Copilot Autopilot Sounds Useful, but Guardrails Matter
Autonomous coding sessions can be useful, but only when teams are clear about permissions, tests, and what still needs a human decision.
Engineering notes from the sharp end
Practical writing about software, systems, delivery, and things that behave strangely under pressure.
wade womersley – york based software engineer I write the way I work: direct, useful, and more interested in what holds up in production than what sounds clever on a slide.
Latest post
Archive
Recent writing
AI, Software Engineer
The best use of AI in code review is not adding more comments. It is finding the few things that actually matter.
AI, Software Engineer
Codex moving beyond code is more interesting than another model benchmark. The harder problem is where the agent sits in the actual workflow.
AI, Programming
After using GPT-5.5 in Codex, the improvement over GPT-5.4 feels less like a benchmark bump and more like better follow-through on real coding work.
AI, Software Engineer
AI coding agents are moving from novelty demos into normal developer infrastructure. The useful question now is how teams manage them properly.
Programming, Software Engineer
Prototypes are allowed to be clever and disposable. Systems are not. The difference shows up when something grows, someone new has to own it, or you need to debug it under pressure.
Programming, Software Engineer
Defaults are useful until they become hidden policy. I usually prefer explicit configuration because it is easier to understand, easier to change, and much less surprising later.
Programming, Support
Good error messages do more than complain. They tell you what happened, what to try next, and what the system needs from you so failure is easier to recover from.
Programming, Software Engineer
Good logging is not just for engineers. It reduces support time, shortens incident diagnosis, and makes a system much easier to trust when something goes wrong.
Programming, Software Engineer
Smaller deployments are usually easier to trust, easier to roll back, and easier to debug. Once a release starts carrying too much change, the team spends more time managing risk than shipping value.