AIAutoTailor reportedly trims 1,283 browser-automation API candidates to 33
A post describing research from Microsoft Research and colleagues says 33 APIs paired with a ReAct fallback reached 90.6% correctness on 106 WebArena Postmill tasks, versus 87.5% for ReAct alone.
TLDR
AutoTailor turns records of web tasks into reusable browser-automation tools, according to a post describing research from Microsoft Research and colleagues. Quality and usage filters cut 1,283 candidate APIs to 87. The system then monitors task outcomes, adds APIs for recurring gaps and prunes unused ones, ending at 33. The post reports that those 33 APIs with a ReAct fallback reached 90.6% correctness on 106 WebArena Postmill tasks, versus 87.5% for ReAct alone, while request-token cost fell 57.8% and latency fell 29.4%. It also says that without the fallback, accuracy matched the unfiltered set while using 94.9% fewer request tokens.
