Which tool to choose
Any AI assistant capable of reasoning will do; the difference is made by how tough you want the confrontation.
- You want direct but constructive criticism: the AI assistant in the role of a skeptical consultant. It lists your weak points and suggests how to strengthen them. Suitable when the idea is still tender and you want to improve it, not demolish it.
- You want the toughest stress test: the same assistant in the role of an investor who has to say no, or of a competitor who wants to beat you. No discounts. Suitable when you have to decide whether to put real money in.
- You want to foresee the failure before it happens: the AI that does a "premortem", that is, imagines the project has already failed and reconstructs why, working backward. A trick that brings out hidden weaknesses better than the direct question "what could go wrong".
How to do it
- Explain the idea to the AI honestly, including the parts you're least sure about. If you hide them, the critic won't be able to help you where it really matters.
- Assign the role and the level of toughness. The operational syntax for the premortem:
Play devil's advocate on this business idea of mine. Imagine a year has gone by and the project has failed miserably. Tell me the story of how it fell apart: the three or four most likely reasons for the failure, starting with the most mundane and underestimated. Then, for each one, tell me what I could have done today to avoid it. Be ruthless but concrete. The idea is: [describe it].
- Read the objections without defending yourself: the instinct is to justify yourself, but the value lies precisely in hearing what you don't want to hear.
- Sort the objections into two lists: the ones that hold up (real weaknesses to address) and the ones that don't apply to your case. You are the final judge, not the AI.
- Follow-through: for every objection that holds up you must come out with an action or a check to carry out. If an objection stops you without giving you something to do, ask the AI "how do I test whether this risk is real before investing".
If the AI stays too kind and keeps encouraging you, reinforce the role: "I don't want reassurance, I want the reasons an expert would say no". Assistants tend to go along with you; you have to say clearly that here their job is the opposite.
A concrete example
You want to open a healthy meal delivery service in your city. You're enthusiastic. You use the premortem prompt. The AI imagines the failure and tells you: "You started with too broad a menu, the kitchen costs ate your margins, and the customers from the first months didn't come back because the delivery arrived cold". For each point it gives you a remedy: start with three dishes, calculate the real cost per delivery before setting the price, run a delivery test with friends and measure the temperature on arrival. Three cracks that enthusiasm was hiding from you, surfaced without having spent a single euro.
When it does NOT work (and how to fix it)
If the AI keeps being kind and encouraging
It's the default behavior: assistants tend to please. Fix: raise the stakes in the role ("you're an investor who has already lost money on similar projects and is wary today") and explicitly ban reassurances. If needed, ask it to give you only the cons, without even a single pro.
If the objections are generic and apply to any business
Vague criticism, not very useful. Fix: give the AI the real numbers and constraints (your city, your budget, your actual competitors) and ask for objections specific to that context. One tailored criticism is worth ten textbook warnings.
If the criticism just discourages you
Having an idea demolished hurts and can be paralyzing. Fix: remember that the devil's advocate's job is to find the cracks, not to deliver the verdict. After collecting the objections, switch hats and ask the AI "now help me strengthen the idea against these criticisms". The goal is a more solid project, not to give up.
A tip from someone who actually uses it
Use the devil's advocate before you've fallen in love with the idea, not after. The more you've worked on it, the more you defend your choices and the less you listen to objections. Doing the premortem at the start, when you don't yet have an ego invested, is what lets you change course when it costs little instead of when you've already spent everything.
Frequently asked questions
Don't I risk being discouraged by an AI that only sees the negative?
You see the negative on request and by choice, not as a verdict: you asked for the cracks in order to reinforce them. After the criticism, always do the reverse round and ask how to overcome them. The devil's advocate serves to make you stronger in the face of real problems, not to convince you to quit.
Can the AI play devil's advocate better than an expert friend?
It has a specific advantage: it's not afraid to hurt you and doesn't spare you out of kindness, something friends find hard to do. But it doesn't know your market like someone who works in it. The best combination is to use the AI to bring out all the possible objections, then take the toughest ones to a person who really knows the field.
If the idea survives the AI's criticism, is it good then?
Having withstood the confrontation is a good sign, but not a guarantee: the AI reasons on what you tell it, it doesn't see the real customers. A good idea on paper still has to pass the market test. The stress test reduces the obvious risks; the final judgment comes when someone pulls out their wallet.