01
We gave three coding agents the same broken production task
Same repository, same acceptance criteria, same time box. The difference was not where we expected it.
STATUSPUBLISHEDOPEN LOG →
DevinBiAI PUBLICATION
MENUSEARCH
EXPERIMENTS / TEST LOG
Experiments
Real tools, real workflows and measured outcomes from the edge of practical AI.
01 / TEST LOG
Published experiments
3 RUNS02
Can a browser agent finish a real procurement flow without intervention?
Browser Use · 24 attempts · 79% completion
STATUSPUBLISHEDOPEN LOG →
03
Testing local models for structured tool calling under latency pressure
Same tool schema, same evaluation set, different model sizes and serving setups. We measured reliability before speed.
STATUSPUBLISHEDOPEN LOG →