The Seat License Is Dead: AI Just Did to Software What DeFi Did to Finance

The most dangerous headline in enterprise software right now contains zero data. It ran on Crypto Briefing on a quiet morning in early May 2026. Eleven words: "AI reshapes software budgets as upstarts gain, established firms cut prices." No named company. No churn chart. No ARR disclosure. No pricing table. No quarter-over-quarter revenue breakdown. Just a claim, delivered with the calm authority of audited fact, and then the machinery of market narrative did what it always does: it started moving money on vibes. I have seen this movie before. In late 2017, during the ICO mania, I co-hosted a podcast called Chain of Thought. We interviewed twelve founders from projects like Golem and Augur. We didn't ask about token price. We asked about what happens when code replaces promises. We built an audience of five thousand listeners who wanted narrative depth over speculation, and I learned something that has never left me: a story without evidence is not an argument. It's a neurological event. So let's treat that headline the way a forensic analyst treats a confession: with empathy for the speaker, and zero trust in the transcript. What is the claim? That artificial intelligence is rewriting how enterprise software budgets get allocated. That nimble AI-native startups are siphoning spend away from legacy vendors. And that the incumbents, feeling the pressure, are slashing prices to hold onto seats. In broad strokes, the direction is plausible. In specific detail, the article is thinner than tissue paper. My own review of the piece — a title-level analysis, because that's all the first phase of information gave us — rates its confidence at D, medium-low. Directionally sound. Evidentially hollow. This matters far beyond the software aisle. Because if the seat license is dying, then the entire architecture of corporate trust is being renegotiated. And I don't mean "trust" as a warm, LinkedIn-buzzword trust. I mean the enforceable, auditable, settlement-grade trust that sits underneath every commercial relationship. That is exactly the territory where I have spent the last eight years. Trust is no longer a promise; it's a protocol. That sentence was true for money in 2020. It's becoming true for software in 2026. Let me unpack what I actually know, what the headline doesn't tell you, and why a crypto education founder is the one holding the flashlight. I ran my first serious audit of enterprise software pricing during the DeFi Summer of 2020. I organized a meetup series in Stockholm called Yield & Connect, and we hosted over three hundred people per event. The founding thesis was that liquidity pools could rebuild community trust. But the deeper observation — the one that stayed with me — was that the software itself was becoming a trust layer. Uniswap didn't ask you to sign a contract in a legal sense. It asked you to sign a transaction. That shift, from relational trust to cryptographic verification, was the quiet revolution. Nobody at the time connected it to the enterprise software that CIOs were buying in ten-thousand-seat increments. I am connecting it now. Here's the frame I want you to hold for the rest of this piece: the AI pricing shake-up is not primarily a technology story. It is not even primarily a software story. It is a story about how value gets measured, who gets to measure it, and which layer of the stack gets paid when measurement becomes automatic. The old model — the seat license — was a proxy. You charged per human head because you couldn't measure the actual value generated by that human clicking through your CRM, your BI dashboard, or your project tracker. The seat was a stand-in for value. It was imprecise, but it was simple, and simplicity was worth billions. AI breaks the proxy. If an AI agent drafts the contract, resolves the support ticket, generates the marketing campaign, and writes the unit tests, then the "user" is no longer the source of productivity. The model is. And you cannot charge per butt-in-seat when the butt belongs to an inference pipeline. The seat becomes fiction. What replaces it is a messy, fragmented, exciting set of pricing experiments that look, to my eyes, strikingly similar to the fragmentation of liquidity that DeFi went through in 2021. Back then, we were told that liquidity fragmentation was a disease that required new products to cure it. I said it was a manufactured narrative from VCs who needed a reason to deploy capital into another fork. The same pattern is emerging here: AI budget fragmentation is being framed as a problem for enterprises to solve with even more software. But fragmentation is not the disease. Fragmentation is the natural consequence of unbundling. What matters is not the fragmentation itself — it is who controls settlement when the fragments start moving. The analysis I performed on the Crypto Briefing piece yielded seven information points. They are: one, AI is reshaping enterprise software budgets. Two, upstarts are gaining share. Three, incumbents are cutting prices. Four, budget shifts are redefining market dynamics. Five, this favors nimble new entrants. Six, it forces established firms to