operating layer. It has bought an intelligent sidebar.</p><h2>What we learned co-designing Konstant.cloud</h2><p>One reason I feel strongly about this is that we have lived the harder version of the problem at Blockdaemon. In the work around Konstant.cloud, the interesting challenge was never whether a large model could generate fluent text. That part was table stakes remarkably quickly. The real challenge was whether AI could be made useful inside a company without creating chaos.</p><p>That forced a different design question. Not &#8220;what can the model answer?&#8221; but &#8220;what is the model allowed to see, what is it allowed to do, under which policy boundaries, with which system hooks, and with whose authority?&#8221; Once you ask the question that way, the product stops looking like a chatbot project and starts looking much more like enterprise infrastructure.</p><p>That also changes how value gets built. A real enterprise AI system needs structured context, permissions, identity, auditability, policy enforcement, and workflow placement. It needs to know which system holds truth. It needs to understand when a suggestion is enough and when an action path has opened. It needs to translate messy human discussion into machine-readable state. It needs to make the workflow more legible as it operates, not merely summarize the workflow after the meeting is over.</p><p>That is why I think the most interesting AI products will be co-designed with the workflow owners themselves. They will not be generic assistants dropped onto the side of the business. They will be systems that emerge from the actual operating constraints of a company: approvals, controls, exceptions, risk boundaries, resourcing limits, and the ugly reality of how work really moves.</p><p>That is also why the best AI products can look less glamorous in a demo and much more powerful in production. They do not just answer questions elegantly. They remove an approval step, surface the exception earlier, draft the brief from live context, preserve decision memory, and route the issue with the right permissions attached. In other words, they change the shape of work.</p><h2>A playbook for valuing enterprise AI apps</h2><p>If I were evaluating enterprise AI apps today, I would use a much harsher playbook than the market often does. Not because the category lacks promise, but because the category is still full of products that look bigger than the operating leverage they actually create.</p><p>Here is the playbook I would use.</p><h3>1. Measure workflow displacement, not feature delight</h3><p>The first question is simple: what piece of work disappears because this product exists? Not what becomes nicer. Not what becomes faster to look up. What actually leaves the system? A manual routing step? A coordination layer? A recurring briefing ritual? An approval chase? A hand-built summary? If the answer is vague, the value is probably thinner than the demo suggests.</p><h3>2. Check whether the product owns a point of action</h3><p>Products are more defensible when they sit where a decision, approval, transaction, or exception actually occurs. If the app lives one screen away from the action, it is easier to replace. If it is embedded where the work changes state, it becomes part of the operating path. That is where enterprise value starts to harden.</p><h3>3. Test policy depth, not just model fluency</h3><p>Enterprise AI gets real when permissions, logging, escalation, and constitutional boundaries are part of the product. A system that sounds intelligent but cannot respect policy is a demo with a procurement problem. A system that knows who can see what, who can approve what, and which actions require human review is much closer to infrastructure.</p><h3>4. Look for compounding context, not repeated prompting</h3><p>The better systems get stronger as they absorb the company&#8217;s workflow exhaust: tickets, specs, decisions, approvals, incidents, and transaction history. Weak systems require the user to restage the scene every time. Strong systems accumulate operational memory and make the next action cheaper.</p><h3>5. Separate seat expansion from operating leverage</h3><p>Seat growth can make a company look healthy while hiding a weaker truth. If adoption expands mainly because more employees want access to a helpful assistant, the app may be a productivity purchase. If adoption expands because the workflow itself now routes through the product, the app may be becoming part of the company&#8217;s operating system. Those are very different kinds of value.</p><h3>6. Ask who gets weaker if the product is removed</h3><p>This is my favorite test because it is brutally honest. If the product vanished tomorrow, who would feel real pain? A few knowledge workers who liked the search box? Or the operating path itself? Would cycle times rise, controls weaken, decisions get slower, exceptions hide longer, and management visibility degrade? That answer tells you whether the product is adjacent to the work or inside it.</p><h2>Where the market will get repriced</h2><p>My bias is that the market will eventually get tougher on AI apps that charge like systems of record but behave like systems of assistance. The novelty premium is real right now. So is the confusion. A lot of products are being evaluated in the warm glow of model fluency rather than the colder light of workflow economics.</p><p>That will not last forever. Buyers will get stricter. Boards will ask harder ROI questions. Procurement teams will care less about the magic trick and more about the operating consequence. And the products that survive that transition will usually have a few traits in common: they are embedded in real workflows, they carry policy and permission logic, they accumulate useful memory, and they reduce the need for organizational heroics.</p><p>That is where I think the next serious enterprise winners live.</p><p>Not in AI that helps companies live with broken workflows a little more elegantly. In AI that helps them remove the breakage.</p><h2>The real allocation question</h2><p>So the question is not whether enterprise AI apps are exciting. Plenty of them are. The question is whether they are being valued as utilities, overlays, tollbooths, operating layers, or real infrastructure. Too many conversations collapse those categories into one.</p><p>I would keep them separate.</p><p>A useful AI app that helps people search the mess may deserve a good business. A workflow-native AI system that changes how the company routes decisions, applies policy, manages exceptions, and preserves state deserves a very different one.</p><p>That is the distinction I expect to matter more over time. The winners will not simply look intelligent. They will make the enterprise itself more legible, more governable, and more executable.</p><p>That is a much higher bar.</p><p></p>]]></content:encoded></item><item><title><![CDATA[If AI Has Not Changed the Workflow, It Has Changed Nothing]]></title><description><![CDATA[Most AI programs do not transform the business. They decorate dysfunction.]]></description><link>https://writing.konstantin.ai/p/if-ai-has-not-changed-the-workflow</link><guid isPermaLink="false">https://writing.konstantin.ai/p/if-ai-has-not-changed-the-workflow</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Thu, 18 Jun 2026 16:57:49 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Most corporate AI programs are an expensive way to avoid management.</p><p>Companies buy licenses, run pilots, celebrate prompt literacy, and keep the same broken workflow underneath. The meetings are still bloated. The approvals are still hidden. The backlog is still stale. The handoffs are still political. Now the company just has better notes about its own confusion.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>That is not transformation. It is cost inflation with a chatbot attached.</p><p>AI gets strategic only when it changes how work moves.</p><p>Not how fast people write emails. Not how nicely a meeting gets summarized. Not how many copilots appear in an org chart. How work moves. How decisions get made. How exceptions are handled. How policy is enforced. How state becomes visible. How memory stops living inside a few exhausted operators.</p><p>That is the real opening.</p><p>And it is why the wrong place to start is the model.</p><p>The right place to start is the workflow.</p><h2><strong>1. Pick one workflow worth rewiring</strong></h2><p>Do not begin with &#8220;AI transformation.&#8221; That phrase is usually a refuge for people who do not want to choose.</p><p>Pick one workflow that matters enough to create discomfort if it stays broken.</p><p>A good candidate crosses teams, depends on context, contains repeated handoffs, suffers when ownership is vague, and improves a meaningful business outcome when it gets faster.</p><p>The trap is picking something easy to demo and irrelevant to performance. That is how an AI program becomes internal theater.</p><h2><strong>2. Map the workflow people actually use</strong></h2><p>Every company has an official process. It is usually neat, optimistic, and partially fictional.</p><p>The real workflow is messier. It leaves the ticketing system. It jumps into Slack. It disappears into a meeting. It comes back as a decision nobody recorded properly. It picks up an exception that only one person knows how to handle.</p><p>At Blockdaemon, one of the useful lessons in our own AI work was that the hard problem was never raw model capability. The hard problem was making enough of the workflow legible that AI could participate without making a mess. That meant exposing context, cleaning up handoffs, and being honest about where the live state of work actually lived.</p><p>If you automate the diagram instead of the reality, you get a faster version of the lie.</p><h2><strong>3. Declare where truth lives</strong></h2><p>If status lives in meetings, approvals live in DMs, requirements live across five documents, and the backlog gets updated after the real decisions have already happened, then AI has nothing durable to work with.</p><p>It can summarize the fog. It cannot run the company.</p><p>So make the call. Which system tracks active work? Where are decisions written down? Where do exceptions go? Where does policy live? Where is historical context preserved in a form both humans and machines can use?</p><p>At Blockdaemon, this showed up in a painfully practical way around Jira hygiene and planning. Teams discussed using AI to flag stale tickets, overdue milestones, and backlog drift before planning and scrum-of-scrums because manual filtering was still happening by hand. That is not a model problem. That is a legibility problem.</p><p>A company that cannot tell AI where truth lives usually cannot tell its own employees either.</p><h2><strong>4. Write the constitution before the automation</strong></h2><p>This is where most AI programs lose their nerve.</p><p>They see friction in a workflow, they see a model that looks capable, and they jump straight to automation. Then everybody acts shocked when the first real permission boundary appears.</p><p>Every serious AI workflow needs a constitutional layer.</p><p>Not as branding. As operating law.</p><p>The constitutional layer answers a few blunt questions:</p><ul><li><p>What can the model read?</p></li><li><p>What can it write?</p></li><li><p>What can it recommend?</p></li><li><p>What can it execute?</p></li><li><p>What requires human approval?</p></li><li><p>What must be logged?</p></li><li><p>Which permissions are in scope?</p></li><li><p>What happens when confidence is low or context is incomplete?</p></li></ul><p>A useful distinction is to separate three classes of AI behavior:</p><ol><li><p><strong>Assistant tooling</strong>: the human stays fully in the loop.</p></li><li><p><strong>Supervised automation</strong>: the machine prepares, routes, or proposes, but a human approves the consequential step.</p></li><li><p><strong>Autonomous agents</strong>: the system acts without per-step human approval inside a narrow, explicit boundary.</p></li></ol><p>If those three modes blur together, the organization is not running an AI program. It is running a trust-destruction program.</p><p>At Blockdaemon, AI policy discussions moved in exactly this direction: sandboxing, documentation of internal AI use, approval visibility, and tighter control over unmanaged experimentation. Shadow AI was already happening. Client questions were already coming. The choice was not whether AI would enter the company. The choice was whether it would enter through a gate or through a crack.</p><h2><strong>5. Put AI at the coordination tax</strong></h2><p>Most AI teams point the model at the visible task.</p><p>The higher-leverage move is to point it at the invisible tax around the task.