Loop

Trust loop

You might say: earned autonomy, opening the throttle

What human review catches becomes new checks, sharper instructions and stronger verification. Fewer bad pull requests reach a person and more merge on evidence, so the team widens the merge policy to the next risk class and the human review queue shrinks. Growth levels off at the work that stays high-risk, where a person still reads the diff.

What it does
Explains how a factory moves from human review on every pull request to level 4, where evidence carries low-risk work and people review only what needs them.
Ask
Which evidence earned the last widening of the merge policy, and what signal would narrow it again?
Trouble
The merge policy widens on hope instead of evidence, and defects reach production that a person would have caught.

Loop: s-curve

Human review → Automated checks → Merge rate → Merge policy → Review queue ↺

Sources

  1. Donella Meadows, Leverage Points: Places to Intervene in a System, 8. The strength of negative feedback loopsNegative feedback loops are ubiquitous in systems. Nature evolves them and humans invent them as controls to keep important system states within safe bounds.
  2. Addy Osmani, Agentic Code Review, Should we just let AI review more of it?How much human you keep is a dial, and you set it by blast radius, not by guilt.
  3. GitHub, Best practices for using GitHub Copilot to work on tasks, Adding custom instructions to your repositoryIf Copilot is able to build, test and validate its changes in its own development environment, it is more likely to produce good pull requests which can be merged quickly.
  4. Anthropic, Building effective AI agents, Combining and customizing these patternsthe key to success, as with any LLM features, is measuring performance and iterating on implementations.
  5. Lauren Tan, pstack, pstackwhen you can go deep on one agent and trust it to write good, verifiable code, you can truly parallelize with confidence.
  6. Lauren Tan, pstack, Earn the trust before the loopA loop you don't trust just produces unchecked work faster, and the mess compounds with every iteration.
  7. Matt Pocock, software-factory skill (pull request #1316), Start slowa human reads the PR before it merges; what it catches becomes new automated checks and sharper automated review.
  8. Matt Pocock, software-factory skill (pull request #1316), Start slowOpen the throttle as the brakes earn trust, and cap every loop of runs at a few tries before a human.
  9. Lauren Tan, Noodle, docs/concepts/modes.mdA mature project with good tests can run in auto. A new project starts in supervised. Critical infrastructure stays in manual. Move the dial as trust builds.

Use it on your project

teardown-apply: hold "Trust loop" (trust-loop) from the teardown at https://teardownatlas.com/software-factory/ against my project

Paste it into a coding agent that has the teardown skills.

An AI agent wrote this map with the teardown skill. It can contain errors: check the sources before you rely on a part. About and legal