A prompt that tests whether your AI workflow compounds or degrades at scale
Powerful_Creme2224 · reddit · 2026-07-23
The post proposes a prompt for asking an AI to judge whether a real work operation will compound useful structure or compound recovery and coordination costs as projects, agents, handoffs, collaborators, and model changes increase.
It defines three possible outcomes:
- 1.01 — useful structure is likely to compound
- 0.99 — recovery/supervision/coordination costs are likely to compound
- UNKNOWN — the available history is insufficient
The prompt instructs the model to use only accessible work history, avoid invented assumptions, and return a diagnosis based on:
- What is already working
- What strengths should survive scale
- What re-explanation, verification, ownership, and recovery costs will repeat
- The actual history behind each judgment
- Conditions that cannot yet be evaluated
- The minimum checks before expanding
The author also specifies how the result should change the next step: preserve what works for 1.01, make only one reversible change for 0.99, and record more history for UNKNOWN.
More from AGI Musings
- Could an AI ever win a Nobel Prize? — TobyWalsh · 2026-07-23
- AI Compute Wars: Sam Altman Leads as Elon Musk Saved by Cursor Acquisition — iruletheworldmo · 2026-07-23
- Robert Wright and Connor Leahy debate the ideology and culture around AI — WonderFactory · 2026-07-23
- VC Calls Academic Science Fraudulent, Stanford Prof Fires Back on Industry Bias — anshulkundaje · 2026-07-23
- A user says better AI tools are making work feel fun again — carsonfarmer · 2026-07-23
- Joshua Saxe says AI cyber risk needs safety rules that evolve with capability — kuza55 · 2026-07-23