Why building for one AI platform is a losing bet.

The bet was never really about the platform

Building a bespoke integration with one specific AI assistant felt like a complete solution: custom-formatted data, a direct partnership, tailored to exactly how that one assistant wanted information delivered. It worked well, for that one assistant. Then a second AI agent arrived, wanting something differently shaped, and the small distillery behind that first integration, known for a popular tasting-room tour, found itself starting again from close to nothing.

The mistake wasn't choosing the wrong assistant. Any specific choice was always going to be a bet on one winner in a market that hasn't settled on one, and betting on a single platform in an unsettled market is a bet against the odds by definition, not because that particular platform was a poor pick. The distillery would have hit the same wall with any single bespoke partnership, because the problem was never which platform, it was answering "which one" at all.

With hindsight it's tempting to imagine a better bet was available, the platform that would still be the right one two years later. That option was never really there to choose in a genuinely unsettled market; nobody picking a single winner today has a reliable way to know which platform that will turn out to be.

An open standard answers a different question

Building to an open, published way of describing tour availability, group sizes and booking terms, rather than to one assistant's private format, means the work serves whichever surface adopts that standard, not just the one the distillery happened to build for first. A new assistant that speaks the same standard can connect to the distillery's real facts without a single custom integration, because the format was never proprietary to begin with. The first standard-based connection costs roughly the same as the first bespoke one did. Every connection after that costs almost nothing.

Keeping the platform relationships that are still worth having

None of this means ignoring the AI assistants that already send real visitors. The underlying facts, tour times, group sizes, cancellation terms, live in an open, standard shape once, and each specific platform connects to that shared version rather than getting its own custom-built copy. A platform-specific relationship can still be worth cultivating for other reasons, visibility, a partnership, promotion. It just shouldn't be where the distillery's actual facts live, because that part has to serve everyone, not one partner.

The second rebuild is what makes the first mistake visible

The real cost of the bespoke approach rarely shows up the first time. It shows up the second time, and the third, each new surface repeating the same custom work from close to zero. A distillery that's been through that twice already knows exactly what the open-standard alternative would have saved; one that hasn't yet is the one still deciding which bet to make.

A business's real facts, held once in an open, standard shape rather than shaped around one platform's preferences, is the specific thing Selfe builds toward.

Should we avoid partnering with individual AI platforms altogether?

No, a specific partnership can still be worth having for visibility or promotion. It just shouldn't be where the underlying facts live, since that part needs to serve every surface, not one partner.

How do we know if we're already locked into one platform's format?

If connecting a second AI surface would mean redoing most of the original integration work rather than a smaller connection task, that's the sign.