Applied Compute lifts Qwen agent tool-use rate via continual learning

In an AI Engineer talk, Applied Compute's Samuel Denton describes raising a Qwen thinking model's submit-tool call rate from 22% to 60% on SWE-bench — test pass rate flat — by conditioning rollout on an old production trace. The fix targeted agents that took up to 80 turns before submitting.

Featured · Samuel Denton