Post

After years of failed startup ideas, I tried something different. Instead of building on a hunch, I let AI run a full market analysis to find a validated problem worth solving. Here's what I learned and how you might do the same. #buildinpublic

First, a caveat: ChatGPT won't cut it for this. You need an AI that can independently research the web, pull real sources, and come back with evidence -- not just confident-sounding hallucinations. I used Manus AI and gladly paid the $40/month for this.

My initial prompt was simple: "I'm a Ruby developer who wants to build a B2B SaaS. Analyze potential opportunities, validate market need, and recommend ideas with real pain points." Then I pushed it to go deeper on each recommendation.

The AI came back with 7 initial ideas. But I didn't just pick one. I asked it to validate each with real evidence: Reddit threads, G2 reviews, market data. It provided actual links -- not vague claims. I could verify everything myself.

Here's the thing: I had my own ideas too. I pitched a few to the AI, expecting it to validate them. Instead, it shot them down hard. Explained why the market was too crowded, why the problem wasn't painful enough, why I'd struggle to compete.

That was actually reassuring. It wasn't just telling me what I wanted to hear. It was giving it to me straight. And that's the benefit of this approach -- no ego, no confirmation bias, just cold, hard research.

One idea kept surfacing: runbook discovery for on-call engineers. The AI found Reddit threads like "Senior SRE who knew all our incident procedures just left -- now we're screwed" and G2 reviews complaining about scattered documentation.

But I didn't stop there. I asked: "Is this even technically feasible?" The AI researched GitHub, Notion, and Confluence APIs -- rate limits, webhooks, authentication. It came back with a full technical feasibility report. Answer: Yes, very feasible.

Then: "Would this violate any Terms of Service?" It analyzed GitHub's Developer Agreement, Notion's API Terms, Atlassian's policies. Found precedents like Glean and Raycast. Verdict: Low risk, as long as you use official APIs and respect data deletion.

Next question: "Why wouldn't a big company like Atlassian just build this?" The AI explained the Innovator's Dilemma: the market is too small for them to care ($5-10M is life-changing for me, a rounding error for them), and their platforms are too horizontal.

I also asked about pricing strategy. Should it be per-user? Per-integration? Usage-based? The AI analyzed competitor pricing, researched willingness-to-pay studies, and explained why usage-based pricing creates perverse incentives for this product.

The data was compelling: $9,000/minute average downtime cost, $3.3B TAM, and most importantly -- Reddit full of engineers asking "Does anyone have a tool for this?" The problem wasn't theoretical. It was validated by real people with real pain.

Could I have done all this research myself? Sure. But it would've taken weeks, and I would've brought my own biases. I would've hit dead ends and convinced myself to ignore red flags. The AI did it in hours, objectively, and without attachment to any outcome.

Am I 100% certain this will work? No. This isn't a slam dunk. But I'm more confident than I've ever been starting something new. The research is solid, the problem is real, and I've personally felt the pain. That's as good as it gets before you start building.

So now I'm building Runbooky (www.runbooky.com ): a unified discovery platform for runbooks. Find the right doc in seconds, no matter where it lives. It's the tool I wish I had during those 3 AM incidents.