Starting a Business? Let AI Play Devil's Advocate First

You have been turning a business idea over in your head for weeks. You mention it to a friend over coffee, and they say it sounds great. Another friend nods. Nobody asks how you will actually get your first ten paying customers. That silence feels good for about a day, then it starts to itch. What you need is not applause. You need someone to poke holes in the plan while the holes are still cheap to patch.

What the AI actually does when you assign it this role

A conversational AI assistant predicts the next words in a reply based on the instructions you give it. If you ask for feedback, it tends to produce balanced, agreeable feedback, because that is what most people reward. But if you explicitly assign it the role of a harsh critic and tell it to attack your assumptions, the same system shifts. It starts generating objections, counterexamples, and uncomfortable questions instead of reassurance.

The trick is not a secret feature. It is the instruction itself. You are telling the tool what kind of reply counts as a good one. A vague request invites a vague answer. A sharp role with clear rules invites sharp pushback. This is why the same assistant can feel useless one minute and brutally useful the next.

Test one: the polite request

Asking for feedback the usual way

I want to open a small coffee shop in my neighborhood. What do you think of my idea?

That reply sounds encouraging, but it tells you almost nothing you did not already know. It never asks how many coffee shops already sit within walking distance. It never asks what your monthly rent would be or how many cups you would need to sell to cover it. The assistant defaulted to cheerleading because you asked what it thought, not what could go wrong.

Test two: the calibrated devil's advocate

Assigning a strict critic role with rules

Act as a ruthless devil's advocate for my business plan. You are a skeptical investor who has seen many small businesses fail. Do not compliment anything. Ask me one hard question at a time about my coffee shop idea, wait for my answer, then attack the weakest part of what I just said. Start now.

That is a different conversation. The assistant is now working against your idea instead of for it. It asks one question, waits, then presses on the weakest point of your answer. You cannot hide behind a vague plan, because the next question will find the gap. This is the same tool, the same free app, and a completely different result, because the instruction changed what counted as a good reply.

Master template to copy and paste

Fill in the brackets with your own details. Keep the role, the rules, and the one-question-at-a-time rhythm. That rhythm is what stops the assistant from dumping a generic list and forces it to react to what you actually say.

You are a skeptical [INVESTOR / MENTOR / COMPETITOR] who has watched many [TYPE OF BUSINESS] fail. I am planning to launch [YOUR BUSINESS IDEA] in [LOCATION OR MARKET]. Your job is to attack my assumptions, not encourage me. Rules: no compliments, no generic advice, one hard question per reply, wait for my answer, then challenge the weakest point in what I just said. Begin with your first question.

Quick steering commands if the AI goes soft

• That answer was too kind. Drop the encouragement and give me the single strongest reason this idea fails.
• You are drifting into general advice. Stay in character and ask me one uncomfortable question about my numbers.
• Stop summarizing my plan back to me. Attack the part I just said and tell me what a skeptic would not believe.

How you know the training worked

You will know it worked when you feel slightly annoyed. That irritation usually means the assistant found a real gap, not a polite one. If every reply still feels comfortable, your role instructions are too gentle or your answers are too vague. Tighten both. Then take the same exercise into the free voice mode on your phone and answer the questions out loud. Speaking forces you to drop the polished script and find out whether you actually have an answer, or only a hope.