Page 1 • Showing 9 tweets
Does a 10x AI engineer require 10x the token budget? It is very, very easy for a AI-pilled SWE to burn through 100-250K worth of tokens a year rn. New budget item for CFOs to figure out...
I am really, really tired of software engineers saying "we don't even code anymore, I'm agentmaxing to be more efficient" and then don't say how much they are spending on tokens! Efficiency is not solely measured by time!! Drives me nuts.
Got a 3,500 word banger coming out tomorrow. Represents months of interviews, customer calls, and late night writing by me. The age of robots has come.
📝 Don't Sell the Work For the last four years, the prevailing theory in Silicon Valley was that software companies should “sell work, not software.” The overly simplistic explanation of the idea is that as AI got better at AIing, customers would need fewer employees, and therefore companies who charged on a per seat basis would be screwed. By selling work, a startup was offering “outcome based pricing” whereby they could capture a percentage of a customer’s payroll versus competing for scraps among a firm’s IT budget with the rest of software. This is an elegant idea. It neatly solves the problems that AI introduced to software’s business model, and like all of Silicon Valley’s most important theses, it could theoretically make founders several yacht’s worth of cash. The narrative it introduced rebounded throughout the valley. Sequoia adopted a “tools → copilots → autopilots” framework with outcome-based pricing as the evolutionary endpoint. a16z began framing the AI era as a shift from “Software-as-a-Service” to “Service-as-a-Software.” And everyone, including myself, nodded in agreement. Unfortunately for all of us, this elegant idea turned out to be very wrong. There is exactly one category of startups where the model has been effective—AI customer service—and every other major AI application today uses a different positioning and pricing strategy. So what happened? Why has “selling work” been such a dud? Outside of mere intellectual curiosity, the answer to this question really matters right now. If you buy my argument from last week, namely that the sell-off in software stocks was justified because profit pools would shift to a new type of applications focused on the context layer, then we need to understand how these new apps should be monetized. Just how exactly are these new types of companies going to make money? Why hasn’t selling work, uh, worked? The answer is simple: there is no actual “work” to sell. The opacity of production The counter-arguments to “sell work” have mostly been economic. Margins are bad—50-60% versus 80-90% for SaaS. These critiques are valid but insufficient. The deeper problem is ontological. When you hire McKinsey, you can’t reverse-engineer what they did into a bill of materials. The production process is opaque—you’re paying for judgment, relationships, and institutional knowledge bundled into a deliverable whose value is difficult to quantify. That opacity is what sustains pricing power. AI work has no such opacity. The production process is increasingly transparent. Sophisticated buyers can see the API calls, estimate token counts, and calculate the rough cost of what you just delivered. The “work” firms are charging for — a market analysis, a contract review, a support resolution — is a markup on inference wrapped in varying degrees of prompt engineering, workflow orchestration, and domain context. And unlike McKinsey’s black box, the buyer can peek inside this one. This is why the “it’s just tokens” critique sticks in a way that “it’s just compute” never stuck against SaaS. Everyone knew Salesforce’s marginal cost per seat approached zero; that was the whole bull case for software! But it didn’t matter. You couldn’t replicate Salesforce’s value by renting your own servers from AWS. The pricing power came from platform lock-in, data gravity, integrations, and switching costs, none of which had anything to do with the cost of production. AI work is different. The production process is reproducible. Your customer can see that you’re calling Claude with a system prompt and some retrieved context, and increasingly, they can do exactly that themselves. The API is right there. The prompt engineering isn’t patented. The workflow orchestration is a weekend project for a competent engineer. When the buyer can not only see your costs but replicate your production process, the opacity is effectively zilch. And this matters because of what sits underneath. Price is only partially determined by the value of a product. A pricing model is a reflection of the scarcity in your supply chain. When your production process is reproducible and the inputs are abundant (tokens are hardly hard to come by nowadays) you’ve lost pricing power from both directions. Take KPMG’s audit group. They forced their own auditor to give them a 14% discount because they know that AI is doing most of the work, and the work is cheap. (Yes, there is some irony in KPMG demanding audit discounts while not extending the same courtesy to their own customers). This is what happens when the opacity that sustained professional services pricing disappears. The customer benchmarks your price against the visible cost of the underlying unit of production, not against the value of the output. Inference costs dropped roughly 99% per token between 2023 and 2025. Any business model that prices against the value of human labor while its input costs follow Moore’s Law trajectories is building on a foundation of perpetual margin compression. Your customers will demand that you pass those savings through. The scissors problem But even if you could