Where the work goes when an agent writes the code

An agent can produce a patch before I've finished thinking through how I'll review it. I'm finding that the review needs more of my attention than the speed of the first draft suggests.

I start with what changed and what was checked. If there's a test, does it cover the problem in the brief? If the agent says a check couldn't run, I still need to run it or find another way to verify the behaviour. Its summary is a useful starting point for that work.

Reading the diff isn't always enough. I can follow each line and still be unsure why the agent chose that approach. That's when I need to ask questions, try the change or simplify it before handing it to somebody else.

I also have to understand the request and explain the finished change to the next person. The agent helps with parts of that, but reviewing several changes at once still takes time. Quick patches can leave me with a queue of work I haven't properly understood.

For now, I prefer to follow a task through its review before taking on more generated changes. I'm still working out how much I can usefully supervise at once.