Stop Saying 'Make It Better': Build Scoring Loops for AI Agents

Roger_M_Taylor · x · 2026-08-06

Asking an AI agent to 'make it better' fails to provide actionable signals. The author proposes a structured iteration loop: PRODUCE → CRITIQUE with a score → REFINE, repeating until a 9/10 score is reached or plateaus.

Additionally, it recommends building three output intensity modes into skills (e.g., MINIMAL for quick summaries, STANDARD for structured output), transforming one-off prompts into a systematically optimizable skill set.

Original post →

More from coding & agent

coding & agent channel →