innovate on pricing models. Seven, the underlying mechanism is a transition from buying features to buying outcomes. That seventh point is the load-bearing wall. Let me stress-test it, because I have audited enough DeFi protocols to know that when a wall looks clean, the cracks are usually hiding in the foundation. Buying outcomes sounds profound and customer-friendly. Pay per resolved ticket. Pay per generated contract. Pay per completed trade. It is, in theory, the ultimate alignment of vendor and buyer incentive. The vendor doesn't get paid unless the user gets value. In practice, outcome pricing is a transfer of risk. The vendor is now an insurer. And insurance, as any actuary will tell you, is a game of parameter estimation. If the vendor defines the outcome too narrowly, they leave money on the table and the client games the system. If they define it too broadly, they get crushed by ambiguity and client disputes. The measurement layer becomes the battlefield. And measuring outcomes requires infrastructure that the old software stack never needed. You need event streams that prove the ticket was resolved, not just closed. You need a feed that proves the contract was accurate, not just generated. You need reconciliation. You need dispute resolution. You need an immutable, auditable record of what the AI did, when it did it, and whether the result actually met the criteria. In other words: the pricing layer of enterprise AI turns into a settlement layer. And there is only one technology class that natively speaks the language of programmatic settlement. That is the insight the Crypto Briefing story completely misses. The headline treats the budget shift as a purely commercial event, a Darwinian reshuffle between agile startups and lumbering incumbents. But the structural question is deeper. When outcomes become the unit of exchange and AI agents become the counterparties, you need a system that makes promises enforceable without a human in the loop. That system is a blockchain. I know how that sounds. I have spent years being the guy who says "actually, this is a decentralization problem" at conferences full of enterprise architects rolling their eyes. But hear me out, because the technical logic is not ideological. It is practical. Consider what happens when an AI marketing agent negotiates a per-outcome deal with a content platform. The agent generates a hundred pieces of collateral. It agrees to pay a per-piece fee only if the pieces exceed a performance threshold measured by a separate analytics API. Who adjudicates the threshold? A human? The analytics API is itself an AI model exposed behind an endpoint. It can be gamed, hallucinated, or compromised. The content platform and the marketing agent are both non-human. They need a machine-legible, tamper-proof record of the performance metric at the moment of delivery. They need escrow that releases on verified outcome. They need a smart contract. This is not the story the article tells. The article tells a story about market share and discounting. It's a fine top-line story. It's just not the real story. The real story is that AI-native software pricing creates a machine-to-machine economy where traditional contracts, invoices, and procurement cycles are too slow. And the software industry's oldest pricing model — the seat — dies not because of competition, but because it becomes semantically meaningless. Let me now walk through the four pricing models I have mapped, because I believe this taxonomy is the closest thing to a technical core that this story currently has. The first and dying model is the seat license. Charge per human user. Stable revenue, easy forecasting, brutal when AI replaces the human. I have watched the numbers inside legacy SaaS companies: they are starting to report seat shrinkage quietly, buried in the footnotes of earnings releases, while hiding AI add-on attach rates in separate line items. The seat model does not fail overnight. It decays like a loan becoming non-performing. The seats don't vanish; they just stop being renewed. The second model is consumption-based pricing. Charge per token, per API call, per agent task. This is the market darling. It feels fair — use what you pay for. But it transfers forecasting complexity to the customer and margin risk to the vendor. Consumption pricing rewards aggressive usage and punishes optimization. If your AI gets more efficient, your consumption bill drops. If the model gets better at reasoning, it needs fewer calls to solve the same problem. Efficiency is a feature for the user and an annuity killer for the vendor. The cloud providers dealt with this by building integrations and switching costs. The AI-native software vendors do not have that luxury yet. The third model is outcome-based pricing. Pay per completed transaction, per accepted deliverable, per verified resolution. This is the most philosophically attractive and operationally terrifying. It requires a measurement layer that satisfies both parties. It requires contested outcome arbitration. And it requires the vendor to accept the risk that their AI simply doesn't work in a particular environment. In my experience auditing protocols, this is where the bright-eyed