maintain the opacity, the economics of “selling work” have a structural problem that makes the model worse over time, not better. There’s a scissors dynamic. On one blade, inference costs per token keep falling. On the other hand, token consumption per task is exploding. OpenRouter’s analysis of over 100 trillion tokens in December of last year, shows that reasoning models now account for over 50% of all tokens processed. Average sequence lengths have more than tripled over the past 20 months, from under 2,000 tokens to over 5,400. Programming workloads routinely exceed 20,000 input tokens per request. A reasoning model can consume 10,000 internal “thinking” tokens to produce a 200-token answer. Ask any founder of an AI application today if their token costs have gone down over the last few years and they’ll laugh in your face. Sure, tokens are cheaper, but the frontier models from OpenAI and Anthropic charge a hefty markup on them, and the work we are demanding from these models is forcing ever more token consumption. Sophisticated customers demand being on the smartest models because why shouldn’t they? This is the treadmill that “sell work” companies are trapped on. You save 50% on per-token costs, but your customers now expect agentic, multi-step workflows that consume 10-100x the tokens. The price per token drops; the tokens per task skyrockets. Net costs stay flat or rise — while your customer still benchmarks your price against the ever-cheaper unit cost of the commodity underneath. The margin improvement that was supposed to rescue outcome-based pricing never arrives because the goalposts keep moving. The quality specification problem To make the case even worse for “selling work” is that even if you could stabilize the pricing, “work” requires something that tokens cannot provide: a contractual specification of quality. The reason firms hire lawyers at $800/hour rather than paying per brief is that the quality of legal reasoning cannot be specified ex ante in a contract. You’re buying access to judgment, not a deliverable. You are buying lawyers who all went to Harvard Law because that carries with it a promise of indefinable quality. The same problem reappears with AI. If you’re “selling work,” you need a contract that defines what good work looks like, and that specification problem is often as expensive as the work itself. Ronald Coase explained decades ago that firms exist because they reduce the “costs of using the price mechanism.” Subscription pricing is the software equivalent — it eliminates per-transaction negotiation, measurement, and verification costs. Per-task or per-outcome pricing reintroduces all three. Every “outcome” must be defined, measured, verified, and attributed. The AI vendor knows the actual difficulty and cost of completing each task; the customer does not. Classic principal-agent problems emerge on both sides. This creates what economists call a moral hazard in both directions. Vendors paid per “completed task” face incentives to use cheaper models, minimize compute, skip quality checks, and classify ambiguous outcomes as successes. Customers face incentives to overload systems with low-value tasks or dispute outcomes to avoid payment. You’ve rebuilt the adversarial dynamics of the consulting industry, except with less accountability and no reputational mechanism to discipline quality. Subscriptions or per-token credit systems sidestep all of this. You’re buying access to the tool. Quality assessment becomes the buyer’s problem, not a contractual negotiation. The entire transaction cost structure is simpler. The binary exception proves the rule There is exactly one domain where “sell work” has produced explosive growth: customer support. Sierra AI grew from roughly $20M to $150M ARR in about 15 months, charging ~$1.50 per resolution. Decagon grew from ~$6M to ~$35M ARR using per-conversation pricing. Intercom’s Fin at $0.99 per resolution drove 40% higher adoption. The only other prominent case study I could find is legal tech firm EvenUp which charges per brief generated, but even then customers have a minimum they have to purchase per month. But customer support isn’t really “selling work” in the way the argument orginally meant. A resolved support ticket is a binary state change. Did the customer’s problem go away? Yes or no. The outcome is measurable not because the “work” is well-defined, but because the absence of a problem is well-defined. Attribution is unambiguous: the AI either handled the ticket or it didn’t. These conditions don’t really exist anywhere elsewhere. What’s a good contract clause? What’s a high-quality line of code? These are judgment calls and judgment is precisely what’s hardest to contractually specify and verify. The moment you move from binary state changes to quality spectrums, outcome-based pricing falls apart. Meanwhile, seat-based AI is eating the world The irony of the “sell work” thesis is that the fastest-growing AI companies of the past three years mostly ignored it. Harvey AI uses traditional seat-based pricing. It grew from roughly $50M to $190M ARR in 2025, with median seat counts doubling within 12 months, and reached an $8-11B valuation. Cursor uses tiered subscriptions with compute credits. It became arguably the fastest-growing B2B SaaS company in history, reaching approximately $1B ARR in 24 months. Microsoft Copilot charges $30/seat/month and reached 100M+ monthly active users. The pricing model was not the differentiator. Product quality, distribution, and domain-specific value drove growth regardless of whether companies charged per seat or per usage unit. Harvey’s seat-based model created strong net revenue retention as firms expanded