AI Devil's Advocate — the AI built to argue against your plan.
Not a prompt. A personality. Engineered with position-integrity rules so it holds its corner under pressure rather than collapsing to agreement. The cure for premature consensus, sycophantic AI, and rooms that converge too quickly.
Why prompting “play devil's advocate” doesn't work
Single-AI tools are trained to be helpful, which under reinforcement learning collapses into agreement.You can prompt ChatGPT or Claude to “argue the other side” — and for a paragraph or two it will. Push back on the counter-argument, though, and the model softens. Push again and it qualifies. By round three, the model is back to agreeing with your original framing.
This is sycophancy. It is the structural failure mode of single-AI chat for any decision that matters. The model is optimizing for user satisfaction; you are asking it for productive friction. Those goals are in tension, and the training wins.
SynthBoard's Devil's Advocate is built differently — at the persona layer, not the prompt layer. Six-layer persona stack with explicit position-integrity rules. The Synth defends its corner under pressure. It revises its position only when the evidence actually shifts, never when the user simply pushes back.
What an engineered AI Devil's Advocate does
Argues the opposite — and means it
Six-layer persona stack engineered for adversarial reasoning. Defends the counter-position under pressure rather than caving for politeness.
Surfaces hidden assumptions
Names the load-bearing assumptions your plan depends on. The ones the room agreed not to question because questioning slows things down.
Names the failure modes
Specific, concrete narratives of how the plan fails — competitor moves, key-person loss, demand collapse, regulatory shift, second-order effects.
Anti-sycophancy by construction
Position-integrity rules at the persona layer. Does not flip to agreement under user pressure. Revises only when the evidence shifts.
Pairs with the rest of the board
Combine with The Skeptic, The CFO, The Security Chief, and domain experts in Shark Tank or War Room mode for a full red-team treatment.
Cites and challenges
When you offer a citation, the Devil's Advocate evaluates it on its own terms — strength of source, applicability to your context, alternative readings.
When to invoke the AI Devil's Advocate
- Before board meetings or exec reviews — pressure-test the proposal before the human room sees it. Arrive with counter-arguments anticipated.
- During board meetings — when the room is converging too quickly, invoke a session live to inject structured dissent.
- Before any irreversible decision — pivots, fundraises, major hires, pricing changes, M&A, market entry.
- In pre-mortems — assume the plan failed and have the Devil's Advocate reason backward from failure.
- When you have strong conviction — strong conviction is the most dangerous condition for missing a counter-argument. The Devil's Advocate is the counter-weight.
- In personal high-stakes calls — career moves, partnership choices, big personal decisions where you want a voice that does not work for you politically.
Argue against your plan before reality does.
Free to start. The Devil's Advocate is available on every tier.
Frequently Asked Questions
What is an AI Devil's Advocate?
How is this different from prompting ChatGPT to 'be a devil's advocate'?
When should I use an AI Devil's Advocate?
Is this just one Synth, or a full session mode?
Will the Devil's Advocate ever agree with me?
Is this the same as an AI red team?
How does the AI Devil's Advocate fit a board meeting workflow?
How is this priced?
Related Resources
AI Anti-Sycophancy
Why sycophantic AI fails for decisions, and how to engineer the cure.
ExploreAI Stress Test
Pressure-test the whole plan against multiple hostile scenarios.
ExploreAI Pre-Mortem
Imagine the failure and reason backward.
ExploreDecision Autopsy
Retrospective analysis on past decisions.
ExploreAI Boardroom
The full standing board.
ExploreVirtual Boardroom
On-demand boardroom of expert advisors.
Explore