founders get eaten by the tail risk they forgot to model. The fourth model is the hybrid. Base subscription plus credits plus overage. This is what incumbents will actually sell. It preserves the recurring-revenue comfort blanket while allowing flexibility. It is also the most complex to design and the easiest to manipulate. I have seen credit systems engineered so carefully that they become another form of trapping the customer, just with a prettier name. In crypto, we call that a liquidity lockup. In SaaS, they call it a "credit pool." Same bone, different dog. The industry impact map is equally revealing. If I had to bet on where AI budget reallocation lands first, I would look at low-workflow-barrier categories: customer service, content generation, marketing design, code assistance, basic QA. These are high-volume, high-human-cost, easily-quantifiable outcomes. The replacement rate in customer service is already observable. The article's short-term impacts align with my independent observation of what happens when an API gets let loose on a ticket queue. But the article misses the second-order shock. When software budgets shift from traditional licenses to AI consumption, a significant portion of that budget doesn't stay in software at all. It flows to cloud providers and GPU suppliers. The enterprise software budget is becoming an inference compute budget. That is a transfer of value from the application layer to the infrastructure layer. The software company becomes a thin wrapper on rented intelligence. And if the wrapper's gross margins start to look like the infrastructure's gross margins, the valuation multiple starts to slide. This is not speculation. This is the same margin compression I've watched in DeFi yield protocols that depend on external incentivization — what looks like revenue is often just a pass-through with extra steps. There is a direct parallel to my own domain here, and it is the ZK rollup proving cost problem. For the last two years, I have argued that ZK rollup proving costs are absurdly high, and unless gas returns to bull-market levels, operators are bleeding money. The market nodded politely and returned to its price charts. But the same accounting logic applies to AI-native software. If your product's core value is delivered by an expensive inference call, then your gross margin is hostage to a commodity input you do not control. The software company that cuts its price per outcome but cannot cut its inference cost per outcome has built a beautiful money-burning machine. That is not a business. That is a charity with a courtroom. The incumbents understand this. That is why the headline says they are cutting prices. But I would reframe it. Incumbents are not cutting prices because they are weak. They are cutting prices because they have learned the Amazon playbook: commoditize the razor, own the blade. The razor is the seat. The blade is the AI workflow, the data estate, the compliance scaffold, the audit trail. By cutting the seat price, they buy the right to sell the outcome metering. The question is whether they can build the metering before the upstarts reach escape velocity. Let me give you the competitive matrix as I actually see it, without brand names, because the article gives us none anyway. AI-native upstarts have product velocity, pricing agility, and the advantage of a blank ledger. They do not have distribution, trust, compliance, or the embedded workflows. They can win a pilot in two weeks. They will struggle to win a seven-year procurement cycle with a healthcare enterprise. Enterprise-grade retention is a different sport from acquisition. And in a bear market — which, let me remind you, is where we are — the cost of acquiring an enterprise customer without a credible compliance story becomes prohibitive. Mature incumbents have distribution, data, integration depth, and cash flow. They have everything except organically AI-first architecture. Their traditional seat revenue is eroding underneath them, and their first instinct is usually to bolt an AI copilot onto the side of the legacy product and call it innovation. That works for one earnings cycle. It does not work for a structural transition. Some incumbents will successfully rebuild from the inside. Most will either acquire the upstarts once valuations drop, or fade into maintenance-mode cash cows. Cloud and model platforms — the infrastructure layer — hold the most leverage in this entire reallocation. They own the inference tokens. They own the GPUs. They own the model APIs on which nearly every upstart depends. The upstarts get the customer relationship and the gross-margin risk; the platforms get the pass-through revenue and the pricing power. If the price of intelligence goes down, the platforms can sustain it better than any application-layer vendor. If the price goes up, the platforms capture the upside. This is the classic squeegee model, and it is the reason I believe the real winner is not the software company at all. Let me now say something contrarian, because this article needs a moment of uncomfortable honesty. The "upstarts gain, incumbents cut prices" narrative is, at this exact moment, unproven. It is a headline with