The model layer
finally got cheap.

For thirty years personal computing was about the *individual*. Word processors that learned your verb tics. Drawing programs that shaped to your hand. Spreadsheets that matched your way of thinking about money. Software was an instrument; you played it; it took on the shape of you over time.

Then SaaS arrived. The software stopped being yours. It became a system of record for someone else — a manager, a CRM, a board. We performed work into dashboards instead of doing it on instruments.

Two curves crossed in 2025. The model layer collapsed in price by 40× in eighteen months. The compute layer collapsed in price too — a tenant per user, with a real model trained on their data, fits the unit economics.

Before that point, *personal* software was a beautiful idea you couldn't ship. After it, the only reason not to ship personal is inertia. We're trying to be unsentimentally not inert.

We are not a chatbot with a fresh personality stitched on. Not a wrapper around someone else's LLM. Not "AI that works for everyone." It is AI that works for *you* — one person at a time, slowly accumulating the patterns of how you write, schedule, decide, drop the ball.

There is no shared training run, no aggregate insight. You own the model. You can read it. You can edit it. You can delete it.

Personalised software is what most products mean when they say AI. It's a feature on top of someone else's product, tuned with your settings. *Personal* is different. Personal is an instrument that took on your shape because it spent six months reading what you wrote, replied to, scheduled, ignored.

We're betting that the next twenty years of software is the second kind. We're building the second kind first.