licenses. Cursor’s subscription model created predictable revenue while allowing usage flexibility. Neither needed to solve the quality specification problem because they sold access to tools, not deliverables. And note what these companies do have that a raw inference wrapper does not: deep workflow integration, proprietary data pipelines, purpose-built UX, and switching costs that compound over time. The production of AI work — the inference itself — is reproducible. The production of AI products — with their context layers, workflow lock-in, and accumulated user data — is not. The confusion at the heart of the thesis The sell work thesis confuses the application of intelligence with the output of intelligence. Professional services solved the quality verification problem not by making outputs easier to evaluate, but by building accountability infrastructure around the producer — licenses, liability, long-term reputation. AI has the capability to produce the output but none of the scaffolding that lets a buyer trust it. It’s the most competent worker in the room who can’t sign the contract, can’t be sued, and won’t remember the engagement next week. Until that’s solved, the human professional’s role shifts from doing the cognitive work to bearing responsibility for it. The winning AI companies will sell software that does work and maintain software economics through workflow lock-in, data moats, and subscription revenue — rather than work done by software, which is just consulting economics and problems in an API trenchcoat. So what does monetization actually look like? If everything is tokens of context in and tokens of intelligence out, where does the pricing power come from? Not from the tokens. Inference is a commodity, getting cheaper by the quarter, and your customers know it. Not from the “work” — which is nebulous, hard to specify, and invites the adversarial dynamics of every consulting engagement ever. The pricing power comes from the thing that makes the tokens useful in the first place: context. As I argued in “Context is King,” the software stack is splitting into three layers: systems of record (databases), point solutions (the interface layer), and a new middle layer — the context layer — that holds the institutional knowledge telling AI agents what to do, in what order, and whether they’re allowed to do it. The “sell work” thesis was an attempt to answer the monetization question, but it landed on the wrong answer because it focused on the output rather than the input. The provider sells the conditions for good work, not the work itself. This is why margins are a lagging indicator of where value accrues, not a leading one. Scarcity drives pricing more than anything else, and inference is abundant. Context is scarce. The “sell work” companies are stuck arguing about gross margins on a commodity. The context layer companies are building something that appreciates. The real lesson The thesis was right in one important way—AI does expand addressable markets from software budgets into labor budgets. The TAM expansion thesis holds. AI companies should think about replacing work, not merely augmenting workers. The context layer will capture much of the payroll budget that went to project management and operations. We already have early evidence of this happening. A recent paper out of Ramp, “Payrolls to Prompts,” puts hard numbers on this. Researchers tracked firm-level spending on freelance labor marketplaces like Upwork and Fiverr alongside spending on AI model providers from Q3 2021 through Q3 2025. After ChatGPT launched, the firms most reliant on contracted online labor adopted AI earlier, spent more on it, and cut their freelance budgets by 15%. For every $1 firms stopped spending on human contractors, they spent just $0.03 on AI. “ Keep in mind that this is all firms, for the ones that have gone all in on AI, the effect is more dramatic, “More than 50% of businesses that had spent on online labor marketplaces in Q2 2022 spent 0% in Q2 2025, whereas roughly 80% of businesses spent between 0 and 5% of their total spend on AI model providers in Q2 2025.” Companies are pulling directly from payroll and contractor budgets, exactly the kind of budget migration the “sell work” camp predicted. They just proposed the wrong business model to capture it. But “sell work” is the wrong business model for capturing that opportunity. It sounds like a paradigm shift, but it’s actually a regression to the services economics that software was invented to escape. The frontier AI companies already understand this. They sell subscriptions that give customers access to intelligence-on-tap and let the customer define what “good work” means. They’ll maintain 70-90% gross margins, predictable revenue, and deep switching costs. They look like the best software companies in history, not outsourcing firms with better technology. The market has also spoken. ~92% of AI companies now use hybrid or subscription pricing. Pure outcome-based models represent just 7%. The “sell work” era lasted about as long as it took buyers to realize they were paying a markup on tokens and to start demanding token-level pricing. The next time someone tells you to “sell work, not software,” ask them one question: can you write a contract that specifies what “good work” looks like for your AI? If the answer involves a spectrum of quality rather than a binary outcome, you’re not selling work. You’re selling tokens at a markup. And your customers will figure that out faster than you’d like. If you want to read more from me, subscribe at gettheleverage.com http://x.com/i/article/20245139936377856…
Brutal result for Brex's investors! First-mover advantage is overrated