a D-confidence rating. There are no hard numbers. No vendor discloses AI-specific ARR with clean definitions. No buyer survey from Gartner or IDC has yet shown a massive reallocation. What we have is anecdote dressed as trend, which in my world is called a "narrative trade." I have seen narrative trades dissolve when the actual data hits. In 2021, every VC believed in infinite user acquisition. In 2022, that belief was priced as catastrophe. The truth was somewhere in the middle, but the narrative whiplash destroyed portfolios that had no thesis, only direction. The contrarian view, which I hold, is this: the upstarts' budget wins may be real, but revenue quality will be poor. They will grow ARR quickly, then watch gross retention collapse when the customer realizes the output is 80% as good as the incumbent's, or when the outcome-pricing contract produces its first disputed invoice. Disputes kill net revenue retention. And in a bear market, retention is the only number that matters. The incumbents' price cuts, meanwhile, are not a sign of defeat. They are a signal of cash-rich positioning. The incumbents can operate at lower margins indefinitely. The upstarts cannot. An upstart with a 60% gross margin that drops to 40% after inference costs is a startup with a runway of twelve months. An incumbent with a 75% gross margin that drops to 65% is a company that buys back shares. That asymmetry is the entire ballgame. The incumbents are buying time. Time is exactly what they need to figure out the outcome-metering problem. And they will need to figure it out, because the accounting problem is real. How do you price an outcome when the outcome is generated by a model whose behavior is probabilistic? An AI agent that resolves a support ticket might resolve it ninety percent of the time. The vendor charges per resolved ticket. The customer complains about the ten percent failure. Who bears that cost? The vendor. So the vendor embeds a probability discount into the price. The customer demands transparency into the failure rate. The vendor shows logs. The logs are generated by the same AI that generates the tickets. Congratulations: you now have a circular trust problem. And circular trust problems do not get solved by more trust. They get solved by verifiable computation. That is the part that the enterprise software world has not yet admitted. The outcome-pricing transition is a cryptographic problem. To prove that a ticket was resolved per agreed criteria, you need an attestation of the event, a signed record of the criteria, and a mechanism for independent verification. To automate that verification at scale, you need deterministic, auditable state transitions. That is a blockchain. Not a token. A chain. I have spent 2026 running the Human-Centric Blockchain initiative, where we study how AI agents interact autonomously on-chain. We brought together five hundred developers and ethicists in Stockholm to discuss preserving human agency in an AI-driven economy. The most surprising finding from those conversations is not how much agents can do. It is how much they fight over settlement. The agents negotiate, execute, and then dispute. Dispute resolution is the unsolved problem. In the enterprise software world, they are about to rediscover every one of those disputes, just slower and with invoices instead of transactions. Here is the uncomfortable truth I want to leave with you: the blockchain industry spent the last decade trying to convince enterprises that money should be programmable. We were met with suspicion. Now AI is making everything programmable, including corporate outcomes, and the same enterprises are demanding the very thing we were selling. The seat license was a social contract. It relied on yearly renewals, relationship managers, and the implicit threat of switching costs. Outcome pricing is an algorithmic contract. It relies on metrics, metering, and enforcement. The social contract is too slow for machine-paced commerce. The algorithm needs a substrate. Let me be concrete about what this means for capital allocation, because I know the readers of this article hold portfolios, not just philosophies. In a bear market, survival matters more than gains. The data you need to judge which software company is bleeding is the same data you need to judge a yield farm: net revenue retention, gross margin after compute, gross margin after inference, and the ratio of marketing spend to net-new ARR. I watched a protocol lose forty percent of its LPs in a single week in 2022 because the yield dropped and the depositors had no loyalty. Software customers are no different. When the seat price drops and the outcome price rises, the customer's loyalty drops faster than your chart. I will say it plainly: if you are an investor looking at AI-native software companies, ignore the topline growth. Ask about the inference cost per unit of outcome. Ask about the dispute rate on outcome contracts. These are the same diligence questions I ask about DeFi protocols — what is the fee share, what is the treasury drain, what is the actor that keeps the machine alive. Without those answers, the headline is just a song. The policy angle will bite quietly too. Buyers