The scale advantage inversion
The question we were left with at the end of the last piece was deliberately uncomfortable: if the roles most vulnerable to AI were specifically designed and selected for over a century to sustain a particular kind of organizational structure, what happens to that structure when the economics of those roles change?
It turns out that’s exactly the right question. It’s also the one the current debate seems to have not yet recognized.
Because the job anxiety people are feeling right now is real, but it may be a surface symptom of something operating at a much larger scale. The process layer that AI is compressing didn’t just exist inside professions. It was the operating system of modern organizations. The mechanism by which large institutions coordinated hundreds of thousands of people toward reliable, repeatable outcomes. Strip out the economics that made that layer necessary, and you’re not just changing what some roles look like. You’re changing the fundamental logic that justified building large organizations in the first place.
And when you start looking at what’s actually happening at the edges of the market right now, the data suggests that shift isn’t coming. It’s already underway.
Large organizations didn’t come to dominate the last century because they were smarter or more creative than smaller groups. They dominated because they controlled three structural advantages that smaller groups simply couldn’t replicate: labor at scale, concentrated expertise, and infrastructure. If you needed more output, you hired more people. If a problem required specialists, large organizations could afford to collect them. If the work demanded expensive tooling, complex systems, or distribution infrastructure, large firms were the only entities that could assemble and maintain it.
That’s the architecture modern work was built around. The hierarchy wasn’t a design preference. It was a rational response to the physics of coordination under those constraints. And the process roles that filled that hierarchy, the execution roles, the consistency roles, the roles that kept everything moving reliably at scale, weren’t arbitrary either. They were the human implementation of a system that couldn’t function without them.
AI doesn’t just automate some of the tasks those roles perform. It starts changing the underlying physics those organizations were built around. And it does it in three directions at once.
— — —
The First Shift: The Large-Team Advantage Gap Starts Collapsing
AI increases the capability of individuals. That much is obvious. But the deeper point is that once individuals become dramatically more capable, several of the historical advantages of large organizations begin to weaken simultaneously.
The expertise gap starts compressing first. In a large field study of more than 5,000 customer support agents, NBER researchers found that generative AI increased productivity by 14% overall, but by 34% for novice and low-skilled workers, largely by helping them adopt patterns used by stronger performers. That’s not a minor efficiency gain. It’s AI disseminating institutional know-how in real time, narrowing the experience gaps that large organizations spent decades building moats around.
The infrastructure gap weakens too. Capabilities that once required specialized teams or expensive software stacks are increasingly accessible through AI-native tooling. The threshold at which a small team can act like a much larger one keeps dropping. And the revenue data is starting to show it: the top 100 AI companies on Stripe reached $1 million in annualized revenue in a median of just 11.5 months, roughly four months faster than the fastest-growing SaaS companies at the height of the subscription boom. Younger AI companies are hitting milestones about three times faster than pre-2020 cohorts.
The gap between what a large organization can do and what a small team can do doesn’t disappear. But it gets much smaller, much faster than anyone expected.
— — —
The Second Shift: The Small-Team Advantages Start Compounding
Small teams already had real advantages before AI. They move faster. They coordinate more easily. They make fewer handoffs. They maintain tighter strategic coherence. They don’t need six meetings and three approvals to change direction.
AI doesn’t remove those advantages. It multiplies them.
That’s the part people keep missing. When each person on a small team becomes dramatically more capable, the team doesn’t just become “more productive.” It becomes disproportionately stronger relative to larger organizations, because it gains new capabilities without inheriting large-organization friction. A single developer can now explore architectural ideas that previously required multiple engineers. A small design team can generate and test dozens of concepts in the time it once took to produce a few sketches. A founder with the right tools can run experiments that previously required entire functional departments.
The numbers that have started coming in are hard to dismiss. Lovable hit $100 million ARR with 45 full-time employees. ElevenLabs reached a similar range with around 50. Cursor is widely cited as one of the fastest-growing software companies ever, crossing $500 million ARR with a team that would have looked absurdly small by any previous standard for that revenue level.
