What is a Development Automation?

This spans a spectrum from narrow automation (linting, codegen from a schema, auto-formatting) to broad agentic automation, where an AI agent reads a ticket, writes the implementation, runs the test suite, and opens a pull request largely unsupervised. The narrow end has existed for decades; the broad end is new and depends heavily on the agent having real tool access — a shell, a test runner, a diff viewer — not just a chat window. What separates useful development automation from risky development automation is verification: an agent that writes code and claims success without running the tests is not actually automating development, it's guessing. Reliable systems gate any agent-authored change behind the same CI and review bar as human-authored code, and log what the agent actually changed for auditability. The more autonomous the automation, the more the review and rollback tooling around it matters — autonomy without a safety net is just faster mistakes.