Did Codex Reset
GitHub

AutoTailor filters auto-generated web APIs from 1,283 candidates to 33

DAIR.AI

DAIR.AI highlighted AutoTailor, a paper from Microsoft Research and colleagues, as relevant to auto-generating MCP tools from agent trajectories. AutoTailor turns web trajectories into parameterized browser-automation APIs, then applies a quality filter that drops APIs with the wrong granularity or duplicate function and a usage filter that keeps APIs likely to be needed, taking 1,283 candidates down to 87. Online, it watches task outcomes, adds APIs for recurring gaps, and prunes unused ones, ending at 33.

On 106 WebArena Postmill tasks, the 33 APIs with a ReAct fallback reached 90.6% correctness against 87.5% for ReAct alone. Request-token cost dropped 57.8% and latency 29.4%. Without the fallback, accuracy matched the unfiltered set while using 94.9% fewer request tokens. The paper is at academy.dair.ai.

#autotailor#microsoft-research#web-agents#webarena

Only 35.8% of the cleanest public terminal-agent RL collection passed a Salesforce AI Research audit

Salesforce AI Research put that collection at 35.8% clean and two others at 10.1% and 3.3%, after finding reward errors in both directions. The authors say RIVER, which filters defective environments, increases RL gains by 106% on Terminal-Bench-Lite and 30% on Terminal-Bench v2.1 while using fewer than 30% of TMax's environments.