Write the commit description yourself, even when the agent wrote the code

A developer writing at yedhu.me makes a small, practical case in this post: even when an agent writes your code, you should still write the commit description yourself. The reasoning is that a good commit body was never about recording what changed, which the diff already shows, but about explaining why, and forcing you to check that you actually understand the change. Before agents existed, the author spent five to ten minutes on these, pulling scattered context into one place and reflecting on the decision.

Agents break this quietly. They can generate a plausible commit message, but they miss the context spread across chats, tickets, and offhand conversations, so the reasoning they produce can be invented and sometimes contradicts the real driver behind the change. Writing it yourself is a cheap test: if you cannot explain the change, you did not understand it well enough to ship it. The post points to one kind of note that agents almost never capture, the temporary decision with an exit condition, written like "I am doing this until we...", which records why a workaround exists and when it should go away. Those lines rarely live anywhere else, and they are exactly what a future maintainer needs.

Why it matters

If you work with coding agents, keeping the commit description in your own hands is a fast way to catch changes that look right but miss the intent, and to leave behind the reasoning the agent had no way to know.