01
Blog
Notes from building Daniel
Grading AI coding agents with hidden tests: Daniel vs Devin on 6 real bugs
Same six real open-source bugs, same prompt, graded by the maintainers’ own tests. Both agents solved 6 of 6. What the misses taught us.
Why Daniel keeps working when you close the tab
Long engineering tasks shouldn’t depend on a browser tab staying open. How Daniel’s missions run on the server and recover by themselves.