The
Dead Moose.
There is a theory in ecology called the mesopredator release. When an apex predator disappears from an ecosystem, the mid-tier predators beneath it don't simply fill the void — they proliferate, often chaotically, reshaping the environment in ways nobody predicted. The absence of one dominant force doesn't produce a vacuum. It produces a new order.
Something like this is happening in software.
01The talent moat
For thirty years, the dominant competitive moat in technology was talent. The companies that won — the ones that became platforms, then infrastructure, then institutions — won largely because they could aggregate the best engineers in the world. They built campuses and compensation packages and internal tools and cultures designed to attract and retain people who could have worked anywhere. The product was almost secondary. The real asset was the density of exceptional people working on hard problems in the same building.
This was a genuinely defensible advantage. Good engineers are rare. Great engineers cluster. A team of ten exceptional developers will consistently outperform a team of a hundred average ones, and the best companies understood this and acted on it accordingly. Talent density was the moat.
That moat is eroding — not disappearing, but changing shape in ways that haven’t fully registered yet.
02Leverage changes the math
The change is not that AI is replacing engineers. That framing misses what’s actually happening. The change is that the leverage of a single exceptional developer has increased dramatically. A developer who knows how to work with agents effectively — who can decompose problems cleanly, maintain structured context, direct parallel workstreams — is not doing the work of one developer faster. They are doing the work of a small team. Not in every dimension, but in the dimensions that compound: architecture, product decisions, implementation velocity.
This changes the talent equation in both directions.
For incumbents, the math gets complicated. A 500-person engineering organization built for the previous era of development — with its ceremonies and handoffs and coordination overhead — does not automatically become a 500-person organization amplified by AI. It becomes an organization where the overhead is the same and some fraction of the individual contributors are dramatically more productive and some fraction are not. The productivity distribution widens. The coordination costs don’t shrink on their own.
For smaller teams, the opportunity is unlike anything that has existed before. The historical disadvantage of a small team was bandwidth — there were simply things you couldn’t build because you didn’t have enough people. That constraint is loosening faster than most people have absorbed. A team that can deploy agents effectively can now build and maintain systems that would have required ten times the headcount five years ago.
03Incumbents are not standing still
The incumbents are not standing still. The same large technology companies that built their moats on talent aggregation are now the ones training and deploying the models that are changing the leverage equation. This is the part of the story that gets underappreciated in the “AI disrupts big tech” narrative. The model providers are not neutral infrastructure. They are participants with their own interests, their own distribution advantages, and war chests that dwarf anything a startup can deploy.
The moat didn’t disappear. It shifted terrain.
Talent density as a moat is weakening. Model access, distribution, and data as moats are strengthening — and the companies that have spent decades accumulating those advantages are not oblivious to what’s happening. The question isn’t whether they’ll adapt. It’s whether their size makes certain kinds of adaptation structurally difficult in ways that create genuine openings for smaller, faster-moving teams.
History suggests the answer is sometimes yes. Not because large companies are incompetent, but because the software that serves a new paradigm well is often architecturally different from the software that served the previous one. And rewriting from a position of success, with existing customers and existing revenue and existing expectations, is genuinely hard regardless of how much capital you have.
04New paths up the mountain
The developers who understand this moment clearly are not the ones asking whether AI will take their jobs. They’re the ones asking how to deploy it most effectively — how to build systems and workflows that make their judgment count for more, that let a small team punch far above its nominal weight.
The apex predator hasn’t left the ecosystem. But the terrain has shifted enough that there are new paths up the mountain that didn’t exist before.
The teams that find them first will build things that wouldn’t have been possible any other way.