I have been sitting with a lot of business ideas lately. Some exciting, some probably terrible, some that feel interesting but where I cannot tell if that feeling is insight or delusion.
For years, the way you would pressure test a business idea was to find smart people to talk to, do market research, maybe build a rough prototype. Those things still matter. But I have added a step that has become one of the most useful things in my toolkit: I talk to AI about it first.
Not to get the answer. To find the questions.
Why AI Before People
There is a reason I start with AI before I talk to people. When you bring a half formed idea to a smart friend or potential advisor, you get one shot. They listen, they react, and their first impression sets the frame for every subsequent conversation. If you haven’t sharpened the idea first, you burn that first impression on something fuzzy.
AI does not have a first impression problem. You can run the same idea through it twenty times, refining your framing each time, without any social cost. By the time you bring the idea to real people, you have already identified the weak points, sharpened the pitch, and can engage in a much more productive conversation.
Think of it as a sparring partner you can hit as many times as you want without worrying about the relationship.
The Process, Step by Step
Step 1: Start with the steelman. I describe the idea to Claude or ChatGPT and ask it to make the best possible case for why this is a great business. Not to flatter me, but to articulate the idea more clearly than I might on my own. Sometimes AI surfaces angles I had not considered. Sometimes it just helps me see the idea more sharply than my own internal monologue allows.
This step matters because founders are notoriously bad at articulating their own value proposition clearly. The AI steelman forces you to see the strongest version of your idea, which often reveals whether the idea has genuine substance or just feels good in your head.
Step 2: Then ask it to kill the idea. This is where it gets genuinely useful. I ask for the strongest argument against the business. What are the fatal flaws? What assumptions am I making that might be wrong? What has been tried before and failed? A good AI response here is brutal and specific, which is exactly what you want at this stage.
Most people skip this step when they brainstorm alone because our brains are wired to protect ideas we are excited about. Confirmation bias is real, and it is especially dangerous in entrepreneurship. Asking AI to kill your idea is a structured way to bypass that bias.
Step 3: Identify the riskiest assumptions. Every business idea rests on assumptions, and usually a small number of those assumptions carry almost all the risk. I ask AI to identify the three to five most critical ones: the ones where, if they turn out to be wrong, the whole thing falls apart. This step focuses my validation work on what actually matters instead of letting me chase the parts of the idea that feel fun but are not load bearing.
For example, one idea I was exploring recently rested entirely on the assumption that a specific customer segment would pay a premium for a curated experience over a DIY alternative. AI helped me see that this was the single most important thing to validate, and that everything else was secondary until I had evidence on that point.
Step 4: Map the competitive landscape. I ask for existing players in the space, adjacent solutions, and what they have gotten right or wrong. This used to take hours of research. Now I get a useful overview in minutes, which I can then verify and deepen with my own investigation. The AI overview is not a replacement for real competitive research, but it is a remarkably efficient starting point.
Step 5: Ask “what would make this a billion dollar company, and what would make it a complete failure?” This question forces extreme thinking. The answers are rarely literally useful, but they reveal the underlying dynamics of the idea in a way that moderate analysis often misses. The path to massive success and the path to total failure usually hinge on the same few variables, and this question exposes them.
What Makes This Work (and What Does Not)
The quality of this process depends entirely on the quality of your prompts. Vague inputs produce vague outputs. If you say “I want to build an app for busy people,” you will get generic advice. If you say “I want to build a mobile tool that helps working parents coordinate logistics with their coparent, differentiated by its use of AI to predict scheduling conflicts before they happen,” you will get something genuinely useful.
The other thing that matters is intellectual honesty. It is tempting to cherry pick the AI responses that confirm what you already believe and ignore the ones that challenge you. The whole point of this process is to seek disconfirming evidence. If you are only using AI to feel better about your idea, you are wasting the tool’s most valuable capability.
Common Mistakes I Have Made
I want to be honest about the mistakes I have made with this process, because they are instructive.
The biggest one is accepting AI’s framing of the problem without questioning it. When you ask AI to analyze a business idea, it will often organize its analysis around the categories it deems relevant. But sometimes those categories miss the actual crux of the matter. I have had cases where AI gave me a thorough analysis of market size, competition, and unit economics while completely missing the real risk, which was whether the target customer even recognized the problem I was solving. Learning to push back on AI’s framing, to say “you are analyzing the wrong dimension, let me redirect you,” is a skill that takes practice to develop.
The second mistake is using AI as a replacement for talking to real people rather than as preparation for those conversations. The AI analysis should make your human conversations better, not eliminate them. I went through a phase where I was doing so much AI analysis that I delayed customer conversations for weeks. That was exactly backwards. The AI analysis is valuable precisely because it prepares you to ask better questions when you do talk to real people.
The third mistake is not documenting the process. I now keep a running document for each idea where I paste the key AI interactions, my reactions, and the follow up questions that emerged. This creates a decision log that is invaluable when I revisit an idea weeks or months later. Without that documentation, I found myself repeating the same analysis multiple times because I had forgotten what I had already explored.
The fourth mistake, and this one is subtle, is conflating AI’s confidence with actual evidence. AI presents its analysis with a certainty that can feel authoritative, but it is synthesizing patterns from training data, not reporting empirical facts about your specific market. Treating an AI analysis as equivalent to real market validation is a dangerous shortcut that I have caught myself taking more than once.
What AI Still Cannot Do
Here is the thing: AI can reason about ideas, but it cannot validate them. It cannot tell you whether real customers will pay. It cannot gauge whether the timing is right. It cannot assess whether you are the right person to build this specific thing. That still requires getting out of your head and talking to people, running experiments, and testing your assumptions against reality.
AI is also limited by its training data. It is excellent at pattern matching across known domains but weak at identifying genuinely novel opportunities where the patterns have not been established yet. If your idea is truly new, AI may underestimate it because there are no comparable examples to draw from.
The process I have described takes maybe 30 to 45 minutes per idea. It sharpens my thinking, surfaces blind spots, and tells me exactly what to go investigate next. That is not nothing. That is actually a lot. And it means that by the time I sit down with a real advisor, potential customer, or investor, I am coming with a much sharper version of the idea than I would have otherwise.
The best part is that the process is iterative. After real world conversations and research, I bring what I learned back to the AI and run the analysis again. Each cycle gets tighter, clearer, and closer to the truth of whether this idea has legs or not.
If you are exploring business ideas right now, I would strongly recommend building this kind of AI assisted analysis into your process. It will not replace the judgment calls that only you can make, but it will ensure that when you make those calls, you are making them with a much clearer picture of what you are actually getting into. In entrepreneurship, the quality of your decisions in the early stages determines everything that follows. Anything that improves the quality of those early decisions is worth your time.
Laurie Stach 