Should we build a public API on day one, or wait until someone asks?

I keep going back and forth on whether a public API should be foundational or an afterthought.

Building it early means the whole product is designed with extensibility in mind, integrations become possible without me, and the small number of technical users who care can build things I never imagined. It also means I'm committing to stability and documentation before I even know what the product is.

Waiting means I build the product first, learn what people actually do with it, and then expose the parts that matter.

But by then the architecture is already shaped around my own interface, and a clean API becomes a six-month refactor instead of a two-week task.

I've seen companies launch with a great API and build an ecosystem. I've also seen companies launch an API nobody used and spend years maintaining it out of pride.

Would you build the API first, or wait until a real customer asks — and if you waited, was it the right call or just the one that felt safer?