I asked an AI to interrogate my startup idea. Here's what honest product thinking actually feels like.

I walked into the session with a product I thought I understood. I'd written the PRD. I had a lean canvas. I had a competitor analysis. I had a name. Blockt — a habit-tracking and time-blocking app with an AI layer — felt well-defined. I knew what it was.

Two hours later, I understood it differently. Not because the AI told me what to build. But because it kept asking questions I'd been quietly avoiding.

The first question I couldn't dress up

The session opened with a demand question: what's the strongest evidence you have that someone actually wants this — not "is interested," not "signed up for a waitlist," but would be genuinely upset if it disappeared tomorrow?

I pointed to a GitHub project in the same space that had picked up 100,000 views on RedNote in three days and pulled 5,000 people into a group chat — non-technical users learning to deploy software locally just to run it. Strong signal, I thought.

The AI acknowledged it, then pushed: that's effort, that's pull — but what are those people pulling toward, specifically? And how close is that to what my product does?

I had to admit the two products were adjacent, not identical. And that I had no willingness-to-pay evidence at all. Most people would dress that up. I didn't. And the session moved faster because of it.

Then it asked me to name a real person

Not a persona. Not a demographic range. A person — their name, their situation, what keeps them up at night, what would make them pay without hesitating.

I said: me.

"I am so anxious about things I want to do and said I'd do but I never really started them. I tried time-blocking, but I always turn to other things like my phone, and I cannot focus during that time. I procrastinate a lot... I don't know when to relax."

The AI pointed back at one line: "I tried time-blocking, but I always turn to other things like my phone." I had already tried the core mechanic of my own product. And it hadn't fixed my problem. My actual pain was downstream of that. That one observation reframed the rest of the conversation.

The assumptions I'd been carrying quietly

Once I named myself as the primary user, the session shifted from pitch mode to diagnostic mode. We mapped the actual workflow — the reminder apps, the note backlogs, the calendars people already rely on — and traced exactly where the breakdown happens.

A mid-session second opinion from an independent AI agent, which was given no context from our conversation, flagged something I'd mentioned almost in passing: a manual check-in sheet I share with friends every day. It identified that as potentially more significant than anything in the PRD. I'd built a working prototype of one of the product's key ideas without realizing it.

That's the thing about the problems you live inside. You stop seeing your own workarounds as data.

What AI-assisted thinking actually felt like

What this session did that a solo thinking session wouldn't: it held my claims accountable in real time. When I tried to sidestep choosing a target user by saying two different problems "should be considered together," it named the exact product consequences of that choice and asked me to defend it rather than accept the reframe. When I cited viral numbers as demand evidence, it immediately separated interest from intent.

The discomfort was the useful part. Good product conversations should make you sit with things you'd rather skip past. This one did that without feeling adversarial. At one point I pushed back on a reframe it offered, and what came out was a more nuanced version of my original premise that I hadn't been able to articulate before. The resistance produced the clarity.

Three things I'd tell any early-stage founder

1. Name your user early — not as a cop-out, as a diagnostic. If you can't describe their pain in two sentences without hedging, you don't understand the problem yet. If you can, you have the most valuable starting point in early product work.
2. Users' workarounds are your roadmap. The manual system they built for themselves, the thing they do every day that no app quite handles — that's a prototype. The product question is how to make it 10x better, not how to build from scratch.
3. AI thinking partners are most useful when you don't fold to them. The sessions where I pushed back produced the sharpest ideas. The goal isn't agreement with the AI — it's using the friction to pressure-test your own thinking.

Where this left me

I came out with a clearer product, a better-defined user, and a first assignment: find three people outside my immediate circle, put the core flow in front of them, and ask what would make them stop using it after one week.

No launch date. No willingness-to-pay evidence yet. What I have is a version of the product I actually believe in, and a much clearer sense of what I'd been building around rather than building toward. That feels like a real start.