One setup, every surface: what that actually requires.

What running one setup actually looks like

Ask about a canal boat's availability today and the question might come from a direct search, an AI assistant, a comparison tool, or an AI agent nobody at the hire company has ever heard of, checking on behalf of someone planning a week on the water. What decides whether that question gets a true answer isn't which surface asked. It's whether the fact it's asking about, this boat, these dates, has exactly one place where it's actually kept up to date.

In practice, that means naming the fact and naming its home before anything else: availability lives in the booking calendar, cancellation terms live in this one document, anything unusual gets confirmed by a named person. A new surface then connects to that home directly, rather than being handed its own copy to keep current, which is the detail that decides whether one setup stays true six months in, or quietly stops being true the first time someone's in a hurry.

Drift is nobody's mistake, it's what happens by default

The boat company's alternative, a bespoke answer for each surface, a feed for a listings site, a manually updated sheet for an assistant partnership, its own copy of the same terms, doesn't fail because anyone gets careless. It fails because keeping three copies aligned was never actually anyone's job, only an assumption sitting quietly underneath everyone else's.

The drift is rarely anyone's mistake. It's what happens by default when the same fact lives in three places and gets updated in one of them. Nobody at the company decided the feed and the sheet should disagree; they simply stopped being the same document the moment someone updated only one of them, which is the ordinary, unremarkable way separate copies of a fact end up disagreeing eventually. An agent checking both would notice immediately. A person would probably never think to compare them at all.

What one setup actually requires, day to day

The pillar's version of this argument is the decision: work out once what's true and where it genuinely lives. What that decision requires afterwards is less glamorous, and easy to under-invest in. It means resisting the reasonable-sounding shortcut of updating the quick version now and the proper source later, because later is exactly where drift gets in. It means a new connection reading from the live source directly, not from an export somebody remembered to run last Tuesday. And it means treating whether this is still one setup as an ongoing check, not a box ticked the day the boat company first wired everything up.

None of that is difficult in isolation. What makes it hard is that it has to hold on the day nobody's paying attention to it, which is most days.

The payoff compounds precisely because the discipline holds

A company that's done this once doesn't dread the next new AI agent or assistant the way it might have dreaded the last one, because the next connection is small, not a fresh project. That's the advantage a bespoke, surface-by-surface approach never earns: it turns every new surface into a first-time cost instead of a marginal one.

Selfe does the less visible part of this job too, not just building the first connection but keeping the discipline behind it in place, so the boat company's one true answer stays one true answer on the days nobody's watching for drift, not only on the day it was first set up.

Doesn't every surface need different information anyway?

Mostly not. Most surfaces are asking versions of the same underlying questions, availability, terms, price, formatted differently. The underlying facts are shared even when the format isn't.

What happens when a genuinely new kind of surface appears?

The same real facts still answer it. What's new is usually the connection, not the underlying information the company already keeps current.