Systems Thinking Lab Newsletter: Build with AI so it sharpens you.

What to do while the agent runs

The weekly letter from Systems Thinking Lab: systems thinking insights for junior engineers, framed through the seven building blocks.


Scott Batson describes a habit a lot of engineers will recognize. His agent started a loop, and his first move was to start another one. AI had made him, by his count, five times as productive, so he took on five times as many things. A few months in, he writes, he had burned himself out ten times faster than before.

His essay, "What Should Programmers Do While the AI Writes Code? Hint: not start another AI coding loop," is one engineer's account, offered as an option, not a rule. What I like is the turn. Watching his agent spin up sub agents and stack pull requests, he asked a different question: "What if... rather than starting another loop of work, I could just... read something?"

So he did, and he reads documentation more thoroughly now, not just hunting for the snippet. He reports that the reading changed how he directs his agents: he leads them down a path he would have paved himself, because he took the time to digest the full scope of a feature.

That is the part worth borrowing: the minutes while an agent runs are yours to spend.

Here is what I would add. Reading goes faster and sticks longer when you know what to read for. "Go read the code" is a vague job, and it is easy to skim for twenty minutes and come away with nothing. A short list of questions makes it concrete.

Pick the feature the agent is touching right now, and name its parts:

You do not need to understand every line. Put a name on each part and say what it hands to the next. If one of those parts has no home in the code path, that is worth noticing too.

Give it ten minutes per loop. One feature, four questions, a few notes in your own words. Do that each time the agent runs, and by the end of the week you can explain a part of the system you did not type. That is a different kind of ownership than approving a diff.

The reading should feed back into how you work, the way it did for Batson. The next time you brief the agent, you can name the service and the queue instead of describing a vibe.

So here is the practice version, for your next wait. Next time the agent is running, do not open a second task. Open the code path it is changing, and name the blocks.

Kay

P.S. What to read for is what Course I teaches: seven building blocks and three external entities, so you can explain how your system works, whoever typed the code. It is all at https://systemthinkinglab.ai/courses-i-iv.html