How to Build a Team Architecture Around AI Agents (Without Burning Out on Review)

Lately I’ve been struggling seriously to stay focused during pair and mob programming. Sometimes we hand a single task to a coding agent, and while it’s working, we drift into other topics. Sometimes we hand off multiple tasks at once, one moves forward while we try to review another, but this time too much dead time creeps in again, and we can’t actually focus on more than 3 or 4 tasks at once, so we start missing bugs or mistakes.

The thing I’ve noticed most is this: the topics we talk about in these gaps aren’t as fun as they used to be. When we’re curious about a technical or soft topic, we type it into a chatbot, get an answer instantly, flash modes already respond in a blink. What’s left to talk about is usually “will AI take our jobs.” So most of our conversations now revolve around uncertainty. We can’t even overengineer anymore.

Because of this, people are starting to drift away from pairing, myself included. But that’s a trap either way: if we drift away from pairing, knowledge sharing dies and team bonds weaken; if we keep pairing, burnout starts instead. So whichever direction we go, we end up somewhere problematic.

It's Not Just Us

Once I started researching, I realized this isn’t a focus problem unique to us, or to me. LinearB’s analysis of 4,800 teams and 8.1 million pull requests shows that AI-generated code waits 4.6 times longer for review than human-written code. So our feelings have basically turned into data.