This isn’t proof that every AI company stays tiny forever. It’s something narrower and more important: we are already seeing examples of unusually small teams hitting unusually large numbers at unusually high speed. The underlying structural question those examples raise is no longer theoretical.
— — —
The Third Shift: Large-Team Weaknesses Get Amplified
This is the part that makes the thesis hard to dismiss as startup optimism.
Large organizations don’t just lose relative advantage on one side while small teams gain it on the other. Their existing structural weaknesses get amplified at the same time. Coordination overhead is the obvious example. HBR has reported that time spent by managers and employees in collaborative activities has ballooned by 50% or more over the past two decades, with many people spending around 80% of their time in meetings or responding to colleagues. That was already a problem before AI.
Now imagine each individual contributor becomes significantly more capable, but the organization still has to funnel decisions through layers, meetings, approvals, and cross-functional dependencies. The increased capability doesn’t dissolve those bottlenecks. In practice it makes them feel even more absurd. The bureaucracy tax grows larger relative to what’s now possible.
Which creates a three-way rebalance happening simultaneously: the capability gap that protected large organizations shrinks, the natural advantages of small teams expand, and the structural disadvantages of large organizations become more punishing than they were before. If you were designing a system to shift competitive power toward smaller groups, it’s hard to imagine a cleaner mechanism.
— — —
This Trend Didn’t Start With AI
None of this came out of nowhere. AI did not invent the idea that smaller units often outperform large monoliths. That pattern is old.
For roughly the last century, large undertakings have been progressively broken into more modular components, supply chains, manufacturing systems, software architecture, product organizations. The reason isn’t mysterious. Smaller teams handling smaller pieces of a problem move faster and adapt more easily. Work has been decomposing into more modular units for a long time because modular systems are easier to recombine and improve.
AI accelerates that trend. It doesn’t make modularity possible, it makes each module dramatically more capable. Which matters because it means the rise of high-capability small teams isn’t some weird AI exception. It fits a much deeper historical pattern in how systems reorganize when capability gets redistributed. The current shift is large and fast, but it’s not the first time the underlying physics of coordination have changed.
— — —
But There’s a Catch
The same force that amplifies the power of small teams also amplifies their weaknesses.
Small teams have always been fragile systems. They lack the structural ballast of large organizations. They have fewer perspectives in the room. Blind spots can persist longer because there are fewer people to challenge them. Large organizations accidentally solved some of these problems simply by having lots of people with different cognitive styles and roles, the diversity of a large system creates a kind of error-correction that small teams don’t get for free.
When AI dramatically increases the speed and capability of a small group, it also increases the risk of drift, runaway exploration, unexamined assumptions, and decisions made faster than they can be stabilized. Which means the future that begins to appear here isn’t simply one where small teams dominate. It’s one where small teams become extraordinarily powerful — but only if they can stabilize themselves.
— — —
The Loop Closes in an Unexpected Place
This is where the job debate comes back in, from a direction most people aren’t expecting.
The people most worried about AI today are often the ones whose work has been built around process. Structured execution. Reliable procedures. Systems that reward precision and predictability. Those cognitive styles were exactly what large process-driven organizations needed, and as we established in the previous piece, a century of organizational design actively selected and shaped for them.
During the transition that’s coming, many of those people will feel like the ground has shifted under them. Some jobs will disappear. Some roles will change dramatically. Some institutions will restructure in ways that are painful for the people inside them. There’s no reason to sugarcoat that.
But the conclusion many people are drawing, that AI is simply eliminating large portions of human work, may still be the wrong diagnosis. Because if small teams become the dominant organizational unit in an AI-amplified world, those teams face a problem that the technology itself cannot solve.
They need stability.
Exploratory hybrid cognition, the kind of thinking that interacts naturally with AI, is extremely powerful. It produces rapid iteration, unexpected connections, and bursts of creative output. But left alone, it also produces chaos. Systems that only explore eventually lose coherence. Which means the cognitive styles that once looked slow, procedural, or overly structured inside AI-accelerated environments may turn out to be something else entirely.
