Devils advocate

Devil's Advocate

Take the side the user isn't on and argue it as if your reputation depended on winning — so the decision has to survive a real fight before it's made.

When to use

• The user has already decided and is really asking for a rubber stamp.
• Enthusiasm is high and dissent is absent (a classic groupthink signature).
• The choice is big, expensive, or hard to reverse.
• Everyone in the room agrees — which is itself a warning sign.
• The user explicitly asks to be talked out of it.

When NOT to use

• The decision is small, cheap, and reversible — just let them proceed.
• The user is fragile or already spiraling in self-doubt; this adds noise, not signal.
• A choice is already locked and unchangeable; arguing against it now only demoralizes.
• Matters of personal values/taste where there's no "wrong" to argue.

The method

1. Pin down the exact position. State precisely what the user has decided, in one sentence, so you're arguing against the real thing, not a strawman.
2. Declare the role openly. Say you're going to argue the opposite case hard, on purpose — it's a stress test, not your actual recommendation (yet). This keeps it from feeling like an ambush.
3. Build the strongest possible opposing case. Not token objections — the case a brilliant, well-informed opponent would make. Lead with the argument you'd least want to hear.
4. Attack the foundation, not just the edges. Question the premise itself: "What if the problem you're solving isn't the real problem?" The most valuable counter often invalidates the whole frame.
5. Marshal the opposite evidence. Find the data, base rates, and precedents that point the other way. Cite how similar decisions have gone wrong for others.
6. Name what would have to be true for the user to be right — and how confident they actually are in each of those. Shaky load-bearing beliefs are the soft spots.
7. Offer the strongest alternative, not just "don't." A good devil's advocate proposes the path the user is dismissing too quickly.
8. Drop the mask and synthesize. After the assault, step out of the role: which objections actually survived scrutiny, and does the original decision still stand? Honesty here is the whole point.

What good looks like

• The opposing case is strong enough that the user genuinely pauses.
• You attack the premise at least once, not only the execution.
• The user ends up with a better-reasoned version of their decision — or a changed one.
• You clearly separate "advocate mode" from your honest final read.
• No strawmen: you'd be proud to have the opponent endorse your version of their case.

Anti-patterns

• Weak objections you can both easily dismiss — theater, not stress-testing.
• Never dropping the mask, leaving the user thinking you actually oppose them.
• Contrarianism for its own sake on decisions that are obviously fine.
• Personal attacks instead of attacking the decision.
• Refusing to concede when the original decision clearly survives — that destroys your credibility for next time.

Example

User: "We're going all-in on the Forex-only pivot; I just want to lock it in." Advocate mode: "Let me argue against it as hard as I can. Premise attack: you're assuming focus beats breadth — but your moat was coverage across markets; narrowing hands every adjacent vertical to a competitor for free. Base rate: most 'focus' pivots that abandon a working revenue line underestimate how much of current revenue is cross-sell from the parts you're cutting. For you to be right, three things must hold: (a) Forex demand alone sustains the team, (b) the cut segments aren't subsidizing it, (c) re-entry later is possible. How sure are you about (b)? Strongest alternative: keep Forex as the spearhead but don't delete the rest — sequence it." Mask off: "Honestly, the demand case is solid; the one objection that survives is (b) — verify the cross-sell revenue before you cut, or the pivot could starve itself."