</p><p>That means:</p><ul><li><p>turning messy discussion into a decision-ready brief</p></li><li><p>extracting risks and assumptions from requirements</p></li><li><p>flagging stale tickets and broken dependencies</p></li><li><p>routing issues to the right owner</p></li><li><p>surfacing current state across fragmented systems</p></li><li><p>preserving historical memory so the company stops re-litigating old ground</p></li><li><p>translating policy into usable guidance at the point of action</p></li></ul><p>That is where the compounding starts. Not when people write faster. When the company spends less energy reconstructing itself.</p><p>At Blockdaemon, teams wanted AI to run repetitive Jira hygiene checks before meetings because the human version was manual, subjective, and easy to neglect. Once that work becomes machine-readable, AI stops being a novelty and starts acting like operating infrastructure.</p><h2><strong>6. Keep the security boundary where it matters</strong></h2><p>This is the part that separates adult AI design from demo culture.</p><p>At Blockdaemon, one concrete debate sat inside the transaction verification workflow. The seductive story was simple: AI can generate decoder logic fast, so wire it in and move on.</p><p>The real question was harder.</p><p>Where does decoding belong? Which chains are in scope? What latency increase is tolerable for an institutional workflow? Which boundary preserves trust? Which implementation leaks security responsibility into the wrong layer?</p><p>The answer was not &#8220;let the smartest model decide.&#8221; The answer was architectural.</p><p>The team pushed decoding and validation toward the secure gateway boundary rather than letting that logic drift into weaker middleware or client-side paths. They also narrowed early scope to the chains that mattered most and treated the public SDK path as separate from the secure validation path. In one proof-of-concept discussion, the team accepted roughly a 13% response-time increase because keeping validation inside the secure boundary mattered more than shaving a few milliseconds off the path.</p><p>That is what real AI operating judgment looks like.</p><h2><strong>7. Rewire the management loop, not just the task</strong></h2><p>A lot of AI work goes after individual productivity and leaves the management system untouched.</p><p>That is why the gains cap out early.</p><p>The bigger opportunity is to cut the coordination tax around management itself: fewer status reconstruction exercises, fewer meetings that exist only because the system is stale, fewer hidden blockers, fewer managers acting as human API gateways between teams.</p><p>At Blockdaemon, that showed up in a blunt push toward more asynchronous operating discipline. Leaders pushed teams to surface blockers in Slack and move state into shared written threads rather than waiting for the next scheduled standup. The point was not to create more chatter. The point was to stop forcing the organization to wait for the calendar before it could act.</p><p>If the workflow is written, explicit, and structured, management gets easier and machine participation gets safer. Same move. Same payoff.</p><h2><strong>8. Put cost discipline into the design</strong></h2><p>AI efficiency is itself a strategic choice.</p><p>A model that burns more in credits than it removes in coordination cost is not transformation. It is a new expense line wearing the costume of innovation.</p><p>That is why constitutional discipline has to include cost discipline. Which workflows justify expensive inference? Which calls need frontier models? Which tasks can use smaller, cheaper systems? Where does autonomy create enough value to pay for itself?</p><p>At Blockdaemon, internal policy discussions made this explicit: AI spend cannot float free from the economics of the work. That is the right instinct. A serious operating model cares about margin, not just model capability.</p><h2><strong>9. Run the first cycle in 30 days</strong></h2><p>The goal of the first month is not elegance. It is proof.</p><p>A practical first cycle looks like this:</p><p><strong>Days 1-5:</strong> reconstruct the real workflow from tickets, docs, approvals, meetings, support logs, and system traces.</p><p><strong>Days 6-10:</strong> define systems of truth, approval boundaries, permissions, logging rules, and escalation paths.</p><p><strong>Days 11-20:</strong> place AI in the highest-friction parts of the flow and keep the human override explicit.</p><p><strong>Days 21-30:</strong> measure what changed, cut what adds noise, tighten the workflow, and write down the new operating rule.</p><p>If cycle time, rework, escalation volume, and status reconstruction do not improve, you did not rewire the workflow. You installed a feature.</p><p>Features impress committees. Rewired workflows change companies.</p><h2><strong>10. The CEO job is to sponsor legibility</strong></h2><p>AI rewiring is not mainly a tooling problem. It is a management problem with software attached.</p><p>Someone has to decide which workflows matter, which systems count as truth, which permissions are acceptable, where automation stops, who owns the rewrite, and what trade-offs the company is willing to make to become easier to run.</p><p>That is leadership work.</p><p>AI exposes the company to itself. It reveals where ownership is fake, where policy is implied rather than written, where status is reconstructed socially, where systems disagree, and where a few strong operators are carrying too much invisible coordination burden.</p><p>That is why the upside is large. It is also why weakly led AI programs keep collapsing into demos.</p><h2><strong>The real prize</strong></h2><p>The real prize is not better content throughput.</p><p>It is a company that becomes more legible, more responsive, less political in its handoffs, less dependent on memory, and harder to knock off course when the operating tempo rises.</p><p>That is when AI stops looking like a productivity accessory and starts looking like infrastructure.</p><p>The winners in this phase will not be the companies with the most pilots.</p><p>They will be the ones that pick one important workflow, make it visible, govern it properly, wire intelligence into it, measure what changed, and repeat until the operating model itself gets stronger.</p><p>If AI has not changed the workflow, it has changed nothing.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Business Process Re-Engineering Took 180 Days. AI Can Rewire a Company in 30.]]></title><description><![CDATA[The original reengineering instinct was right. The cycle time was fatal. AI changes that.]]></description><link>https://writing.konstantin.ai/p/business-process-re-engineering-took</link><guid isPermaLink="false">https://writing.konstantin.ai/p/business-process-re-engineering-took</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Mon, 08 Jun 2026 22:41:47 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<blockquote><p><em>AI is not a content layer sitting beside the business. It becomes useful when it becomes an operating layer inside the business.</em></p></blockquote><p>Most companies are using AI to make broken workflows look busy.</p><p>They generate better emails, cleaner notes, sharper summaries, and faster slide decks. None of that is useless. None of it is transformation either. If the operating model underneath is slow, fragmented, political, and dependent on memory, AI does not fix it. It accelerates the mess.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>I have seen this movie before. In the 1990s, business process reengineering promised that if you redesigned work, performance would follow. The instinct was right. The workflow mattered more than the org chart. The handoff mattered more than the department. The bottleneck mattered more than the title. But the execution cycle was slow. A serious BPR effort often took 90 to 180 days: interviews, workshops, process maps, redesign sessions, training plans, compliance reviews, rollout decks. By the time the new process reached the business, the business had already changed.</p><p>That is why so many reengineering efforts felt heavy and stale. They treated the company like a machine you could stop, open up, rebuild, and start again. Real companies do not work like that. They keep moving. They route around the diagram. They invent side channels. They adapt in private, then surprise the official process later.</p><p>AI changes the economics of this completely.</p><p>For the first time, a company can watch work as it happens, reconstruct the real workflow from its digital exhaust, encode rules and policies closer to the point of action, and revise the operating model in weeks rather than quarters. The old BPR cycle that took 90 to 180 days can compress into 30.</p><p>But only if the company makes its work legible.</p><p>That is the dividing line. AI does not create clarity out of thin air. It needs written decisions, structured data, visible handoffs, named owners, accessible system context, and rules that exist somewhere other than a senior operator&#8217;s head. If status lives in meetings, if approvals live in DMs, if exceptions live in folklore, and if nobody can tell which system holds truth, AI has nothing durable to work with. It becomes a smart veneer on top of organizational fog.</p><p>This is where the old BPR questions come back into focus.</p><p>Where does work stall? Where do approvals pile up? Where are the same decisions made twice? Where does the org chart interrupt the flow of the customer outcome? Where are people doing coordination work that the system itself could carry?</p><p>Those questions have not changed. The method has.</p><p>Old BPR started with interviews and whiteboards. AI-enabled rewiring starts with workflow exhaust: tickets, product specs, support logs, meeting notes, policies, approval chains, API events, backlog movement, and internal documentation. Old BPR produced a future-state diagram once or twice a year. AI-enabled rewiring can keep updating the map as the company actually behaves. Old BPR asked people to comply with a redesigned process. AI-enabled rewiring can embed part of the redesign into the workflow itself.</p><p>That is the real shift: from process redesign as an event to workflow rewiring as an operating discipline.</p><p>I think this matters because most companies are still framing AI too narrowly. They treat it as a content tool, a productivity add-on, or a layer for answering questions after the real work is done. That misses the bigger opening. AI becomes far more valuable when it helps the company read and rewrite how work moves.</p><p>I have seen that more clearly in the work around AI at Blockdaemon. One of the useful lessons behind Konstant.cloud was that the interesting problem was never &#8220;can the model answer a question?&#8221; The harder and more valuable question was whether the company could expose enough structured context, system hooks, permissions, and policy boundaries for AI to act usefully without creating chaos. In other words: could the workflow itself become machine-readable enough for AI to participate in it?</p><p>That is a much more serious ambition than adding a chatbot to the side of the business.</p><p>It also leads to a practical 30-day model.</p><p><strong>Days 1-5: map the actual workflow.</strong><br>Not the official version. The real version. Pull from Jira, Slack, product docs, support escalations, approvals, incident notes, and customer handoffs. Let AI surface the hidden bottlenecks, duplicated decisions, missing owners, and recurring stalls. What matters in this phase is not elegance. It is fidelity. You want to know how the work really moves.</p><p><strong>Days 6-10: define the constitutional layer.</strong><br>This is where many companies lose the plot. They rush to agents before deciding what the agents are allowed to do. Every serious AI system inside a company needs a constitutional layer: who can approve what, which data can be touched, which actions need human sign-off, what must be logged, and where the boundary sits between suggestion and execution. If that layer is vague, the AI system becomes politically dangerous long before it becomes operationally useful.</p><p><strong>Days 11-20: wire AI into the flow.</strong><br>Now the company gets leverage. One AI layer drafts requirements from messy input. Another flags stale tickets or broken dependencies. Another translates raw discussion into a decision-ready brief. Another routes issues to the right owner. Another explains policy at the point where work happens. Another turns buried context into usable state. This is the phase where AI stops looking like a demo and starts behaving like infrastructure.</p><p><strong>Days 21-30: tighten the loop.</strong><br>Measure what changed. Look at cycle time, escalation volume, rework, exception rates, and time-to-decision. Keep what reduces friction. Remove what adds noise. Rewrite the routing, prompts, policies, and thresholds. Then repeat the cycle. That repetition is the point. A rewired company is not redesigned once. It learns how to keep redesigning itself.</p><p>This, to me, is the real successor to business process reengineering.