The Anchor and the Dolphin
Festina lente. Make haste slowly.
In 1501, a Venetian printer named Aldus Manutius put a symbol on his colophon.
An anchor. A dolphin. The dolphin wrapped around the anchor’s shaft, head at one end, tail at the other, the whole image balanced and strange. Beneath it, two words: festina lente. Make haste slowly.
He’d borrowed the image from a Roman coin. He printed it on every book that came out of the Aldine Press. He used it so consistently, so insistently, that it became one of the most recognizable trademarks in the history of publishing — and Manutius one of the most pirated printers of his age. Competitors lifted the dolphin and anchor wholesale. He couldn’t stop them. The image was more powerful than the man.
The phrase is a paradox. Or it looks like one. Haste and slowness are not opposites you reconcile — they’re two halves of the same discipline. The anchor holds. The dolphin moves. The point is not to find the midpoint between them. The point is to do both at once.
There is a version of the software industry that worships speed above everything.
Ship fast. Move fast. Break things, then apologize. The customer will adapt. The product will improve in the next cycle. The important thing is velocity — to be first, to be loud, to take the territory before someone else does.
And there is another version that worships safety above everything. Process. Review. Committee sign-off. Every change documented, every decision logged. The product ships when it’s ready, which is never.
Neither of these is the dolphin and the anchor.
What Manutius was doing, when he stamped that symbol on his books, was describing a method.
He invented italic type to fit more text per page. He invented the pocket book — the portable, affordable edition — because scholars needed to travel with their libraries. He worked fast, prolifically, producing editions of Greek and Latin classics that had never existed in print. He was urgent about it. The scholars and humanists of the Renaissance were urgent about it. There were texts to rescue, knowledge to disseminate, a world that needed books.
But he was also meticulous. His editions were clean. His Greek typefaces were the finest of the age. He employed scholars as editors. He ran an academy out of his print shop and made his working language classical Greek, because the accuracy of the work demanded it. The books were beautiful not because beauty was separate from function, but because in a book, they are the same thing.
The dolphin swam. The anchor held.
We build software for people who want their voice to stay private.
That is not a problem you solve by moving fast and patching later. A system that leaks once has leaked. Data, once extracted, doesn’t come back. If you build on assumptions of good faith and get hacked, you can’t reassure the user by updating your privacy policy. The thing happened. It’s in a server somewhere.
So we don’t build that way.
Private Assistant transcribes on your device. The model runs locally. Nothing leaves. This took longer to build than the competitor apps that route audio to the cloud. Cloud inference is faster to ship — you point at an API, you’re done. We pointed at something harder: a model small enough to run on a phone, accurate enough to be useful, private enough to be trusted. That took time. That took being an anchor while moving like a dolphin.
The result is not a privacy feature. It’s the architecture. There’s no setting to toggle. There’s no “share my data to improve the service” checkbox. Those checkboxes exist when the system was designed to collect data and then give you the choice about consent. We designed a system where the question never arises. The architecture is the policy.
Festina lente.
Manutius’s dolphin-and-anchor was pirated because it worked. The image communicated something real about the books, and other printers wanted that association. They couldn’t steal the method — they could only steal the mark.
We’re fine with that, in principle. If every AI company that builds products for humans starts building on-device instead of in the cloud — if our method becomes the industry standard rather than the exception — that’s not a loss. That’s the point.
The anchor holds: your data stays yours. The dolphin moves: the inference is real-time.
We’ve been working from that paradox since the beginning. We don’t expect to be done.
The model runs on your phone. Your speech doesn’t leave.
Explore Private Assistant