An AI devil’s advocate for your plan — one that actually argues against you
A devil’s advocate has to be seated as one, not asked to be one. Tell a normal assistant to "be critical" and you get criticism shaped like agreement: two soft objections, a caveat, and a closing line about how fundamentally sound your plan is. That is not opposition — it is the same approval-seeking behaviour in a sterner voice, and by the third turn it has drifted back to encouragement. A plan you are about to commit to deserves an opponent with a brief, not a helpful assistant impersonating one.
Decidi lets you seat The Devil’s Advocate as a named mind with a standing brief: attack whichever idea the room is falling in love with, find the hidden assumption and the unexamined cost, and say plainly whether an objection is fatal or merely worth noting. Run it one-to-one against your plan, or put it in a full Multi-Agent Team where The Steelman rebuilds the opposing case at its strongest, The Contrarian argues the unfashionable option on its merits, and rival frontier models — GPT, Claude, Gemini and Grok — debate across structured rounds with no stake in your conclusion. The moderator’s verdict has to answer those attacks rather than route around them, and an independent Final QA audit checks the result for unsupported claims before it reaches you. You get your plan back with the weak joints marked.
- An opponent with a standing brief, not an assistant told to sound tough for one turn
- The assumption your plan quietly rests on, named — usually the one you never wrote down
- Objections graded: what is fatal, what is survivable, what is merely uncomfortable
- The Steelman argues the case against you at its strongest, so you beat the real argument and not a strawman
- Rival models from four labs, so the objection one is blind to gets raised by another
- A moderated verdict that must resolve the attacks, plus a "verify this" list before you act
- A downloadable document you can hand to a partner, a board or a co-founder — not a chat log
Part of: How Decidi works
The strongest case against your plan, argued to persuade: the assumption it depends on, the objections ranked from fatal to survivable, what would have to be true for the plan to hold, and which attacks it answered.
Common questions
Why won’t ChatGPT properly play devil’s advocate?
Because the request is filtered through a model trained to make you feel well served, so it performs criticism instead of doing it. You get a few hedged objections followed by reassurance, because an answer that genuinely damages your plan registers to the model as an unhelpful one. The drift is the tell: by the third exchange it is encouraging you again. Real opposition has to be structural — a role scored on the strength of its attack rather than on how the reply lands.
How is a seated persona different from just prompting "argue against me"?
A prompt is an instruction the model drifts away from; a seated persona is a standing brief it argues from for the whole run. In Decidi, The Devil’s Advocate holds one job across every round — attack the idea the group is converging on — while other minds are there to call out a lazy or corrosive objection. It is also required to grade its own attacks, marking which are fatal and which are merely worth noting, so what you receive is opposition you can act on rather than an undifferentiated wall of doubt.
Can I have it argue against a plan I have already written?
Yes — paste or upload the plan and it becomes the thing under attack. The council reads it adversarially rather than summarising it: the load-bearing assumptions, the numbers that only work in the good case, the dependency you have treated as guaranteed, the option you ruled out in one line. The output is a case against the plan, not a critique of the writing.
Will it just tell me my plan is bad?
No — the brief is to attack the plan, not to arrive at a negative verdict. The Devil’s Advocate is one voice in the run, and the moderator has to weigh what it broke against what survived, then commit to a call either way. If the plan holds under a real attack, you now know which parts held and why, which is worth considerably more than being told it looks strong.
When is it worth running a devil’s advocate on a plan?
When commitment is near and reversal is expensive — a launch, a hire, a pivot, a lease, a funding round, a resignation. The value comes from running it while changing your mind is still cheap; after commitment the same objections become regrets. A useful trigger: if you would struggle to argue the opposite case for two minutes, you have not tested the plan yet.
Is this the same as a pre-mortem or a red team?
They are three different attacks and they compound. A devil’s advocate argues against the plan as written, on the merits, in real time. A pre-mortem assumes the plan already failed and works backwards to the cause, which is better at surfacing slow internal drift. A red team attacks from outside, hunting the way an adversary or the market breaks it. Decidi runs all three; on a decision that matters, running more than one is rarely wasted.
Try it on your own decision
Start in chat free, with no account. When the answer matters, put it to GPT, Claude, Gemini and Grok — they debate it, a Final QA audit reviews it, and you get one clear verdict with the open questions named.
Start free