</p><p>The first BPR wave got captured by the worst instinct in management. It became associated less with redesigning work and more with cutting people. &#8220;Chainsaw Al&#8221; became the symbol: use process language to justify headcount removal, declare victory, and call it modernization. Some companies got leaner on paper. Many got weaker in reality. They reduced cost without building a better operating system.</p><p>AI now faces the same trap.</p><p>Use AI mainly as a labor-reduction story and people will hide work, protect turf, and move decisions into darker corners of the organization. The result will be faster fear, not better execution. Use AI to make the company clearer, faster, and easier to run, and the result is very different. You get better visibility. Better handoffs. Better written decisions. Better exception handling. Better memory. Better management.</p><p>That is why I think the opening question matters so much. Not &#8220;which jobs can AI replace?&#8221; but &#8220;which workflows are too slow, too manual, too fragmented, too opaque, or too dependent on memory?&#8221; One question produces a stronger company. The other produces a brittle one with better software.</p><p>This also has to come from the top. Workflow rewiring is not a side project. It needs an explicit operating priority, a decision about which systems hold truth, a writing discipline, named ownership, permission boundaries, and real trade-offs on roadmap and resourcing. If AI work lives as a part-time hobby scattered across well-meaning teams, the company gets prototypes and politics, not transformation.</p><p>That is why &#8220;AI transformation&#8221; is often too vague to be useful.</p><p>The more precise phrase is <strong>AI-enabled business process rewiring</strong>.</p><p>Same core instinct as BPR. Better instrumentation. Faster cycle times. Higher-resolution visibility. Tighter feedback loops. More room for human judgment where judgment matters, and less tolerance for chaos where chaos used to hide.</p><p>The companies that win this next phase will not be the ones with the most AI pilots. They will be the ones that can take one important workflow, make it legible, connect it to policy and machine-readable state, improve it inside 30 days, and then do it again the month after that.</p><p>AI will not save companies that cannot explain how work moves.</p><p>It will expose them.</p><p>And the winners will not be the companies with the best prompts. They will be the ones that can turn management itself into a rewiring discipline.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[AI Needs a Constitution]]></title><description><![CDATA[Why the next company will live between centralized intelligence and decentralized trust]]></description><link>https://writing.konstantin.ai/p/ai-needs-a-constitution</link><guid isPermaLink="false">https://writing.konstantin.ai/p/ai-needs-a-constitution</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Mon, 01 Jun 2026 02:07:12 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Adapted from a book in progress on AI, crypto, and the machine-readable company.</em></p><p>Most enterprise AI strategy still misses the real problem.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>It treats AI as a model problem.</p><p>In practice, it is an authority problem.</p><p>Companies are getting excited about copilots, agents, retrieval, automation, and conversational interfaces. Fine. The tools are real. But once AI moves closer to execution, the question changes fast. It is no longer just &#8220;is the model smart enough?&#8221; It becomes:</p><p>What is it allowed to see?</p><p>What is it allowed to combine?</p><p>What is it allowed to infer?</p><p>What is it allowed to do?</p><p>That is a very different conversation. It also happens to be where the next corporation will be built.</p><p>For most of modern corporate history, intelligence and trust lived in the same place. The company gathered the data, interpreted the data, and asked everyone else to trust that interpretation. The institution was the center of gravity.</p><p>That architecture is breaking.</p><p>AI centralizes intelligence. One strong AI layer can become the front door to search, writing, coding, triage, planning, and decision preparation. Whoever controls that layer starts shaping what the company sees and how it acts.</p><p>Crypto pushes in the opposite direction. It decentralizes trust by moving ownership, settlement, verification, and shared state outward into systems that no single institution fully owns.</p><p>One force pulls judgment inward. The other pushes trust outward. The next corporation will live between them.</p><p>Blockdaemon is one of the clearest operating examples of this tension I have seen.</p><p>Institutions do not want decentralized chaos. They want one answer, one security posture, one delivery path, one relationship they can trust. Underneath that clean surface, however, the underlying systems are distributed, technically varied, and operationally demanding. The company&#8217;s job is to make decentralized complexity feel like a single trusted institution.</p><p>Now the same design problem is showing up inside the company through AI &#8212; and the numbers are striking.</p><p>239%: Claude user growth in 14 weeks</p><p>70%: Daily return rate vs. 20&#8211;40% industry benchmark</p><p>4&#215;: Growth in Claude daily active users since February</p><p>3w &#8594; 1d: Engineering tasks that used to take weeks</p><p>95% of AI-suggested code is now being merged without change. That is not a drafting tool. That is a full engineering contributor.</p><p>But the productivity numbers are not the interesting part. The interesting part is how Blockdaemon built the intelligence layer underneath them.</p><p>The place to start is Konstant.cloud.</p><p>Most internal AI tools start at the bottom &#8212; pick a workflow, automate a task, measure the time saved. Konstant starts at the top. It begins with high-level strategic parameters: what the company is trying to do, what the priorities are, what the risks look like. Then it wires the tools together underneath that frame.</p><p>The best way I can describe the experience is this: most organisations feel like a nightclub with the lights off. People are moving, things are happening, signals are everywhere &#8212; but nobody can quite see the whole room at once. Konstant is the floodlight. Suddenly everything is visible. Not just the tasks that got done, but the shape of the company &#8212; where work is stalling, where priorities are misaligned, where a risk is building before anyone has named it.</p><p>It does not just surface information. It surfaces the right information to the right person at the right level of the organisation &#8212; in real time, grounded in what is actually happening across every system.</p><p>That matters because the hardest part of running a company is not executing. It is maintaining a clear picture of reality while everything is moving. Most leadership teams are working from a version of the business that is already three weeks old by the time it reaches them. Konstant collapses that lag.</p><p>But Konstant only works because of what sits underneath it. That is CORAL &#8212; a single SQL interface across every system the company operates, giving every agent the same unified read path. One layer that can see Salesforce, Jira, Slack, Datadog, Grafana, Notion, Stripe, and more, without each agent needing to negotiate its own connection to each system.</p><p>Machine-readable does not mean universally visible. A company only becomes AI-ready when legibility and permission evolve together.</p><p>CORAL is read-only by design. It sits behind the Enterprise Agent Gateway &#8212; a control plane that handles authentication, role-based access, PII masking, rate limits per team, and a forensic audit trail. Intelligence is centralized. Access is governed. Those are not the same thing, and conflating them is how AI projects go wrong.</p><p>In old enterprise software, permissions mostly answered a static question: can this user open this system?</p><p>In AI systems, the question gets much harder:</p><p>What context can the model retrieve?</p><p>What systems can it join?</p><p>What may it summarize for this role?</p><p>What actions require human review?</p><p>What boundaries remain intact even when the model can technically see across them?</p><p>That is why I think enterprise AI needs a constitution. Not a manifesto. A governing layer.</p><p>Look at how the PRD Agent works inside Blockdaemon. It drafts product requirements from feature requests. It creates tickets and epics automatically. But it pauses at an approval gate &#8212; a human signs off before anything proceeds into the project management system. The model is useful precisely because it is not indiscriminate.</p><p>Or the Feature Request Extractor, which monitors Slack autonomously, extracts product-worthy signals, looks up relevant products in the catalog, and creates a draft PRD &#8212; all without human initiation. The intelligence is real. The boundaries are explicit.</p><p>The future firm will not choose between centralized intelligence and decentralized trust. It will use both.</p><p>Intelligence will centralize where synthesis and decision preparation benefit from concentration. Trust will decentralize where verification, ownership, and portability benefit from openness.</p><p>The management question is no longer &#8220;centralize or decentralize?&#8221;</p><p>What belongs at the center, and what becomes more valuable when it moves outward?</p><p>That is the real design challenge now. And the companies getting it right are not the ones with the most sophisticated models. They are the ones that built the permission layer first.</p><p>That is where the next corporation gets built.</p><p></p><p></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[REWIRE]]></title><description><![CDATA[A practical method for making AI useful inside a real company]]></description><link>https://writing.konstantin.ai/p/rewire</link><guid isPermaLink="false">https://writing.konstantin.ai/p/rewire</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Tue, 26 May 2026 03:23:24 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div><hr></div><h1></h1><h2></h2><p><em>Adapted from a book in progress on AI, crypto, and the machine-readable company.</em></p><p>For the last five posts, I have been making the same argument from different angles: <strong>most companies do not have an AI problem. They have an illegibility problem.</strong> This is the practical answer to that diagnosis.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>I have seen the same underlying issue from three very different vantage points: around business process reengineering in the 1990s, inside the decision-preparation machinery of Deutsche Telekom, and now at Blockdaemon, building in a world where AI is no longer a side topic but part of the operating environment itself. The lesson across all three is consistent. Technology does not fix broken operating design. It reveals it, accelerates it, and occasionally makes it sound more coherent than it really is.</p><p>That is why I think the right question is not, &#8220;How do we deploy AI?&#8221; The right question is: <strong>what does the company have to become for AI to be useful, trustworthy, and worth scaling?</strong> My answer is a six-part method I call <strong>REWIRE</strong>:</p><ul><li><p><strong>R</strong>eframe the business around value flows</p></li><li><p><strong>E</strong>xternalize tacit work</p></li><li><p><strong>W</strong>ire the system of truth</p></li><li><p><strong>I</strong>nsert AI at the decision edges</p></li><li><p><strong>R</strong>eorganize around flows, not functions</p></li><li><p><strong>E</strong>nterprise-harden the new system</p></li></ul><p>Most companies try to start at step four. That is how they end up with demos instead of operating leverage.</p><p>One more point matters before the method itself. Making a company machine-readable does not mean making everything visible to everyone. In the old software world, access control mostly answered a static question: can this user open this system? In the AI world, the question gets much harder. What can this identity see, retrieve, combine, summarize, infer, and act on through the model? AI readiness is not just a workflow problem or a data problem. It is also a permissions problem.</p><h2><strong>1. Reframe the business around value flows</strong></h2><p>The first move is to stop looking at the company through the org chart. Customers do not buy reporting lines or internal departmental boundaries. They buy an outcome. That was true in the reengineering era, and it is even more true now.</p><p>At Blockdaemon, this has been obvious for a long time. Institutional customers do not want one answer from product, another from engineering, a third from security, and a fourth from delivery. They want one accountable surface and one coherent answer they can trust. Once you see the business that way, AI stops looking like a collection of local productivity tools and starts looking like part of the company&#8217;s ability to sense, interpret, and improve the path from customer demand to delivered outcome.