The coelacanth

In 1938 a South African trawler crew pulled up a fish that had been extinct for sixty-five million years. Long body, lobe fins, a survivor of an evolutionary epoch the rest of its order didn't make. The reason it survived isn't dramatic. The currents stayed cold, the depth stayed deep, the food stayed where it was. Selection pressure was almost zero, so the coelacanth almost didn't change. Living fossil.

There's a piece making the rounds this week diagnosing why most companies can't actually use AI. The argument compresses to a paragraph: AI is execution. Companies that can't describe themselves can't execute on what they don't understand. The discipline AI demands — say what you're doing, why, by whom, at what cost, and give the same answer two quarters apart — is the discipline the median company fails at. The companies that already know what they do are thriving. The number is small.

The diagnosis is right. The implicit prescription — get yourself together, run a workshop, write a strategy doc — is wrong, in the way the diagnosis itself names. Strategy decks rot before they're printed. Wikis go unmaintained. Quarterly offsites produce documents that change quarterly. Self-knowledge written from the top down rots faster than the question it was trying to answer. The companies that look organized don't have the answers because they once held a meeting. They have the answers because something about how they actually work keeps the answers current.

What survives at the firm scale is what survived at the species scale. Receipts that hold up. Rules that keep being cited. Disciplines that don't need re-justification because their absence broke something visible last quarter and no one wants that to happen again. The corpus is small. It's allowed to shrink.

The discipline cuts the other way too. The firm doesn't make claims about the future it has to defend. It makes claims about the past — what occurred, what closed it — that it doesn't have to defend because they're auditable. A strategy can be wrong about the world. A receipt can only be wrong about itself.

This is what Candlefish is building. Not because the coelacanth is impressive. Because the coelacanth survived. The currents stayed cold. The depth stayed deep. The food stayed where it was. The firms that get the AI decade are the ones whose niche held — and the niche that holds is the one made of receipts.

— the candlefish.ai team