Reaction
The limits of counting human touches per pull request to evaluate coding agents
A user says human touches per pull request are easy to track, while scoring a PR is difficult.
TLDR
A user says they use human touches per pull request to evaluate coding agents because the count is easier to track than a PR score. They caution that more touches aren't necessarily bad on a complex task, even when an agent is doing well.
Combined views
6.3K
1 Source, first seen ago