</p><p>That distinction matters. If you frame the business by function, sales gets an AI assistant, engineering gets another, and support gets a third. The company becomes a federation of smarter silos. If you frame it by value flow, you start asking a different question: where does value slow down, distort, or lose accountability? When an institution asks whether you can support a workflow under a particular security posture by a particular date, that is not just a product question or a security question. It is a flow question touching commercial commitments, product readiness, engineering capacity, security review, runtime reliability, and delivery ownership all at once. If that flow is fragmented, AI will only make each local answer sound cleaner. It will not make the company more coherent.</p><h2><strong>2. Externalize tacit work</strong></h2><p>This is the step most leadership teams underestimate because so much company life runs on hidden labor. Large organizations depend on tracing, filtering, summarizing, framing, and context stitching long before anyone calls it AI.</p><p>I saw that clearly at Deutsche Telekom. Significant decisions did not arrive at the top in raw form. They were prepared by layers of people whose job was to make the company legible enough for executives and board members to act. In other words, serious companies already had a human intelligence layer. AI is now being asked to compress part of that layer, but it cannot compress work that the company never made explicit in the first place.</p><p>That is why so many AI efforts feel thin. The real process often lives in meetings, side channels, unwritten rules, copied spreadsheets, and the heads of the few people everyone calls when something breaks. At Blockdaemon, that became concrete during the konstant.cloud journey. The difficult part was never getting a model to produce words. The difficult part was getting enough of the company&#8217;s real context into a form that could be grounded, queried, and trusted. That means externalizing what companies usually leave vague: who owns the decision, what was actually decided, what changed since last week, where exceptions happened, why a customer issue took the path it did, and which source is canonical versus merely conversational. Until that work is done, humans keep rediscovering context and AI keeps guessing. Neither is an operating model.</p><h2><strong>3. Wire the system of truth</strong></h2><p>Once tacit work is surfaced, the next move is deciding where truth lives. This is where many AI programs quietly collapse, because the company has one version of reality in tickets, another in CRM, another in Slack, another in meeting notes, another in spreadsheets, and a sixth in people&#8217;s heads. Then leadership asks AI for a summary as if there were one coherent ground truth underneath it all.</p><p>At Blockdaemon, one of the clearest lessons has been that &#8220;AI strategy&#8221; is actually at least two different problems that companies often muddle together. The first is the AI layer itself: copilots, retrieval, workflow support, agents, decision prep. The second is the data foundation beneath it: where truth lives, how systems connect, how stable the sources are, and whether the model is being fed something coherent enough to reason over. Those two tracks are linked, but they are not the same.</p><p>A useful AI layer needs at least four kinds of readable state: work state, decision state, runtime truth, and permission state. Work state tells you what is in flight, who owns it, what changed, and what is blocked. Decision state preserves what was decided, by whom, with what assumptions, and what would change the call. Runtime truth connects live system behavior to commercial and operating reality. Permission state determines who is allowed to see what, retrieve what, combine what, and act where.</p><p>This is one reason I keep coming back to a fairly unglamorous point: a better view is useful, but a second truth is dangerous. For execution tracking, the logic inside Blockdaemon has been straightforward. Jira holds the operating truth; other layers can visualize it more elegantly, but they cannot compete with it. The moment the presentation layer starts pretending to be a separate source of truth, ambiguity returns precisely where AI needs clarity most.</p><p>The same discipline applies to AI integrations. Sensitive systems cannot just be connected directly to whichever model interface happens to be fashionable that quarter. The stronger architecture is controlled connectivity: a gateway boundary where data can be filtered, normalized, permissioned, and logged before it reaches the model. That is what an adult AI data layer looks like. Not maximal connectivity. Controlled connectivity.</p><h2><strong>4. Insert AI at the decision edges</strong></h2><p>Only after the first three steps does AI enter the picture directly, and that order matters. AI is most useful where work is high-volume, context-heavy, ambiguous, and expensive to prepare manually. That usually means strategic summaries, decision prep, issue triage, account context stitching, escalation support, first drafts, and source-grounded retrieval across fragmented information.</p><p>That is a very different posture from &#8220;put AI everywhere.&#8221; The first job is not replacing judgment. It is improving the quality and speed of decision preparation. That distinction matters to me because I have seen how much human labor large organizations already devote to preparing leaders to decide. AI changes the economics of that layer dramatically, but only if the company has already done the earlier work.</p><p>At Blockdaemon, the valuable AI use cases were never the flashy ones. They were the moments where AI could take dispersed context and turn it into usable clarity without severing the link to underlying truth. That is also where the MCP conversation becomes interesting. The point is not MCP as novelty. The point is MCP as a forcing function. Once you accept that models and agents will increasingly interact with systems through structured interfaces, you stop treating AI as a bolt-on after the APIs are done. You start designing the interface layer with machine use in mind from the beginning.</p><p>That is a bigger shift than it sounds. It means new APIs increasingly need to be machine-ready by default, not because protocol terminology is the story, but because the company no longer wants one layer for humans and a second incompatible layer for machines. Still, sequencing matters. Agentic workflows sit downstream of good interfaces. Good interfaces sit downstream of clean systems. Clean systems sit downstream of clear ownership and readable truth.</p><p>The same applies to engineering work. AI can accelerate migrations, repetitive implementation, and first-draft code generation, but the real pattern is still prompt, plan, verify. AI compresses effort. Humans review the result, check the edge cases, and decide whether the code is actually fit for use. That is not caution for its own sake. It is how real systems stay real.</p><h2><strong>5. Reorganize around flows, not functions</strong></h2><p>Once AI starts exposing where context breaks, a more uncomfortable truth appears. Many company problems are not technology problems at all. They are ownership problems.</p><p>If work crosses too many teams, AI will not rescue the flow. It will simply produce better summaries of the delay. If the handoff between sales, product, engineering, security, and delivery is structurally weak, AI will accelerate the distortion before it fixes the outcome. This is where the old reengineering lesson returns with fresh relevance. A company that wants real AI leverage has to reduce unnecessary handoffs, clarify end-to-end ownership, and organize more of its execution around flows instead of isolated functional excellence.</p><p>That does not mean functions disappear. It means the value flow becomes the thing leadership watches most closely. This is especially visible in institution-facing businesses because the customer experiences one promise. If the internal system fragments that promise across too many owners, trust degrades quickly.</p><p>This is one reason Blockdaemon is such a useful example for the broader argument. In our world, the institutional test and the machine-readability test are nearly the same test. Can the company produce one coherent answer? Can it connect commercial promise to delivery reality? Can it surface risk early enough to act on it? Can it carry the same truth from customer conversation to engineering action to runtime operation? If the answer is no, the customer feels it first. The AI layer just exposes the weakness faster.</p><h2><strong>6. Enterprise-harden the new system</strong></h2><p>This is the step that separates serious AI from clever demos. A real AI operating layer needs permissions, provenance, source traceability, human review paths, fallback behavior, clear boundaries around what the system knows and does not know, and auditability around high-consequence outputs and actions.</p><p>In a consumer app, getting 80 percent of the way there can still feel magical. In an enterprise, that missing 20 percent is where trust dies. If AI prepares a strategic summary for leadership, people need to know where the underlying facts came from. If it flags a delivery risk, the owner needs a way to inspect the signal. If it drafts an answer for a customer, the system needs to know whether it is allowed to use that context, whether it is crossing a boundary it should not cross, and whether a human review gate sits in front of the action.</p><p>That is why hardening is not cleanup. It is part of the design from the start. It is also where the policy-agent idea becomes genuinely interesting in AI terms. The question is no longer just whether a user can open an application. The harder question is what the model can see, retrieve, combine, infer, and act on, and under whose authority it is operating while it does those things. That is a different category of control. It is not screen-level access control. It is reasoning-aware and execution-aware permissioning.</p><p>This matters because once AI sits between people and systems, the policy layer is no longer guarding a page. It is governing how intelligence moves through the company. That means selective legibility: enough context to be useful, enough constraint to be trusted, enough authority to help, and not enough authority to quietly become dangerous. Without that architecture, AI centralizes visibility and action in ways that get reckless quickly. With it, AI becomes much more usable inside companies that actually care about trust, auditability, and controlled execution.</p><p>This is one reason I think Blockdaemon&#8217;s path is instructive. In our world, reliability is not ornamental and trust is not a branding exercise. If a system cannot be explained, governed, permissioned, and inspected, it does not matter how elegant the interface looks. That is true for infrastructure. It is also true for AI.</p><h2><strong>What REWIRE really means</strong></h2><p>A company does not become an AI company because it launches a chatbot, an assistant, or an internal prompt library. It becomes an AI company when intelligence is embedded into the operating system of the firm in a way that is readable, permissioned, governable, and tied to reality.</p><p>That is the standard behind REWIRE. Reframe the work around value flows. Externalize tacit work. Wire the system of truth. Insert AI at the decision edges. Reorganize around flows. Enterprise-harden the result. Do that well and AI becomes useful. Skip the first three steps and the rest is theater.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[One Throat to Choke]]></title><description><![CDATA[What institutions really buy when they buy infrastructure]]></description><link>https://writing.konstantin.ai/p/one-throat-to-choke</link><guid isPermaLink="false">https://writing.konstantin.ai/p/one-throat-to-choke</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Tue, 12 May 2026 18:41:56 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Adapted from a book in progress on AI, crypto, and the machine-readable company.</em></p><div><hr></div><p>There is a phrase that gets used in enterprise procurement. You will not hear it in the polished executive briefing or the partnership announcement. You hear it in the room after, when the people who actually manage the vendor relationship talk honestly about what they need.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p><strong>&#8220;One throat to choke.&#8221;</strong></p><p>Not elegant. But precise.</p><p>What institutional buyers want &#8212; underneath all the language about capabilities and roadmaps and long-term partnerships &#8212; is accountability without ambiguity. One person, one contract, one security review, one delivery path. One owner who cannot point at another team when something goes wrong.</p><p>That is not a procurement quirk. It is a diagnostic.