Patrick Zandl

questions about that 35% figure: What’s the breakdown and revert rate? If agents handle boilerplate while humans do architecture, “35% of PRs” tells a very different story than the headline suggests. Show the error rate too — wins without failure data is a pitch, not evidence.