Software engineers keep the factory running. They set up AGENTS.md, knowledge bases, and prompt libraries that let users request changes and ship them instantly. Automated evaluation layers—tests, linting, type systems, and AI‑based evals—watch over the code, keeping agents on track even when the underlying model is simple.

Writing code remains valuable. It forces developers to think directly in the execution environment, rather than through vague English. Hands‑on coding sharpens ownership, helps spot fragile design choices, and lets engineers debug and tighten testing strategies. When a developer refactors a buggy local storage pattern, the resulting architecture is clearer and more reliable.

Agents act like interns that read incomplete, sometimes sloppy code. They lean on conservative changes that preserve the current state. High‑quality code, low‑level control, and clear architecture reduce the agent’s risk surface. CI pipelines, early in side projects, become a standard practice, raising overall software quality.

See how agents can improve search at Cheat at Search with Agents.