Me-First Prompt Collaboration

Form your own answer before consulting the model, start with oversimplified prompts, and review outputs with distrust heuristics to avoid anchoring.

When humans collaborate with GenAI (or approve agent plans), Me First: produce an independent answer before reading the model’s, to avoid anchoring. Start with an overly simple prompt to establish a doubting mentality rather than accepting a polished first draft. Review heuristics: distrust first impressions; probe understanding with self-validating questions; compare against your independent answer; ignore flattery and projected confidence; ask how you can evaluate the answer; check stability by re-asking differently.

After a productive session, ask the model to summarize questions asked and turn that summary into a reusable prompt — feed the prompt catalog. This is operator discipline for human approval gates and offline critique, not a substitute for offline prompt evaluation oracles. It counters the AI productivity paradox pressure to skip review.

---

Sources

• Taking Testing Seriously (Bach & Bolton), ch. 7