in finance, healthcare, and government will demand data isolation, audit trails, and regulatory compliance. EU AI Act obligations, model filing regimes, and the simple reality that a U.S. public company cannot outsource core accounting to a black-box model will create de facto moats for incumbents who have compliance spine. The AI-native upstart that cannot produce SOC 2 and ISO 27001 will not scale past the pilot. Regulation is the slow-moving glacier that the narrative trade always forgets. But I am not here to be the dreary auditor at the party. I am here to tell you what the pivot looks like. I learned to stop preaching and start listening during the 2022 bear market. After the growth years, I burned out, stepped away from the charts, and spent three months in art installations across Europe, documenting a blog series called Finding Humanity in the Void. That hiatus rewired how I think about decentralization. It isn't machinery. It's the act of distributing trust so that no single point of failure — human or corporate — can freeze the system. AI-native software pricing is about to do the same thing to procurement that crypto did to banking: remove the trusted intermediary and replace it with a verifiable mechanism. The difference is that this time, the intermediary isn't a bank. It's the vendor's finance department. So when an incumbent software company tells you they are cutting prices, don't applaud the discount. Ask what the pricing model is underneath. If they are moving to outcome-based pricing, ask what the measurement layer is. If they can't describe it in terms of attested, auditable events, they are winging it. And if the upstart claims it will eat the incumbent's lunch, ask what happens when the incumbent calls the upstart's API a billion times a month just to test it. Ask who pays for that. Or better yet, ask who verifies the result. There is a reason I keep coming back to the verification problem. In 2024, with the approval of spot Bitcoin ETFs, I launched a webinar series for traditional finance professionals called The Ethical Investor. My co-host and I translated regulatory frameworks into stories about inclusion and transparency. We had over two hundred institutional players in the room in Dubai and Miami. And every single one of them asked the same question about AI adoption in their own systems: how do we prove to the auditor what the model did? That question is not a compliance footnote. That question is the product. The enterprise software company that solves the proof problem does not need to cut prices. It can name any price. Let me also address the elephant called open source. The article's logic implies that AI-native upstarts have an insurmountable cost advantage because they can build on open-weight models. That is partly true. Open models lower the barrier to entry. But they also lower the barrier to differentiation. If everyone uses the same open-weight model, the competitive moat reverts to distribution, data, and trust — exactly the incumbent's home turf. The upstarts win the first round on speed. The incumbents win the second on scale. The match goes to whoever builds the strongest metering apparatus. And you know where you can build a metering apparatus without asking permission? On-chain. Here is the cleanest version of the thesis, and I will put it in bold because it deserves to be the core insight of this entire piece: The pricing transition from seats to outcomes is a transition from social settlement to algorithmic settlement. Every outcome needs a verifiable event. Every event needs an attestation. Every attestation needs an environment that makes the record tamper-evident. The enterprise software stack has no such environment. The crypto stack does. This is not a bull thesis for a token. It is a functional argument for a settlement substrate. I am not saying every software company will put its invoices on a public chain. Privacy, latency, and regulatory requirements will keep many ledgers private. But the escape valve from the circular trust problem — where the AI both performs and reports the outcome — will be some form of verifiable, attested execution. That is the same logical space as ZK rollups. And I have watched the ZK space struggle with proving costs. I have watched operators bleed. I have watched everyone hope for the next bull market. That experience has taught me to respect the cost of verification. Verification is not free. The enterprise software companies transitioning to outcome pricing will discover the same truth. And they will discover it the hard way. Now, the contrarian angle in full, because I promised you pragmatism tests and not just philosophy. The pragmatic test for the upstart thesis is not whether they can win a pilot. It is whether they can survive a nine-month procurement freeze. When the bear market bites, enterprises delay decisions. AI budgets get frozen along with everything else. The upstart with six large logos and negative churn is the upstart with eighteen months of cash and no margin for error. The incumbent with ninety logos can wait. This is the difference between activated and actualized value, and it is the same difference between a token's market cap and