AI agents are reshaping what good software engineering looks like. Rather than rewarding clever, idiosyncratic code, agents favor explicit, well-documented, consistently structured codebases. Evidence includes TypeScript overtaking Python/JavaScript on GitHub (up 66% YoY) due to its AI-friendliness, and a contrast between a METR study showing naive AI use slowed developers by 19% versus OpenAI's Codex pilot where a small team merged ~1,500 PRs by redesigning workflows around agents. The shift means repositories are becoming operating manuals for agents, context is infrastructure, and developers must invest heavily in evals, documentation, and specifications. The agent era rewards engineering discipline—explicitness, testability, consistency—over personal cleverness.
Sort: