Argue from N perspectives

Last tested 2026-04-26

The prompt

I have a decision to make: {{decision}}.

Argue the case from each of these perspectives, one at a time:
{{perspectives}}

Then produce a 'conflict map' showing where the perspectives actually disagree, and a synthesized recommendation that takes the disagreement seriously - not one that smooths it over.

Why it works

Decisions made by a single mind have a single set of blind spots. This prompt simulates a panel: it forces the model to argue the case from N specified perspectives, then synthesize the disagreement. The two outputs that matter are the conflict map (where the perspectives actually disagree) and the recommendation under each lens. You'll notice the recommendations diverge - that's the point. A choice that looks obvious from one perspective often looks insane from another. Use this for hiring decisions, architecture choices, pricing, positioning, anything irreversible. Do not use it for things you can A/B test cheaply.

Example output

The decision: charge $59/yr or $99/yr for the annual plan.

Founder lens: $99 - we under-priced the last product, do not repeat that.

CFO lens: $99 - LTV math only works above $80.

User-research lens: $59 - early users report $99 feels speculative for an unproven app.

Conflict map: founders/CFO want signal-of-quality and unit economics; user-research wants conversion in cold-start phase.

Synthesis: ship $59 for the first 1,000 customers as a 'founders rate,' then move to $99. Captures both.

Common mistakes

Don't pick perspectives that all share the same priors ('founder, investor, advisor' are functionally one perspective). Pick lenses that genuinely conflict: founder vs. user, engineer vs. salesperson, optimist vs. risk-officer. Also: do not let the model collapse the disagreement too quickly. If the synthesis looks suspiciously clean, ask 'where did the [losing perspective] still have a point?' to recover the texture.