Open source killed it
You’re working on an exciting problem with real user traction. You built a model, ran trials, feedback looks great. Then a big tech company releases “the same model” for free (open source), but they poured 100x more compute into it because they can. It’s better.
Why don’t they monetize? Because a trillion-dollar company doesn’t care about a sub-$100M market. They release it to sell compute, gain reputation, or simply because they can.
§
I’ve been there. We were building AI-generated children’s booklets. We solved many problems. Eventually Nano Banana wiped out all the progress. By then we had shut down. Some time wasted, but more time saved. The models were going to catch up.
§
The wrapper pivot is one option. Add better UX, tighter integrations, ship faster than the model providers. It works. For a while. But your margins eventually collapse to whatever the model provider decides to charge.
Cursor did this right. They started as a UX wrapper, then used the data they collected to build their own models. The wrapper was a bridge, not a destination.
If you build a wrapper, have a medium-term plan for what comes after.
§
The real lesson: some problems are more defensible than others - they’re harder to commoditize or less appealing to Big Tech bros.
Before you commit, ask: if a well-funded lab open-sources a better version tomorrow, what’s left? Data, distribution, expertise?