its realized volume. The pragmatic test for the incumbent thesis is whether their price cut is improving net revenue retention or just papering over seat decay. I have audited companies — not in crypto, but in SaaS — and the hard lesson is always the same: a discount is a delayed revenue problem. You can cut the price to hold the customer, but you have The pragmatic test for the intermediary thesis — the one I am advancing — is whether the cost of verification will drop faster than the price of outcomes. I believe it will, but only in specific areas. Model routing, caching, quantized models, and specialized small models are making inference dramatically cheaper. The same trajectory is hitting proving systems. The unit economics of verification are improving. But they are improving from a high base, and the price of software is being pushed down by competition. The margin squeeze may last longer than crypto-native humans expect. Let me also acknowledge the risk that I am completely wrong about the settlement layer. It is possible that the enterprise world simply refuses to use blockchain rails, and instead settles AI outcomes through old-fashioned centralized audit processes. In that world, the winners are not the method actors of decentralization. The winners are the centralized oracle vendors, the escrow services, the reconciliation platforms. It is possible. But I would note: centralized audit proved fragile in the 2008 crisis. It proved fragile in the 2022 crypto contagion. It is proving fragile every time a software vendor disputes an AI outcome and there is no senior engineer who can explain what the model did. The human layer is the failing layer. The writers of the Crypto Briefing piece may not realize they were writing about the death of the human layer in procurement. They thought they were writing about budget share. But every technological shift has a dehumanizing undercurrent, and the one thing blockchain offers that no other stack offers is the ability to preserve human agency without requiring human presence. That sounds paradoxical, but I have been inside enough protocols to know that the curse of decentralization is not machines taking over — it is machines being indecipherable. My 2026 manifesto, The Soul of the Code, argued exactly that: verifiability is a human right. If an AI commits a company to spend money, the human affected by that AI spending deserves a witness. Not a log. A witness. So here is the takeaway, and I will make it as direct as I can. When you read the next headline about AI reshaping software budgets, do not ask whether the upstart or the incumbent won the quarter. Ask who settled the outcome. Ask what the measurement architecture looked like. Ask whether the pricing was a discount on a seat or a payment on a proof. The companies that survive this transition will be the ones that treat the pricing problem as an accounting problem and the accounting problem as a settlement problem. The software industry is about to discover that selling outcomes is harder than selling features. It is going to require a discipline new to the application layer. In my world — the trustless world — that discipline is not new. We have been building it for ten years. I have spent a decade evangelizing decentralization, and I have learned that trustless systems require trusting relationships. This is the secret of the entire transition. The enterprise world maliciously trusts its software vendors because it pays them not to betray. The crypto world is built on verification because it cannot pay enough to prevent betrayal. AI-native software pricing merges these two worlds. The contract runs itself. The outcome measures itself. The dispute settles itself. You cannot stop this transition any more than you could stop the internet from unmaking the travel agent. Code is law, but empathy is the interface. Keep that ready, because the budget wars are about to teach the enterprise how much of its legacy procurement was going to human work that never should have existed in the first place. The AI upstarts are not gaining because they are magical. They are gaining because they point at the wasted seat and finally press the delete key. The incumbents are not cutting prices because they are generous. They are cutting prices because the seat is a blade that is already dull. The question is not whether software budgets will be reshaped. They already are being reshaped, quarter by quarter, contract by contract. The question is whether anyone will bother to build the verification layer that tells the truth about what the outcome was worth. And that layer, I promise you, will not be another seat. It will be a protocol. The pivot wasn't from SaaS to AI. The pivot was from trusting the vendor to verifying the outcome. I built my career on the belief that this moment would come. Now that it's here, the only disappointing thing is that nobody in the enterprise software press has said it yet. So I will say it for them. The seat license is dead. What replaces it is not a subscription. It is a settlement. And the winners will be the ones who understand that the product is not the software. The product is the proof. Everything else is just a headline.