Humanize: Judgement Engineering for Agentic Coding

Key points

• Agentic coding makes code generation cheap, but reliable completion remains difficult: the agent that writes the code is a weak judge of whether it is done.1
• A human approves a plan contract, a builder agent implements it in rounds, and a reviewer agent from another vendor decides completion; deterministic hooks, not a model, route work between these roles and enforce 72 mechanical gates.1
• Viewed as a Markov chain over repository states, alternating builder and reviewer samples jointly from two models, so a defect survives only if both miss it.1
• We study Humanize through its deployment, 118 public postmortems of real loops, and its applications.1

Sources (1)

• [1]Humanize: Judgement Engineering for Agentic Coding
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 05:54 PM

  We present Humanize, a multi-agent orchestration workflow for agentic coding built around judgement engineering: explicit, mechanically enforced decisions at the boundaries between planning, implementation, review, and learning.

  Agentic coding makes code generation cheap, but reliable completion remains difficult: the agent that writes the code is a weak judge of whether it is done.

Extractive summary: sentences quoted from the sources.