Two Types of AI Loops: Deterministic vs. LLM-as-Judge
See more in Agents & MCP →Exact instruction
- When setting up an AI loop:
- First ask: can the goal be measured with a concrete value? (load time, error count, coverage %) → Use a deterministic loop with a numeric target Claude checks itself.
- If no clean metric exists (code quality, doc accuracy, prose clarity) → Use an LLM-as-judge loop, but write a detailed rubric for Claude to test its work against.
- Default rule: try to reduce the goal to something deterministic first. Only hand judgment to the AI when it genuinely can't be measured.
Suggested prompt
I want to set up a loop for [task]. Help me decide: is this a deterministic loop (measurable target) or an LLM-as-judge loop (needs a rubric)? If deterministic, define the target. If LLM-as-judge, write the rubric.
Adopt?
Yes: This deterministic vs. LLM-as-judge framework applies directly to this workflow and any Claude Code project. General Claude Code update — use this mental model when designing any looping or self-checking agent task.
Find the resource
Creator @buildwitholi offers a full guide when you comment "rubric". Search: "buildwitholi Claude loops rubric deterministic LLM judge" for any published version.
show original captionhide original caption
Comment "rubric" for the full guide on how to set up loops!