our mission
Pave the shortest path to an AI-based economic shift by making intelligence composable — so a lot of software, written by a lot of people, can actually use it.
We already have general intelligence. The moving target called AGI is not the bottleneck. Today’s models already know enough to create real economic value. Most software is still not intelligent. Life is still mostly the same. That is a packaging problem, not an IQ problem.
Early cars were horseless carriages: the old shape, a new motor, even a whip socket. Chat is that stage of AI. A helpful assistant assumes a human on the other side. Software never worked that way. It is simple logic, stacked, every branch inspectable.
We want a primitive any programmer can call for judgment, while code keeps doing exact work. Computers already branch on bits. They should also be able to branch on common sense.
Safety here is practical. You let a component run unattended if it is reliable. You bury it five layers down only if you can test it and constrain it. That is how databases and protocols became things nobody designed top-down, and still ran the world.
Appendix, in short: the missing piece is automation; the economic bet is TFP actually showing up; the method is preconditions, not a single cathedral; RLHF optimized for being liked; the dream is a smart if-statement.
Early access is opening in waves.