HarnessMM cuts agent tokens per task by up to 88% while raising success rates
omarsar0 · x · 2026-10-26
A Tsinghua/Microsoft Research Asia paper proposes HarnessMM, which stores settled decisions in an Executable Decision Graph and reuses them when guard conditions hold, avoiding redundant model calls after tool feedback. A recursive self-improvement loop tunes the harness config without touching model weights. On a DeepSeek harness, tokens per task fall from 96.28k to 11.62k (87.93% savings) while success rises from 71.43% to 85.71%; on Codex/HumanEval tokens drop 26.52% with a 29.82% saving reported. Ablations show reuse is 26.91% cheaper than single calls plus callbacks.
Related event: HarnessMM Cuts Agent Harness Token Use by 88%(2 posts)→
More from coding & agent
- Dev mixes Claude specs with abliterated GLM to build SS7 threat detector — l4rz · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- GitHub Next's Chopin prototype explores real-time multiplayer agentic planning — max__drake · 2026-10-27
- TMLR paper: enforced thinking doesn't reliably make LLM agents disclose key info to users — SharonYixuanLi · 2026-10-27
- LLM-written Rust library bbtree verified: speed claims hold, memory savings reverse — lmoroney · 2026-10-27
- Resin lets an agent build standalone LLM inference engines; vLLM and SGLang optimize DeepSeek kernels differently — AccBalanced · 2026-10-27