Blogmark

Justin describes some of ways that LLM coding agents used as pair programmers can lead to bad patterns in pair programming. In particular, one person taking over while the other person disengages or becomes a rubber stamp for their changes.

He then throws out some ideas for ways that LLM coding assistants can be improved with what we’ve learned over the years from person-to-person pairing.

My preference is what he describes here:

throttle down from the semi-autonomous "Agent" mode to the turn-based "Edit" or "Ask" modes. You'll go slower, and that's the point.

One suggestion is something we can do now (out-of-band or at least front-loaded).

Design agents to act with less self-confidence and more self-doubt. They should frequently stop to converse: validate why we're building this, solicit advice on the best approach, and express concern when we're going in the wrong direction

Tell the LLM you want to approach things this way, that you want to break the problem down and tease out any issues with the proposed approach. Engage in some back and forth, have it generate a plan, and then go from there.