The trust problem in code review agents

This is a placeholder post. Replace the body with the real write-up; the structure below is here to show the format.

The brief

Describe the agent, who it is for, and what “world-class” means for this job. Be concrete: the inputs it receives, the artifacts it produces, and the decision someone makes from them.

The first version

Describe the first working version and the first thing it got wrong. Keep it short — one example is worth more than a feature list.

Defining quality

Describe how you turned “good” into cases and graders. What does a correct answer look like? What separates a plausible answer from a useful one?

Climbing

Describe the loop: which traces you read, what you changed, and what the eval said. Include one change that won and one that did not.

What’s next

Say what you’re working on and how readers can get involved. If you’d like to talk, send us a note.