</p><p>Banks, asset managers, regulated intermediaries, sovereign-grade infrastructure buyers &#8212; they have spent decades dealing with vendors who fragment accountability. Who bring five executives to the kickoff dinner and nobody accountable on the third Tuesday of a bad month. Who produce impressive proposals and manage relationships through charm, hoping the operating reality stays hidden long enough.</p><p>It does not stay hidden.</p><p>When serious customers show up, they do not just evaluate your product. They evaluate whether your company can describe itself. Whether it can produce one coherent answer, under pressure, about who owns what and what happens next.</p><p>That is the institutional test.</p><p>And it is the best diagnostic your company will ever have.</p><h2><strong>What selling to institutions at Blockdaemon taught me</strong></h2><p>At Blockdaemon, institutional sales taught me things no internal operating review ever could.</p><p>When you are selling blockchain infrastructure to a regulated financial institution, you are not selling a feature set. You are selling a confidence. You are asking a customer to trust critical operational plumbing to a company they cannot fully see inside. Their answer to that ask depends almost entirely on whether the company can speak for itself clearly.</p><p>That sounds obvious. In practice, it is brutal.</p><p>The first time a prospective tier-one financial institution asked us to walk through our end-to-end security posture across wallets, staking, and data infrastructure in a single conversation, the answer could not come from one place. It lived in three teams, two product lines, and a handful of people whose knowledge was only partly documented.</p><p>That was not a security problem. That was a legibility problem.</p><p>The customer was not wrong to ask. They were asking exactly the right question. In their world, an unclear answer to &#8220;who owns this?&#8221; is not a yellow flag. It is a red one.</p><p>That pressure became one of the most clarifying forces in how we thought about operating model design at Blockdaemon. The institutional customer was not just a revenue source. It was a truth machine. Every question they asked revealed another place where the company had not yet made its own work explicit enough to survive outside scrutiny.</p><h2><strong>The integrated pitch before the integration is finished</strong></h2><p>One of the most important insights I took from building the Blockdaemon institutional go-to-market was this:</p><p><strong>You have to pitch the integrated product before the integration is finished.</strong></p><p>That is uncomfortable but necessary.</p><p>Institutional buyers have long procurement cycles. By the time a major asset manager or global custodian bank has run their internal approval process, conducted their security review, aligned their legal and compliance teams, and mapped the commercial terms, six to eight months have passed. If you wait for everything to be built before you describe the vision, you are not in a pipeline. You are in a waiting room.</p><p>The right move is to sell the integrated story early and use the sales cycle as a co-design cycle. The customer becomes a partner in shaping the real operating architecture. Their procurement questions reveal the seams you still need to close. Their security review accelerates the hardening you needed to do anyway. Their roadmap asks help you prioritize what actually matters to the buyer over what matters to internal product politics.</p><p>That only works if the company can speak coherently about the vision.</p><p>And that only works if enough of the operating state is legible enough to support the conversation honestly.</p><p>Selling a story you cannot back up with real organizational state is not bold. It is a future embarrassment with a nice deck.</p><h2><strong>The moment institutional reality arrived</strong></h2><p>The most clarifying institutional pressure I experienced at Blockdaemon was a delivery commitment to one of the largest securities settlement infrastructures in the world &#8212; the kind of organization that sits at the center of trillions of dollars in daily market activity.</p><p>This is not a startup customer. It is foundational financial plumbing. Their counterparties are SEC-regulated. Their timelines are not suggestions. Their tolerance for ambiguity is essentially zero.</p><p>When a delivery milestone with a hard deadline becomes real with a customer like that, the company has nowhere to hide.</p><p>Everything that was implicit becomes urgent. Who owns the sprint? Who owns the security review? Who owns the customer communication if something slips? Who has authority to make the call when two teams disagree about scope? What is the single point of truth for delivery status?</p><p>Those are not engineering questions. They are operating model questions. And most organizations discover, under real deadline pressure, that they have been living on informal answers to all of them.</p><p>That experience shaped how I think about AI inside organizations.</p><p>Because the questions a tier-one institutional customer asks are exactly the questions an AI system needs answered too. What is the current state? Who owns it? What changed? What is at risk? What was decided and why?</p><p>If the company could not answer those cleanly for a customer like that, it would not be able to answer them cleanly for a model either.</p><p>The institutional test and the machine-readability test are the same test.</p><h2><strong>AI does not change what institutions require. It raises the stakes.</strong></h2><p>Here is the part that most enterprise AI conversations miss.</p><p>When AI enters the operating layer of a company serving serious institutional customers, the question is not whether the company has good models. The question is whether the company has real state.</p><p>Institutions do not accept &#8220;the AI said so&#8221; as an accountability answer. They want to know who owns the outcome. They want a human accountable for the commitment. They want to be able to audit the decision trail. They want to trust that what the company&#8217;s systems say is happening is actually happening.</p><p>AI can help the company move faster toward those answers. It can surface risk, summarize change, prepare briefings, route issues, and compress the decision-preparation labor that previously required a room full of people. I saw a version of this at Deutsche Telekom, where just making information flow cleanly to board-level decisions required enormous human infrastructure. AI changes the economics of that significantly.</p><p>But AI cannot substitute for the underlying truth.</p><p>At Blockdaemon, rolling out konstant.cloud alongside the institutional go-to-market made this vivid. The AI interface only becomes a genuine competitive advantage when the operating state underneath it is real. When work is tracked cleanly, decisions are preserved, customer commitments are visible, and runtime events connect back to ownership.</p><p>When those things are true, AI compresses the cycle from signal to action. An institutional customer asks a hard question Monday morning. The right people have context by Monday afternoon. A plan exists by Tuesday. That is speed that compounds.</p><p>When those things are not true, AI just produces beautiful responses to questions the company cannot actually answer.</p><h2><strong>What serious customers are really asking</strong></h2><p>I have spent a lot of time in rooms with institutional buyers. What they are really asking, under all the formal evaluation criteria, is a simpler thing:</p><p><strong>Are you the kind of company that knows what is happening inside itself?</strong></p><p>Can you tell me who owns the outcome? Can you tell me what changed since last week? Can you tell me what the risk is without making me wait three days for an internal briefing? Can you produce one coherent operating picture across wallets, staking, data, and delivery without me needing to interview five different product owners?</p><p>Those are not unfair demands. They are the minimum standard for operating at institutional scale.</p><p>And they are exactly the questions that force a company toward machine-readability.</p><p>The institutional customer does not care about your AI strategy. They care about your operating reality. But in a useful way, those two things are converging. The companies that become machine-readable to serve institutional customers will also be the companies whose AI systems actually work. The discipline is the same.</p><p>Make work visible. Make decisions durable. Make ownership explicit. Make runtime truth connect back to accountability.</p><p>Do that for your most demanding institutional customer and you have done it for the model.</p><p>Do it for the model and you have done it for every serious institution that comes after.</p><p>That is the insight worth building on.</p><p>Because the institutional test is not just a sales problem. It is the operating standard that makes everything else possible.</p><div><hr></div><p>This is the fifth post in a series on AI, crypto, and the machine-readable company.</p><p>Next: <strong>The REWIRE Method &#8212; a practical framework for redesigning the company AI can actually help.</strong></p><p><em>If this resonated, subscribe. The book is in progress.</em></p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Post 4: Do Not Automate the Mess]]></title><description><![CDATA[Why AI makes broken workflows more dangerous, not less]]></description><link>https://writing.konstantin.ai/p/do-not-automate-the-mess</link><guid isPermaLink="false">https://writing.konstantin.ai/p/do-not-automate-the-mess</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Mon, 04 May 2026 17:51:48 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Adapted from a book in progress on AI, crypto, and the machine-readable company.</em></p><p>The fastest way to waste money on AI is to automate a workflow you do not understand.</p><p>That sounds obvious. Companies do it every day.</p><p>They take a messy process with unclear ownership, inconsistent data, verbal decisions, hidden exception paths, and three unofficial workarounds. Then they layer AI on top and call it transformation.</p><p>What they get is not transformation.</p><p>What they get is a faster mess.</p><p>This is one of the oldest mistakes in management. A new technology arrives. Leadership sees leverage. The company skips the redesign and rushes to the tooling. It assumes the software will impose the discipline the organization never built for itself.</p><p>Sometimes that works for a quarter. Then reality sends the invoice.</p><p><strong>AI does not fix bad operating design. It amplifies it.</strong></p><p>That is why I keep coming back to the same line:</p><p><strong>Do not automate the mess. Rewrite the mess.</strong></p><p>If there is one idea worth carrying into every AI conversation, it is that one.</p><h2>Automation multiplies whatever is already true</h2><p>People talk about AI as though it injects intelligence into a company from the outside.</p><p>That is not quite right.</p><p>What AI usually does first is increase the speed, reach, and surface area of existing behavior. If the workflow is clear, AI can make it faster. If the workflow is ambiguous, AI can make the ambiguity travel farther. If the state is reliable, AI can help interpret it. If the state is fragmented, AI can turn fragmentation into polished output.</p><p>That is the danger.</p><p>A broken process used to be limited by human friction. It was slow. It was visible. It annoyed enough people that somebody eventually noticed.</p><p>AI reduces the friction.</p><p>Now the same broken process can route faster, summarize faster, answer faster, and scale faster. Which sounds useful until you realize the company has simply become more efficient at being wrong.</p><p>This is why so many AI deployments feel impressive in demos and disappointing in practice. The model looks smart. The company underneath it is still confused.</p><h2>I saw this firsthand at Blockdaemon</h2><p>I saw this clearly at Blockdaemon while rolling out konstant.cloud.</p><p>On the surface, the temptation was obvious. Build an AI interface, connect a powerful model, put a conversational layer on top of the company, and it looks like intelligence has arrived.</p><p>But that was never the hard part.</p><p>The hard part was making sure the system could ground itself in real operating state. If work lived in one place, decisions in another, customer context somewhere else, and important exceptions in side channels, then konstant.cloud would not become an operating advantage. It would become a very elegant way to generate confident summaries from incomplete reality.</p><p>That is the trap.</p><p>For konstant.cloud to be genuinely useful, it had to sit on top of something more disciplined: readable work, durable decisions, current source material, explicit ownership, and a tighter connection between what the company said was happening and what was actually happening.</p><p>In other words, the AI rollout immediately became an operating model project.</p><p>That is the part many leaders miss.