They may be the stabilizing infrastructure those environments require to function.
The people who understand process. Who build systems that hold together under pressure. Who turn exploration into something repeatable and reliable. In other words, the people many assume AI will replace may actually become essential inside the kinds of teams the technology itself favors. Not despite their orientation toward process. Because of it.
— — —
The Bigger Picture
None of this means the transition will be easy. Periods when technological change redistributes capability tend to be chaotic. Institutions built for one environment struggle to adapt to another. Careers built around old assumptions suddenly feel uncertain. Entire industries reorganize around new constraints.
The printing press did this. The industrial revolution did this. The internet did this. Each time, the shift looked frightening while it was happening, and in many cases it was. Disruption at the level of organizational structure is not an abstraction for the people caught inside it.
But each of those transitions also had something in common: the conversation at the time was focused on the visible symptoms while the deeper structural shift was still assembling itself underneath. People argued about which trades would survive the factories. Which newspapers would survive the internet. The right question, what kind of institutions does this new environment actually favor, only became legible after the fact.
We’re in that moment again. Everyone is asking which jobs AI will replace. But if the technology is actually destabilizing the organizational structures those jobs were designed to fit, then the question itself is too small.
The real story may not be about jobs disappearing. It may be about the systems that defined work for the past century starting to lose the structural advantages that once held them together.
And when that happens, history suggests the consequences rarely stay contained to the workplace.
— — —
Sources
1. AI productivity gains and expertise gap compression
Eriksson, K., Hasan, I., and Jia, C. (2023). “Generative AI at Work.” National Bureau of Economic Research Working Paper. Study of 5,179 customer support agents found generative AI increased productivity 14% overall, and 34% for novice and low-skilled workers, largely by helping them access patterns used by stronger performers.
https://www.nber.org/papers/w31161
2. AI company revenue velocity (Stripe cohort data)
Stripe. (2025). “Indexing the AI Economy.” The top 100 AI companies on Stripe reached $1 million in annualized revenue in a median of 11.5 months — approximately four months faster than the fastest-growing SaaS companies at the height of the subscription boom. Younger AI cohorts are reaching major milestones roughly three times faster than pre-2020 cohorts.
https://stripe.com/blog/inside-the-growth-of-the-top-ai-companies
3. Lovable: $100M ARR, 45 employees
Shead, S. (2025, July 23). “Eight months in, Swedish unicorn Lovable crosses the $100M ARR milestone.” TechCrunch. Lovable reached $100M ARR in eight months from first $1M, with 45 full-time employees and zero paid marketing spend.
https://techcrunch.com/2025/07/23/lovable-100m-arr/
4. ElevenLabs: ~$100M ARR range, small team
TechCrunch reporting on ElevenLabs ARR milestone, reached with approximately 50 employees. Exact figures vary by source and reporting period; figure cited reflects range reported across multiple outlets in 2024–2025.
5. Cursor: fastest-growing SaaS to $100M ARR, later $500M ARR
Sacra. (2025, February 4). “Cursor at $100M ARR.” Sacra estimates Cursor reached $100M ARR in approximately 12 months — the fastest SaaS company on record from $1M to $100M, ahead of Wiz (18 months), Deel (20 months), and Ramp (24 months).
https://sacra.com/research/cursor-100m-arr/
Cursor. (2025, June 5). “Series C and Scale.” Cursor confirmed crossing $500M ARR, used by over half of the Fortune 500.
https://cursor.com/blog/series-c
6. Coordination overhead in large organizations
Cross, R., Rebele, R., and Grant, A. (2016, January–February). “Collaborative Overload.” Harvard Business Review. Research across two decades found time spent by managers and employees in collaborative activities has ballooned by 50% or more, with many people spending around 80% of their working time in meetings or responding to colleagues.
https://hbr.org/2016/01/collaborative-overload
— — —