HomeAgent failure modes
Agent failure modes, measured
The failure modes people hit with AI coding agents, one page each. Every page carries a measurement we ran ourselves rather than a recap of someone else's blog post.
what a page here has to clear
- A number we measured on a real repository, not an estimate and not a quote from another article.
- The method printed on the page, so you can run it against your own sessions and check us.
- Published as measured — including the page where the data contradicts our own marketing.
Why there aren't more pages here yet
Each of these takes a measurement to write, and we publish them as the measurements land rather than filling the section first and finding the numbers later. Context loss, hallucinated endpoints and silent regressions are next.