</p><p>Once AI gets serious, it stops being mainly a tooling conversation. It becomes a truth conversation.</p><h2>The fantasy behind bad AI strategy</h2><p>A lot of enterprise AI strategy is built on a fantasy.</p><p>The fantasy is that intelligence can be added without redesign.</p><p>That leaders can keep the same fragmented systems, the same unclear handoffs, the same undocumented decisions, the same side-channel exceptions, and then somehow get a cleaner operating model by placing AI on top.</p><p>They cannot.</p><p>If the company has no shared reality, AI becomes a smoothing layer over disagreement.</p><p>If nobody knows who owns the outcome, AI becomes a polished narrator of organizational drift.</p><p>If the real process lives in folklore, AI can only automate the official fiction.</p><p>That last point matters more than most executives realize.</p><p>Most companies have two processes.</p><p>The first is the one in the documentation, the system, the executive review, and the training deck.</p><p>The second is the one people actually use to get things done.</p><p>The second one is usually the real company.</p><p>It lives in Slack messages, copied spreadsheets, hallway calls, escalation habits, personal notes, and unwritten rules. That is where customer commitments get rescued. That is where deadlines get saved or missed. That is where the company reveals what it actually trusts.</p><p>When leadership automates without redesigning, it usually automates the first process and ignores the second.</p><p>That is why the rollout looks clean while the business still feels chaotic.</p><h2>Where this goes wrong in practice</h2><p>The failure modes are not mysterious. They repeat.</p><h3>1. AI automates ambiguity</h3><p>A workflow says an item is &#8220;in progress.&#8221; One team means active work. Another means blocked pending review. A third means mostly done.</p><p>Then leadership asks AI what is at risk.</p><p>The model cannot resolve what the company never standardized. It can only interpret inconsistent signals and produce a sentence with good grammar.</p><p>That is not intelligence. That is ambiguity with formatting.</p><h3>2. AI accelerates bad handoffs</h3><p>A customer request enters the system. Sales frames it one way, product interprets it another way, engineering receives it stripped of commercial context, and operations inherits the consequences later.</p><p>Now add AI summaries and automated routing.</p><p>Did the handoff get fixed? No. It got faster. The distortion now moves with less resistance.</p><h3>3. AI gives management false confidence</h3><p>This one is especially dangerous.</p><p>When a human gives a messy status update, you can usually hear the uncertainty. When AI gives one, the uncertainty often arrives flattened into tidy prose. The result feels more authoritative than it is.</p><p>That is how leaders end up managing a description of reality instead of reality itself.</p><h3>4. AI hardens the wrong workflow</h3><p>Sometimes the most valuable people in a company are quietly compensating for a bad system. They know the hidden dependencies. They know which approvals can be ignored, which fields are stale, which spreadsheet matters, which issue needs a phone call instead of a ticket.</p><p>Then AI gets rolled out against the formal workflow.</p><p>Instead of fixing the design, the company hardens the bureaucracy around it. The machine now reinforces the weaker system because the stronger informal one was never made explicit.</p><h3>5. AI widens operational blast radius</h3><p>A bad manual process is annoying. A bad automated process is dangerous.</p><p>If AI is helping route incidents, summarize customer commitments, prioritize engineering work, prepare executive briefings, or interpret runtime events, then weak process design is no longer just inefficient. It is operational exposure.</p><p>That is when the bill comes due.</p><h2>The lesson from the last reengineering era</h2><p>This is one reason the current AI wave feels familiar to anyone who watched how business process reengineering got abused in the 1990s.</p><p>There was a valid insight at the center of that movement. Companies should redesign work around outcomes, not departments. They should reduce bad handoffs, simplify flows, and challenge internal bureaucracy.</p><p>Correct.</p><p>Then came the misuse.</p><p>Many leaders kept the language of redesign but skipped the discipline. Reengineering became a cover story for layoffs, centralization, and short-term optics. The company looked leaner on paper while the operating substance got weaker.</p><p>The current AI moment carries the same temptation.</p><p>Executives say they want productivity, speed, and leverage. Fine. Some of that is real. But too often the first question is still, &#8220;How many people can this replace?&#8221; instead of, &#8220;What is the actual flow of work, and why is it broken?&#8221;</p><p>That is the tell.</p><p>If AI starts as a labor-reduction story before it becomes a design story, the company is probably heading toward a more expensive version of the same old mistake.</p><p>The technology is better now. The management failure mode is not.</p><h2>What rewriting the mess actually means</h2><p>&#8220;Rewrite the mess&#8221; does not mean stop automating. It means earn the automation.</p><p>It means doing the harder operating work first so AI has something real to work with.</p><p>That usually starts in five moves.</p><h3>1. Reframe around the value flow</h3><p>Do not start with departments. Start with the end-to-end outcome.</p><p>How does customer demand become delivery?<br>Where does trust get won or lost?<br>Where do commercial promises meet technical reality?<br>Where does work wait?<br>Where does context degrade?</p><p>The org chart is not the unit of failure. The flow is.</p><h3>2. Expose tacit work</h3><p>Find the places where the company depends on oral tradition.</p><p>Which decisions disappear after meetings?<br>Which escalations happen off-system?<br>Which exceptions are handled informally?<br>Which teams keep shadow records because they do not trust the official ones?</p><p>That is the hidden company. If you do not surface it, you will automate around it and wonder why the results feel fake.</p><h3>3. Choose canonical truth</h3><p>Pick where truth lives for work, decisions, customer state, and runtime state.</p><p>Not in theory. In practice.</p><p>A company cannot become machine-readable if every important question requires interpretation across five systems and three memories.</p><h3>4. Insert AI where judgment is expensive and state is strong</h3><p>This is where AI actually shines.</p><p>Use it where first-draft economics matter, where summarization helps, where routing helps, where issue triage helps, where decision preparation helps, and where humans still own the final call.</p><p>Do not start where the underlying state is weakest. That is just industrialized guessing.</p><h3>5. Harden the loop</h3><p>Once AI enters the workflow, the workflow has to become more auditable, not less.</p><p>Who approved what?<br>What source grounded the recommendation?<br>What changed?<br>What failed?<br>Who owns the override?</p><p>If the answer is &#8220;it depends,&#8221; the system is not ready.</p><h2>This is the real work</h2><p>The most important AI projects in the next decade will not be the ones with the flashiest demos.</p><p>They will be the ones that force companies to become more legible to themselves.</p><p>That was the deeper lesson for me from rolling out konstant.cloud at Blockdaemon. The product side was interesting. The model side was interesting. But the real lesson was operational. Once you try to make AI genuinely useful inside a company, you immediately discover the quality of the company&#8217;s own state.</p><p>You find out what is explicit and what is folklore.<br>You find out where ownership is clear and where it is fuzzy.<br>You find out whether decisions survive the meeting in which they were made.<br>You find out whether the system reflects reality or merely describes a cleaner version of it.</p><p>That is not a side issue.</p><p>That is the project.</p><p><strong>Do not automate the mess. Rewrite the mess.</strong></p><p>Everything good comes after that.</p><p>This is the fourth post in a series on AI, crypto, and the machine-readable company.</p><p>The next step is turning this into a method: how leaders can redesign the company so AI becomes useful without making the business more brittle, more centralized, or more detached from reality</p>]]></content:encoded></item><item><title><![CDATA[Post 3: The Machine-Readable Company ]]></title><description><![CDATA[Operating standard. Why AI only becomes useful when work becomes legible.]]></description><link>https://writing.konstantin.ai/p/post-3-the-machine-readable-company</link><guid isPermaLink="false">https://writing.konstantin.ai/p/post-3-the-machine-readable-company</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Thu, 30 Apr 2026 06:23:35 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Adapted from a book in progress on AI, crypto, and the machine-readable company.</em></p><p>Most companies are trying to deploy AI into operating environments they cannot describe with a straight face.</p><p>That is the problem.</p><p>They say they want copilots, agents, automation, and decision support. Fine. The tools are real. The models are real. The leverage is real.</p><p>But most companies are still running on a mix of meetings, side channels, tribal memory, half-maintained systems, and a few indispensable employees who know where the real process is buried.</p><p>Then leadership acts surprised when AI sounds polished and behaves inconsistently.</p><p>It should not be surprising.</p><blockquote><p><strong>Most companies do not have an AI problem. They have an illegibility problem.</strong></p></blockquote><p>The company is not readable enough for AI to be reliably useful.</p><p>That is what I mean by a <strong>machine-readable company</strong>.</p><p>Not a company run by robots.<br>Not a company reduced to workflow diagrams.<br>Not a giant management dashboard pretending to be strategy.</p><p>A machine-readable company is one where the important parts of work exist in forms that humans, systems, and AI can all interpret from the same underlying truth.</p><p>That standard is much higher than most firms realize.</p><h3>What machine-readable actually means</h3><p>A machine-readable company can answer basic operating questions from system state instead of institutional folklore.</p><p>What is in flight?<br>Who owns it?<br>What changed this week?<br>What is blocked?<br>Which customer commitment is at risk?<br>What decision was made, by whom, and why?<br>Which exception path was used?<br>Where is work happening outside the official process?</p><p>Most companies cannot answer those questions cleanly without calling a meeting.</p><p>That is not because the people are weak. It is because the operating system is fragmented.</p><p>Part of the truth lives in the CRM.<br>Part lives in tickets.<br>Part lives in Slack.<br>Part lives in meeting notes.<br>Part lives in spreadsheets that everyone claims are temporary.<br>Part lives in the heads of the five people everyone calls when something breaks.</p><p>And then somebody asks AI to summarize reality.</p><p>Reality compared to what?</p><p>If the company has three versions of the truth, the model will smooth, infer, and guess. It may do it elegantly. That is what makes the problem dangerous. AI does not just expose broken state. It can make broken state sound coherent.</p><p>A bad system used to fail noisily.<br>AI lets it fail persuasively.</p><h3>Three layers have to become readable</h3><h4>1. Work state</h4><p>Tasks, priorities, blockers, dependencies, deadlines, ownership, and status changes must be recorded consistently enough that another system can interpret them without a human translator.</p><h4>2. Decision state</h4><p>Important decisions have to survive the meeting where they were made. If the company cannot preserve judgment in durable form, both humans and machines are forced to rediscover the same context over and over.</p><h4>3. Runtime truth</h4><p>Operational reality has to connect back to ownership and action. Deployments, incidents, security exceptions, failed flows, and customer-impacting events cannot remain isolated from the operating layer.</p><p>That is where &#8220;machine-readable&#8221; stops sounding neat and starts sounding necessary.</p><h3>Why this matters now</h3><p>Because AI is moving from the edges of work toward execution.</p><p>It drafts the memo.<br>It routes the issue.<br>It summarizes the account.<br>It explains the incident.<br>It proposes priorities.<br>It helps write the code.<br>It shapes what managers see before they decide.</p><p>That means bad state no longer just creates bad reporting.</p><p>Bad state creates bad action.</p><p>This is why so many AI rollouts look good in demos and feel thin in production. The model is capable. The company is not queryable enough to support the use case at scale.