Not everything survives.

A SEMrush for stock media. It analyzed searches across the major stock platforms to surface high-demand, low-competition niches, so photographers could shoot what actually sells instead of guessing.

It died the month generative AI went mainstream. The whole premise was helping people find the right photo to shoot. When anyone could generate the exact image they needed on demand, the market I was optimizing for stopped existing. Not a pivot. The ground moved.

What it taught me: you can execute well and still be on the wrong side of a shift you didn't see coming. Timing isn't a detail. It's the whole bet.

A context-aware reminder assistant. You'd send it a task by Telegram, a screenshot, or a note, and it would surface the right one based on where you were and what your day looked like. Walk into a shop, your list appears. Capturing cost nothing, remembering cost nothing.

It worked. For me. Generalizing it for other people would have meant a lot more engineering, and I never got any market validation to justify the work. So I stopped building it as a product. I still use it every day.

What it taught me: I'll happily pour engineering into a problem before checking whether anyone else has it. Validation first, build second. (Also: a tool that only serves you isn't a failure. It just isn't a business.)