AI

Humanize: Judgement Engineering for Agentic Coding

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 02:33:24 UTC.

Briefings

Humanize is a multi-agent workflow for agentic coding that uses judgement engineering to enforce deterministic decision boundaries between planning, implementation, and review phases. Deployed over 108 days, it achieved 672/672 on PutnamBench and ranked first on Lean-Eval, while participating in top three contests of the MLSys 2026 FlashInfer and winning gold medals in international Olympiads. The system uses 72 mechanical gates to route work between roles, ensuring defects survive only if both the builder and reviewer miss them.

Key points

• Humanize implements a multi-agent orchestration workflow with explicit mechanical decision boundaries between planning, implementation, and review phases.
• The system achieved 672/672 on PutnamBench and ranked first on Lean-Eval, competing with professional mathematicians.

Context

Agentic coding has made code generation cheap but unreliable, as agents often fail to judge their own completion. Humanize addresses this by using judgement engineering to enforce deterministic decision boundaries between planning, implementation, and review phases.

Limitations

Evidence is observational, not a controlled comparison of workflows. Deployment data spans 108 days with 118 postmortems, and some performance metrics like IOI and IMO scores are from specific contests rather than general performance.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 26/10/2026, 00:44:48 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeihuqccfhnualmubgudz7x2ssgprmcp7ds6x2gkomo2add4i7jduga