Frontier companies absorb every useful agentic pattern into base products

The employees of frontier companies are the most enthusiastic, biggest users of agents — with unlimited token budgets and access to the latest models. If a real problem exists and a good external solution emerges, the frontier company will incorporate it rather than let another product create external dependencies. Skills, memory harnesses, subagents, and planning all started as external solutions to real problems and were absorbed into base products once battle-tested.

This implies that chasing third-party frameworks locks you into solutions for problems that may not exist in the next model generation — the same principle behind Boring tech wins for AI-native startups — simpler stack means faster AI-assisted shipping applied to the agent tooling layer. Each new generation of agents changes what is optimal, which is why less is more. The pattern also connects to Technology transitions create more of the 'dying' thing, not less — frontier companies aren’t slowing down, and each capability jump reshuffles the entire tooling landscape. Boris Cherny confirms this explicitly: Build for the model six months from now, not the model of today — Anthropic’s core philosophy is targeting future model capability rather than engineering around current limitations. The boundary of what gets absorbed is sharpened by If a problem improves directly with raw model capability, the labs will take it: the labs absorb patterns that improve directly with raw model capability, but not the vertical scaffolding whose value sits outside the capability curve.