Every quarter, a new model writes marginally better code on benchmarks. And every quarter, enterprise teams are stuck on the same problems. The agent does not know where the types live. It cannot run the test suite. It ignores the conventions in the README. Nobody knows which agent told a customer the next event was in London.
The teams getting actual value are not the ones with the best model. They invested in context engineering. They built the agents.md files. They configured the LSP. They set up the VM environments so the agent can run the app. They created the feedback loops so every failed PR makes the next one better. They picked workflows narrow enough that learning compounds. They gave each user their own instance.
The hard part is not the AI. It is the engineering around the AI — the infrastructure, the context, the controllability, the integration with the systems your team already uses. Agents are software. Like all software, they are only as good as the environment they run in.
I've argued elsewhere that the harness is the product and controllability is not optional. Both of those are downstream of one truth: the codebase is the territory and the agent needs a map. Nobody is going to draw it for you. The frontier model providers will not. The platform vendors will not, except inside their own walled gardens. Your team is the only group with the institutional knowledge to put structural, navigational, and operational context into a form the agent can use.
The codebase is the territory. The agent needs a map. Your job is to draw it. Start this week.
Sources
Related Essays
Context Engineering Is the Hard Problem
Models keep getting better, but agents without deep codebase and organizational context are just expensive autocomplete. Context engineering is the bottleneck nobody has productized.
Tech Context Is Tractable. Org Context Is Not
The hardest unsolved problem in agent infrastructure is not compute or sandboxing. It is context — and most of that context lives in people, not repos.
Three Layers of Agent Context, and Most Agents Have Zero
The coding-agent debate is dominated by model capability. Wrong bottleneck. Context splits into structural, navigational, and operational layers — and most agents are missing all three.
Key takeaways
- Teams getting actual value from agents did not pick the best model. They invested in context engineering.
- They built agents.md files, configured the LSP, set up VMs that run the app, created feedback loops, narrowed workflows, and gave each user their own instance.
- The hard part is not the AI. It is the engineering around the AI — infrastructure, context, controllability, integration.
FAQ
What separates teams getting value from teams that have stalled?
Investment in context engineering. The teams getting value built the maps — agents.md files, LSP configuration, running app environments, feedback loops, narrow workflows, per-user instances. The stalled teams kept upgrading the model.
Is the model getting better solving the problem?
Marginally. Benchmarks improve every quarter, but enterprise teams stall on the same context problems regardless of model version. Agents are software, and software is only as good as the environment it runs in.