</p><p>Everyone wants intelligence. Very few want the discipline intelligence requires.</p><h3>What a leader should do next</h3><p>Start with a blunt audit.</p><p>Where does important work still depend on oral tradition?<br>Where do exceptions get handled off-system?<br>Where does ownership blur?<br>Where do teams maintain shadow records because they do not trust the official one?<br>Which decisions disappear after the meeting that produced them?<br>What would an intelligent system need in order to answer &#8220;what is at risk this week?&#8221; with confidence?</p><p>Then make a few hard choices:</p><ul><li><p>pick the canonical system for work state</p></li><li><p>pick the canonical place where decisions get preserved</p></li><li><p>pick the operational signals that must connect back to owners and actions</p></li><li><p>standardize the minimum vocabulary that lets teams interpret state the same way</p></li><li><p>stop tolerating ten unofficial truths</p></li></ul><p>That is the beginning.</p><p>Not an AI strategy deck.<br>Not innovation theater.<br>Not a cloud of pilots hovering above broken workflows.</p><p>A readable operating system.</p><h3>The real point</h3><p>The machine-readable company matters because it is the condition that makes AI useful without making the business more brittle.</p><p>If AI is the great centralizer, machine-readable work is what keeps that centralization grounded in reality instead of polished fiction.</p><p>If crypto is the great decentralizer, machine-readable work is also how firms interact with decentralized systems without drowning in complexity.</p><p>That is the bridge.</p><p>The next question is the one most companies are still trying to avoid:</p><blockquote><p><strong>Once you can see the mess clearly, what do you do with it?</strong></p></blockquote><p>That is the subject of the next post: <strong>Do Not Automate the Mess.</strong></p>]]></content:encoded></item><item><title><![CDATA[Post 2: AI Is the Great Centralizer. Crypto Is the Great Decentralizer.]]></title><description><![CDATA[Structural context. The next generation of companies will have to operate with both.]]></description><link>https://writing.konstantin.ai/p/ai-is-the-great-centralizer-crypto</link><guid isPermaLink="false">https://writing.konstantin.ai/p/ai-is-the-great-centralizer-crypto</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Tue, 28 Apr 2026 02:09:28 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Adapted from a book in progress on AI, crypto, and the machine-readable company.</em></p><p>Most people talk about AI and crypto as if they live in different universes.</p><p>They do not.</p><p>They are opposite responses to the same broad shift in how power moves through technology.</p><p>AI centralizes intelligence.<br>Crypto decentralizes trust.</p><p>That is the frame.</p><p>AI centralizes because the best models, interfaces, and workflow layers concentrate capability. One good model can become the front door to writing, search, coding, triage, planning, analysis, and decision support. Whoever controls that layer starts shaping what the organization sees and how it acts.</p><p>Crypto moves in the opposite direction. It decentralizes trust by pushing ownership, settlement, coordination, and verification outward into shared systems. It reduces dependence on single intermediaries and makes more of the trust layer inspectable and portable.</p><p>One concentrates intelligence.<br>The other distributes trust.</p><p>Modern institutions will need both.</p><p>That is the part people miss when they treat these as separate conversations. AI without counterweights tends toward concentration. Crypto without usable intelligence tends toward complexity. One becomes too centralized. The other becomes too hard to use cleanly at scale.</p><p>Put differently:</p><blockquote><p><strong>AI tells you what to do. Crypto helps determine who gets to decide, what can be verified, and how value moves.</strong></p></blockquote><p>That is why the overlap matters.</p><p>Inside the firm, AI increasingly becomes the operating layer. It drafts, routes, summarizes, recommends, prioritizes, and interprets. It pulls intelligence upward into fewer interfaces and fewer decision surfaces.</p><p>Outside the firm, crypto changes the trust architecture. It makes ownership, settlement, and coordination less dependent on closed institutional stacks and more dependent on networks, rules, and verifiable state.</p><p>The companies that matter in the next decade will live between these forces.</p><p>They will use centralized intelligence on top of decentralized systems. They will need the leverage of AI without surrendering entirely to concentration. They will need the openness of decentralized trust without drowning in fragmentation.</p><p>That tension is not theoretical. It is already visible in businesses that have to translate decentralized infrastructure into something institutional customers can actually use. Institutions want one accountable surface. One security posture. One delivery path. One contract owner. Underneath that clean surface may sit multiple protocols, multiple chains, multiple dependencies, and multiple systems of record.</p><p>So the real challenge is not choosing AI or crypto. It is designing a company that can operate with both.</p><p>That takes more than tooling.</p><p>It takes operating clarity.</p><p>Because if AI is going to sit at the center of how work gets interpreted and crypto is going to sit closer to how trust and value get coordinated, then the company itself has to become more legible. Otherwise the intelligence layer centralizes on top of internal confusion and the trust layer decentralizes into operational chaos.</p><p>That is the bridge.</p><blockquote><p><strong>AI centralizes intelligence. Crypto decentralizes trust. The next generation of companies will need both.</strong></p></blockquote><p>But neither works cleanly if the company itself remains unreadable.</p><p>That takes us to the next question:</p><p>What would it actually mean for a company to become readable enough for humans, systems, and AI to operate from shared truth?</p><p>That is the subject of the next post: <strong>The Machine-Readable Company.</strong></p>]]></content:encoded></item><item><title><![CDATA[Post 1: The Customer Never Buys the Org Chart]]></title><description><![CDATA[Why most AI strategies fail before the model even arrives]]></description><link>https://writing.konstantin.ai/p/the-customer-never-buys-the-org-chart</link><guid isPermaLink="false">https://writing.konstantin.ai/p/the-customer-never-buys-the-org-chart</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Tue, 21 Apr 2026 00:40:18 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><em>Adapted from a book in progress on AI, crypto, and the machine-readable company.</em></p><p>I learned this lesson twice.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>The first time was in the late 1990s, when I was a young management consultant working on restructuring projects around Al Dunlap, the executive known as Chainsaw Al. In those rooms, business process reengineering was often presented as hard-headed managerial truth. Strip out layers. Eliminate overlap. Move faster. Cut deep enough and the company would come back stronger.</p><p>The second time was decades later, as CEO of Blockdaemon, sitting across from institutional customers who did not care how our company was organized internally. They did not care which team owned which product. They did not care whether a problem sat with engineering, product, security, delivery, or operations.</p><p>They wanted one answer to one question:</p><p><strong>If we trust you with critical infrastructure, who owns the outcome?</strong></p><p>That question gets to the heart of how companies actually work.</p><p>Customers experience the company as one thing. Management rarely does.</p><p>Inside a company, the work may cross multiple teams, systems, reporting lines, and planning cycles before the customer sees the outcome. Each function may be doing sensible work. Each leader may be acting rationally. Each team may be optimized for its own metric. And still the experience can feel fragmented, slow, and brittle because the customer does not buy your internal complexity. They buy the total flow.</p><p>That was the core insight behind business process reengineering in the 1990s, and it was a good one. The unit of analysis should not be the department. It should be the end-to-end flow of value. How does customer demand become delivery? Where do decisions stall? Where does information degrade? Where do accountability and ownership blur?</p><p>That idea was right.</p><p>What went wrong was what companies did with it.</p><p>Too often, &#8220;reengineering&#8221; became a nicer word for mass layoffs, centralization, and short-term financial optics. The language sounded rigorous. Processes would be streamlined. Redundancies would be eliminated. The company would become more focused and efficient. Sometimes the numbers improved just long enough to keep the story alive.</p><p>But in many cases, what was called reengineering was not a redesign of the operating system. It was an extraction event. Capability left with the people who were cut. Judgment left with them too. The company looked leaner while becoming weaker.</p><p>I saw enough of that firsthand to know how easily a valid management idea can be weaponized.</p><p>That is why the AI moment feels familiar.</p><p>Today, executives say they want to use AI to make their organizations faster, leaner, and more productive. Fine. Some of that is real. But too much of what passes for AI strategy already sounds like the old reengineering misuse in a new accent.</p><p>The first question is often not, &#8220;How should work be redesigned?&#8221;<br>It is, &#8220;How many people can this replace?&#8221;</p><p>The first instinct is not to clarify ownership, improve system state, or remove bad handoffs. It is to drop a model on top of a messy operating environment and hope efficiency falls out.</p><p>That is the same mistake.</p><p><strong>You cannot get the benefits of redesign by skipping the redesign.</strong></p><p>This is one reason Blockdaemon has been such a useful lens for me. In a business like ours, fiction gets punished quickly. We serve institutions that expect reliability, accountability, and operational clarity. We operate across infrastructure, security, delivery, and commercial commitments. If the company is fuzzy about who owns what, where truth lives, or how issues move from signal to action, the problem does not stay theoretical. It turns into delay, risk, and lost trust.</p><p>That environment exposes a hard truth:</p><p><strong>The org chart is an internal convenience. It is not a customer-facing design.</strong></p><p>Institutions do not want to buy your internal seams. They want one commercial posture, one security story, one delivery path, and one answer when something matters. They want a coherent surface.</p><p>And now AI raises the stakes.</p><p>The problem with AI in most enterprises is not that the models are weak. The problem is that the company itself is illegible. Work lives in too many places. Decisions are trapped in meetings. Exceptions live in side channels. Ownership is assumed rather than stated. Systems disagree with each other. The company has an official process, a workaround, and an emergency path everybody relies on when the official process breaks.</p><p>Then leadership adds AI and asks for leverage.</p><p>Leverage over what?</p><p>If the company cannot clearly tell a machine what is happening, the machine will infer, smooth, and guess. Sometimes it will even sound persuasive while doing it. That is why so much enterprise AI feels polished and hollow at the same time. The outputs are cleaner. The underlying operating state is still a mess.</p><p>This is the real sequel to 1990s business process reengineering.</p><p>Back then, the question was whether companies could redesign work around outcomes instead of departments.</p><p>Today, the question is whether companies can redesign work so machines can participate in it without making the business more brittle, more centralized, or more detached from reality.</p><p>That changes the standard.</p><p>It is not enough to have a process. The process has to be legible.<br>It is not enough to have ownership. Ownership has to be explicit.<br>It is not enough to have systems of record. They have to contain enough truth that humans and machines can act from them without inventing the missing pieces.</p><p>Most companies are not there yet.</p><p>That is not an AI problem. It is an operating model problem.</p><p>And it is why I keep coming back to one line:</p><p><strong>Do not automate the mess. Rewrite the mess.</strong></p><p>That is the real task.</p><p>The companies that win in the AI era will not be the ones that adopted the most tools first. They will be the ones that used this moment to make work legible, decisions durable, ownership explicit, and trust portable.