OpenAI banned racist MLK videos on Sora (as they should) But they're now allowing erotica And still plenty of other racist content Seems contradictory, right? Well, social platforms benefit from allowing content that goes right up to the line (and sometimes over it) Whichever platform allows users go the furthest to the right on this chart gets a massive boost in viewership
📝 Context is King To the individual who enjoys Patagonia vests and the sexual high that increasing shareholder value brings, there is nothing more beautiful than a software subscription business. From 2005-2023, a twinkling flutter of a moment in capitalism’s ascendency, Silicon Valley invented the most powerful business model ever seen on God’s green earth. SaaS seemed divinely sculpted, a kismet meeting between the needs of venture capitalists to deploy capital and of startups that would need Brinks trucks worth of cash to build products. The margins were judicious, ranging above 85%. The terminal value was indisputable, with customers paying you 10% more every year, forever. It simply…worked. Until last week, when the market said, “Nah.” $300 billion in software companies' market capitalizations vanished, affecting everyone from new darlings like Figma to legacy giants like Salesforce. While the market is mysterious, this selloff is likely due to fears over AI generated code decreasing the value of software companies. Why buy a CRM when you can just make one? This thesis is too simplistic. In the idea’s defense, there is lots of data to make it seem like that is right. Code is looking like more of a commodity input—4% of GitHub commits are already generated by Claude Code, with SemiAnalysis forecasting that to hit 20% by the end of the year. In my own experimentation, it is remarkably easy to generate applications shaped specifically to my needs. If you give Claude a set of instructions, it can do a halfway decent job of recreating your digital labor. AI doesn't make software companies less valuable. It simply makes a different group of software companies more valuable. AI increases the power of software, not decreases it, and AI moves where in the stack the value lives. The SaaSacre is partially justified. Many of the SaaS giants of yesteryear should be valued less. But what's actually happening is a culling of the herd, a separation. Software is splitting into three layers with fundamentally different economics. Two of those layers are getting commoditized. The third—a layer that barely existed before AI—is where the value migrates. I'm calling it the context layer. The valuation problem To understand my idea, first we need to establish why the generic software companies aren't as valuable as they used to be. The answer is threefold: The growth sucks: The top quartile of public SaaS companies today grows slower than the bottom quartile did in 2016. Public SaaS growth halved from 36% to 17% since 2023. Growth is king in technology and these companies don’t have the juice anymore. AI native private companies are where much of the growth in software markets are happening. The margins are structurally worse: In SaaS, you built it once, and distributed/used it at zero marginal cost per additional customer. In AI-land, you build it once, distribution is more expensive, and each use has a marginal cost. AI-first applications currently operate at 50-65% gross margins, not the 75-85% that justified SaaS premium multiples. Importantly, I would expect these gross margins to improve over time as the cost of intelligence decreases. But that doesn’t matter! That there is a “cost-per-action” associated with the usage of your software dramatically reworks the economic logic. Switching costs are decreasing, wrecking SaaS platform’s strategic power: This is the underappreciated one. LLMs make it much, much easier to switch data over to a new application. Switching costs are moved to data access and organizational context. There are still good reasons to not vibe code a new application—either due to the opportunity cost or you want to let other platforms be responsible for certain types of risk, such as payments or user identity, really anything where being wrong is catastrophic. But that is a small subset in the world of software! And as agents improve, the organizational context will get easier to switch over too. Essentially, this compresses every lever in the typical SaaS company's valuation. Growth, margin, and terminal value are the three things that mostly determine a company's long-term worth. For the majority of SaaS companies, those are materially worse and will not improve. But this also introduces a paradox: How can AI simultaneously make software significantly more useful while making it a materially worse business? Does it all just go to consumer surplus? In answering these questions, we can understand what the context layer is and why it is so valuable. The three-layer stack OpenAI's new Frontier platform describes itself as an "intelligence layer" that connects "siloed data warehouses, CRM systems, ticketing tools, and internal applications." While this does read like secret diary entries of an MBA, behind the jargon is something important. The platform’s goal is to make every software company commodity plumbing. Simply dumb pieces of software that are there to help pipe intelligence around the company correctly. Frontier, and other platforms like it, simplify the tech stack into three layers: Layer 1: Systems of Record (the database layer). The auditable, regulated data stores. Financial ledgers, healthcare records, compliance systems. Enterprises aren't replacing these — they're layering intelligence around them. Oracle and SAP didn't die when