Most startup advice sounds reasonable on the surface. “Find a problem.” “Scratch your own itch.” “Build something people want.” I followed all of it, and still watched smart founders, including myself, spend months building things that quietly went nowhere.

That contradiction bothered me enough to dig deeper. I started reading failure postmortems, dead Product Hunt launches, abandoned GitHub repos, and long Hacker News threads where people explain why they passed on an idea.

After a while, a pattern emerged. The problem isn’t that people lack ideas. It’s that we evaluate ideas using storytelling logic instead of survival logic.

We choose ideas that sound interesting, feel ambitious, or look good in a pitch, but collapse the moment they meet real-world behaviour.

What kills most ideas isn’t competition or execution. It’s that they don’t replace anything urgent.

They don’t map to an existing habit, a recurring cost, or a painful workaround someone is already using.

When you ask simple questions like who is paying, what they stop doing when they adopt this, and why now, most ideas fall apart very quickly.

To stop repeating this mistake, I began writing ideas down as hypotheses instead of inspiration.

Each idea had to survive a few uncomfortable questions: what existing behavior does this replace, what would kill it in the first thirty days, and what is the smallest experiment that could prove or disprove demand.

Most ideas failed immediately. A few survived longer than expected.

Over time this turned into a private database I used to avoid wasting months on weak ideas. It wasn’t a collection of “great ideas.”

It was a record of ideas that survived brutal filtering, along with many that didn’t. Eventually I cleaned it up into something others could browse, now called startupideasdbcom (google it), mostly because I kept wishing something like this existed earlier.

If you’re stuck choosing what to build, or tired of clever ideas that die quietly, this might save you some time.

And if you disagree with the framework, I’m genuinely interested in where it breaks, Hacker News usually finds the flaws faster than anywhere else.

The View from Inside the AI Bubble