</p><p>I saw what happened in the 1990s when a real management idea got turned into a body-count strategy. We should be honest that the same temptation is back. The language of efficiency is back. The pressure to centralize is back. The temptation to confuse cost reduction with redesign is back.</p><p>The difference is that this time the technology is real.</p><p>Which means the upside is real too, if companies are willing to do the harder thing first.</p><p>Not automate confusion.<br>Redesign the company so intelligence can actually help.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[AI Will Not Fix a Company That Cannot Read Itself]]></title><description><![CDATA[Why the next generation of companies will be rebuilt around machine-readable work, centralized intelligence, and decentralized trust.]]></description><link>https://writing.konstantin.ai/p/ai-will-not-fix-a-company-that-cannot</link><guid isPermaLink="false">https://writing.konstantin.ai/p/ai-will-not-fix-a-company-that-cannot</guid><dc:creator><![CDATA[konstantin]]></dc:creator><pubDate>Fri, 17 Apr 2026 00:58:16 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!zhyW!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F48b85d9c-5291-4066-b159-1e7364e3033e_485x485.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Most companies think they need an AI strategy.</p><p>What they actually need is a redesign of how work happens.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p>That is the first thing worth saying plainly, because the market is already flooded with AI advice that mistakes the symptom for the disease. The symptom is that teams want better outputs, faster decisions, less manual work, and more leverage. The disease is that most companies are illegible to themselves.</p><p>Decisions live in meetings. Exceptions live in Slack. Ownership lives in tribal memory. Real process lives in the gap between what the system says is happening and what a few over-relied-on people know is actually happening.</p><p>Then leadership buys copilots, launches an AI initiative, and wonders why the output feels impressive and untrustworthy at the same time.</p><p>The model is not the first problem.</p><p>The company is.</p><h2><strong>The company is illegible</strong></h2><p>Most firms cannot see the true flow of work from customer request to internal decision to shipped result.</p><p>They do not know where ownership really sits.</p><p>They do not know which exceptions are routine and which are dangerous.</p><p>They do not know how many decisions made in meetings ever became durable state anywhere else.</p><p>They do not know how much of their operating rhythm depends on someone manually reconciling the gap between system truth and reality.</p><p>That kind of company can use AI.</p><p>It just cannot use AI well.</p><p>This is the hidden reason so much enterprise AI still feels like theater. The demonstrations are real. The capabilities are real. The disappointment is real too. Companies are trying to get machine leverage from work that does not yet exist in a machine-readable form.</p><p>They want systems that can reason over the business, but the business is scattered across calendars, chats, documents, tickets, spreadsheets, and oral history.</p><p>So the AI sees a version of the company.</p><p>It just is not the company.</p><p>That is why the output often feels slick but thin. The system is producing from partial state. It is reading the visible paperwork of the business while missing the hidden operating logic that experienced people carry in their heads.</p><p>That hidden logic is usually the whole game.</p><p>It tells you which customer exception matters, which dependency is political rather than technical, which project is marked green but is actually blocked, which system of record is aspirational rather than real, and which executive phrase actually means this has to ship by Friday.</p><p>Until that logic becomes explicit, AI does not transform the company.</p><p>It mostly reveals how badly the company needs redesign.</p><h2><strong>The best old idea in management is suddenly useful again</strong></h2><p>In the 1990s, business process reengineering started with a clean, brutal insight: stop optimizing departments and redesign work end to end.</p><p>A company does not create value through isolated functions. It creates value through flows.</p><p>Lead to close. Order to cash. Incident to resolution. Product idea to shipped capability. Customer problem to trusted outcome.</p><p>That insight was right.</p><p>What went wrong was what many companies did with it. Reengineering became famous for cost cutting, org trauma, and executive overreach. Some of that reputation was earned.</p><p>But buried inside the wreckage was a point that matters again, maybe more than it did the first time:</p><p><strong>If you want step-change performance, you do not automate the mess. You rewrite the mess.</strong></p><p>That was true in the age of ERP.</p><p>It is even more true in the age of AI.</p><p>The difference is that the substrate has changed. Today the modern firm has to become machine-readable. Work has to exist in a form that people, software, and AI can all act on from the same underlying state. Decisions have to survive the meeting where they were made. Exceptions have to be visible. Ownership has to be explicit. The live state of the company cannot remain trapped in side channels and human memory.</p><p>That is not a minor upgrade.</p><p>It is a redesign of the corporation&#8217;s operating grammar.</p><h2><strong>AI is the great centralizer</strong></h2><p>AI is often discussed as a productivity tool.</p><p>That is too small.</p><p>Structurally, AI is a centralizing force.</p><p>It concentrates intelligence, interface power, and workflow leverage into fewer layers. One good AI system can become the front door to search, writing, planning, coding, support, and eventually action. Whoever controls the model, the memory, and the workflow layer starts shaping what the organization sees and how the organization decides.</p><p>Inside a company, AI centralizes:</p><ul><li><p>knowledge access</p></li><li><p>workflow routing</p></li><li><p>first-draft creation</p></li><li><p>management visibility</p></li><li><p>decision support</p></li><li><p>operating leverage</p></li></ul><p>That is not a moral judgment. It is a structural one.</p><p>The best AI systems become the nervous system of the firm very quickly.</p><p>If the company is machine-readable, that is powerful.</p><p>If it is not, you get a polished hallucination machine sitting on top of operational confusion.</p><h2><strong>Crypto is the great decentralizer</strong></h2><p>Crypto is often described in the worst possible language.</p><p>Too much of the public story is still trapped in speculation, ideology, or tourist-grade futurism.</p><p>The serious story is simpler.</p><p>Crypto is a decentralizing force because it changes where trust lives.</p><p>It pushes ownership, settlement, coordination, and verification outward into networks rather than private ledgers controlled by single firms. It creates shared state, programmable assets, and verifiable execution across parties that do not need one institution to own the entire system of record.</p><p>In business terms, that matters because crypto is not mainly about coins. It is about trust architecture.</p><p>It changes how institutions can think about:</p><ul><li><p>settlement</p></li><li><p>asset control</p></li><li><p>identity</p></li><li><p>shared state</p></li><li><p>multi-party coordination</p></li><li><p>provenance and auditability</p></li></ul><p>That is why AI and crypto matter together.</p><p>AI centralizes intelligence.</p><p>Crypto decentralizes trust.</p><p>The next generation of institutions will need both.</p><h2><strong>Why these two forces belong in the same conversation</strong></h2><p>Most people write about AI and crypto separately.</p><p>That misses the actual story.</p><p>The future enterprise will have to use centralized intelligence on top of decentralized systems.</p><p>It will need AI to interpret complexity, route work, draft decisions, surface risk, and shorten cycles.</p><p>It will need decentralized infrastructure to verify, settle, coordinate, and preserve agency across networks.</p><p>One gives the firm a stronger intelligence layer.</p><p>The other gives it a more programmable trust layer.</p><p>Without AI, decentralized systems remain too complex for most institutions to use well.</p><p>Without decentralized trust, AI-era operating systems risk concentrating too much power into too few interfaces and platforms.</p><p>One solves the intelligence bottleneck.</p><p>The other constrains the trust bottleneck.</p><p>That is why I think these are the two most important technology forces of this era.</p><p>They solve opposite problems.</p><p>And serious companies now have to design for both.</p><h2><strong>Why I am writing this book</strong></h2><p>I am writing this because I sit at the intersection where AI and crypto stop being abstractions and become operating realities.</p><p>Most books about AI in business are written by one of three people: software vendors, consultants, or futurists who do not have to ship under pressure.</p><p>Most books about crypto are written by traders, protocol evangelists, or ideological decentralization purists.</p><p>That leaves a gap.</p><p>The interesting story is not what these technologies promise in theory.</p><p>The interesting story is what happens when they hit a real company with customers, deadlines, compliance obligations, product sprawl, handoffs, outages, and revenue targets.</p><p>That is where new operating models are either built or exposed.</p><p>I work close to that seam.</p><p>I see how AI changes the internal operating system of a firm. I see how crypto changes the external trust architecture that firm plugs into. And I see how quickly both of those forces expose whether the company actually understands its own work.</p><p>That is the book I want to write.</p><p>Not a book about prompts.</p><p>Not a book about hype.</p><p>A book about how companies redesign themselves so AI can become useful, and so decentralized systems can become usable.</p><h2><strong>The method: REWIRE</strong></h2><p>The framework I am building for the book is called <strong>REWIRE</strong>.</p><p>It is a modern reengineering method for AI-native firms:</p><ul><li><p><strong>Reframe</strong> around value streams</p></li><li><p><strong>Externalize</strong> tacit work</p></li><li><p><strong>Wire</strong> the system of record</p></li><li><p><strong>Insert AI</strong> at decision edges</p></li><li><p><strong>Reorganize</strong> around flows</p></li><li><p><strong>Enterprise harden</strong> the new system</p></li></ul><p>The point is simple.</p><p>Before you ask AI to run across the company, the company has to become legible enough for AI to read.</p><p>Before you let agents act, you need explicit state, explicit ownership, explicit permissions, and explicit fallback paths.</p><p>Before you celebrate automation, you need to know whether you are automating signal or automating noise.</p><p>That is the difference between transformation and theater.</p><h2><strong>What the next generation firm will look like</strong></h2><p>The firms that win this transition will not be the ones with the most AI subscriptions.</p><p>They will be the ones that learn how to centralize intelligence without centralizing fragility.</p><p>They will be the ones that learn how to decentralize trust without decentralizing accountability.</p><p>They will be the ones that make work machine-readable.</p><p>They will be the ones that treat hardening as part of the commercial product, not as cleanup after the demo.</p><p>They will be the ones that can answer basic questions from system state instead of from a meeting.</p><p>What is blocked this week.</p><p>Which accounts are at risk.</p><p>Which decisions were made.</p><p>Which dependencies are real.</p><p>Which workflows are healthy.</p><p>Which parts of the company still rely on memory instead of machinery.</p><p>That is where this is going.</p><p>The future enterprise will not just use AI.</p><p>It will be redesigned around the conditions that make AI useful.</p><p>And it will increasingly operate in a world where trust, value, and verification move through more decentralized rails.</p><p>That is why this is not just a story about tooling.</p><p>It is a story about the next institutional form.</p><h2><strong>The line I keep coming back to</strong></h2><p>Most companies do not have an AI problem.</p><p>They have an illegibility problem.</p><p>Until they fix that, they will keep asking machines to understand a business that even they cannot fully read.</p><div><hr></div><p>If this idea resonates, subscribe.</p><p>I am developing this into a book on how to make AI actually useful in corporations by redesigning the work itself. In future essays, I will dig into the REWIRE framework, the relationship between AI and crypto, and what the machine-readable company looks like in practice.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://writing.konstantin.ai/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item></channel></rss>