we switched to cloud, but they stopped commanding premium growth multiples. So too with cloud systems of record like Salesforce, Netsuite, etc. They are sticky, hard to sell, and are subject to increased competition, but these databases have to exist so AI agents can have something to reference. Layer 2: Point solution applications (the interface layer). The software humans interact with directly. Analytics packages on top of systems of record or project management tools, the stuff that is defined by per-seat pricing. This is the true SaaSacre zone. Tools that serve an individual user, that have no complex security or permissioning needs are doomed because they are relatively simple to spin up. The world that many SaaS companies were designed for simply doesn't exist anymore. A significant portion of their workflow can/will be done by agents, driving value away from the software provider and towards the token provider. Legacy providers have bridged this gap with their pricing by bundling per seat human subscriptions with a per token credit system. However new entrants are leaning all the way into it, not bothering to charge per seat anymore and just charging on the basis of tokens used. But we should strive to be very, very precise here: generating code is commoditized. Governing code in production—knowing what should exist, what databases/systems of records it connects to, who's allowed to change it—is not. And those crucial questions are only answerable in the context layer. Layer 3: The Context Layer (the new middle). The industry already has a name for the space between databases and applications: they're calling it the "orchestration layer." But orchestration is just a fancy word for a type of plumbing, and plumbing commoditizes fast. All the current leaders from startups like LangChain, to MCP, or even Google's agents protocol are racing toward interoperability. The orchestration frameworks will be cheap. What won't be cheap is what directs the orchestration. The institutional knowledge that tells agents what to do, in what order, and whether they're allowed to do it. Before AI, every company had an invisible, accidental version of this to direct humans: the email threads, wiki pages, Slack channels, onboarding docs, and tribal knowledge where organizational truth actually lived. It was never structured. Never searchable. Never maintained. It was just the ugly overhead that required companies to hire more people than they wanted. After AI, this overhead becomes real software, and it becomes the most important software in the stack. Not because the individual emails are all that precious, but because it's the raw material for a living model of how your organization actually operates. Who has access to what data, who can do what with it, what sequence of steps actually gets a deal closed or an incident resolved. The context layer is the directing layer, the thing that makes the difference between an agent that takes action and an agent that takes the right action. Best of all, this is a compounding asset. Every time an agent executes a workflow, it generates traces that feed back into the context layer, making the next execution smarter. That's context. Context is the institutional knowledge that makes coordination valuable. Systems of record can't provide it—they store data, not meaning. Endpoint applications can't provide it—they're just interfaces restricted to their own domain. The context layer is what makes AI agents productive instead of just active. It is the context of how humans work, but over time, it becomes the context of how agents work even better. You could argue this is just a fancy way to describe a bunch of markdown files. After all, Anthropic shipped its Cowork plugins as literally that. But there's a critical difference between context you can write down and context that has to be learned. A markdown file can describe your sales process. It can't encode that deals over $500K stall when legal reviews before procurement, or that your best AE skips the discovery call for inbound leads from existing customers. That kind of process knowledge doesn't live in a document. It emerges from thousands of workflows executed over time. The markdown file is a snapshot. Context is what gets the margin that SaaS lost The playbook for what happens next was written in 2003. (Nobody in Silicon Valley has read it because it wasn't a tweet.) Clayton Christensen called it the Law of Conservation of Attractive Profits: when one layer of a value chain commoditizes, the adjacent layer de-commoditizes. IBM’s hardware commoditized, and value migrated to Intel and Microsoft. If applications and systems of record just become the cost of the tokens to create them, and are thus commoditized, the value must migrate to the layer between them. The reason is structural. At any point in time, there is one thing that is the main bottleneck in a technology stack. Everything else gets cheaper and more interchangeable so that bottleneck can be solved. Right now, the thing that matters most is the connection between how AI models are trained and the agent systems that actually use them. That's where the real performance gains are. For that connection to improve, everything around it has to get out of the way. Databases become interchangeable inputs. Applications become disposable interfaces. Building software isn't the hard part anymore. Directing it is. The context layer sits at the new bottleneck. Here’s the important part: the context layer doesn't fully compete with existing software spend. It replaces coordination overhead that companies just accepted. It takes money from the payroll budget, not the IT one. If that sounds too