HarnessMM from Tsinghua and Microsoft cuts agent tokens 87.93% while boosting success

A Tsinghua–Microsoft paper introduces HarnessMM, which combines guarded decision reuse and recursive self-improvement to cut agent harness token usage by 87.93% while actually increasing task success rates.

2026-11-02 ~ 2026-11-02 · 3 related posts

• Tsinghua & Microsoft's HarnessMM cuts agent token use by up to 87.93% while raising success rate — rohanpaul_ai · 2026-11-02

2 near-duplicate retellings: rohanpaul_ai · rohanpaul_ai