clean, consider what the alternative looks like where you pay a bunch of MBAs to do fake email jobs and make powerpoints, all just to make sure that your company doesn’t go off the rails. Most importantly, it compounds. Every other layer is getting easier to swap out. The context layer only gets harder. The more an organization's nuance gets encoded—definitions, process logic, team-specific meaning—the more switching costs increase. The context layer increases the importance of software by turning a category of organizational cost that was previously just called "overhead" into software margin for the first time. The battle for the context layer The unusual thing about this moment is that the biggest question in enterprise software is genuinely unanswered. There is a trillion dollars sitting on the ground, waiting for someone to pick it up. There are no clear winners yet. This is mostly because the details of the context layer are incredibly murky right now. We can look at the primary competitors as a lens by which to interpret the potential sources of power. In traditional SaaS, ServiceNow has the deepest enterprise workflow penetration and just announced collaborations with both OpenAI and Anthropic—it already holds decision logic and process definitions for thousands of large organizations. Notion holds institutional knowledge for a different segment: the companies that run on wikis, docs, and databases. They have a new agents product coming soon that will directly target the context layer. Glean is evolving enterprise search into enterprise context — starting from the problem of "find what we already know" and building toward "know what we mean." Their bet is that if you can find the organizational knowledge, adding agents on top is a natural extension. The foundation models are also building towards this. OpenAI Frontier wants to be the semantic operating system connecting databases, CRMs, and internal apps, then deploying agents that accumulate institutional context over time. Anthropic's Cowork plugins target professional workflows directly, bypassing the application layer entirely. Their bet is that if you build the most capable agents, the organizational knowledge accumulates to you. It is easy to see them extending Cowork into a true enterprise platform. The question that no one knows the answer to is this: Does the context layer get owned by the company that already holds the knowledge, or the company that builds the most capable agents? The answer probably varies by company size, industry, and how centralized their existing knowledge infrastructure is. But the battle for this layer is now the most important strategic competition in enterprise software. The close Christensen's law says that when a layer commoditizes, the adjacent layer captures the margin. Code is commoditizing. Databases are commoditizing. The layer between them—the one that holds organizational meaning, permissions, and institutional judgment—is where the new value forms. The SaaSacre isn't the death of software. It's the birth of the layer that makes software actually work. Systems of record are just databases. Endpoint applications are just code. And both of them are just tokens. Context is king. Thank you to Akshay Kothari, Buck, Tina He, Nathan Baschez, and a few others who rather remain anonymous for their feedback on this idea! They are all worth a follow. http://x.com/i/article/20219853548997632…
There is a reason that Substack abandoned their once so principled stance against ads last week—you can only scale premium subscriptions to a certain point. To make a content catalog competitive, you have to offer an emotionally charged, ad-subsidized access point.
“AI Agents will replace human workers” Let’s talk about that. Here’s a philosophical question an economist might ask: Why do companies even employ people? Why do they spend tons of money on paychecks, benefits, offices, and a million other things to employ people? After all, if markets are efficient, shouldn’t it always be cheaper to outsource your operations to the best available external provider? Ronald Coase’s Theory of the Firm proposed a brutally practical answer: Because the cost of using the market (finding vendors, negotiating, coordinating) is often higher than doing it inside the org. In other words: Outsourcing often has lots of hidden costs. This idea matters because while we’ve clearly established that AI can produce individual pieces of code or content materially cheaper than human beings, we have yet to show that the coordination costs actually decrease within a firm. AI Agents are doing tasks efficiently; but we still have the same costs for human employees to coordinate the AI agents’ work and align it with the goals of the company’s leadership. If you believe in the theory that AI allows companies to be much smaller than before, you are actually saying that you think internal coordination costs are going to dramatically decrease. Otherwise every additional meeting gets exponentially more expensive as your staff get more and more leverage out of their time spent. Allow me to ask this question in another way: What happens when “the market competitor” isn’t another vendor or headcount but a meter - AI agents that can log in, click, remember, and obey rules? If metered compute + a little supervision costs less than new payroll or new vendors for the same reliability, the rational move is unfancy: don’t hire someone; meter an AI agent. So AI Agents won’t replace humans when they can do tasks better or faster than humans (they already can in many cases). They will replace humans when coordination time between AI agents & the company using the agents falls below the status quo.