our manifesto
Composable AI: Build Prod, Not God
The only evidence AGI isn't already here is the embarrassing lack of impact modern AI has had. TypeSafe AI is focused on how to unlock the intelligence already inside frontier models into something far more useful for the world.
Our mission is to pave the shortest path to an AI-based economic revolution by making intelligence composable to steward a Cambrian explosion of intelligent software. ¹
We already have general intelligence
General intelligence is a spectrum, and we're clearly somewhere far along it. Today's models know an enormous amount, reason across domains, and do things that looked impossible a few years ago.
Yet after years and trillions of dollars invested in the industry, most software still isn't meaningfully intelligent, and life is mostly the same for the average person. That should tell us something: the bottleneck isn't raw intelligence. It's that today's intelligence is hard to build on.
Beyond horseless carriages
Early cars were designed as horseless carriages: rather than reimagining transportation from scratch, inventors took the familiar carriage and replaced the horse with a motor, while keeping the high seats, the buggy springs, and even the whip socket in some models. It’s a clear example of how new technologies are often forced into the shape and assumptions of the thing they replace before finding their own native form.
Current AI is trained to be a helpful, articulate, pleasant assistant. ² Reasonable goals if you assume a human is on the other side of the model. Yet the foreseeable consequence is AI that requires humans in the loop instead of running in the background.
Software has never worked that way. Even the most complex software is built out of simple logic and layered abstractions, with every branch auditable. We want AI to work alongside existing software as a primitive that any programmer can invoke for semantic judgement and decisions, while still using code for what it’s best at: exact computation.
Computers can do so much by just branching on bits, imagine if they could also branch on common sense. ³
This is the opportunity for true, machine-native composable AI.
Early cars were designed as horseless carriages: rather than reimagining transportation from scratch, inventors took the familiar carriage and replaced the horse with a motor, while keeping the high seats, the buggy springs, and even the whip socket in some models. It’s a clear example of how new technologies are often forced into the shape and assumptions of the thing they replace before finding their own native form.
Current AI is trained to be a helpful, articulate, pleasant assistant. ² Reasonable goals if you assume a human is on the other side of the model. Yet the foreseeable consequence is AI that requires humans in the loop instead of running in the background.
Software has never worked that way. Even the most complex software is built out of simple logic and layered abstractions, with every branch auditable. We want AI to work alongside existing software as a primitive that any programmer can invoke for semantic judgement and decisions, while still using code for what it’s best at: exact computation.
Computers can do so much by just branching on bits, imagine if they could also branch on common sense. ³
This is the opportunity for true, machine-native composable AI.
Composability leads to safe emergence
The people who built databases didn't imagine Google. The people who built internet protocols didn't envision Stripe. They made lower-level capabilities so dependable that they could not only run in the background, but also be layered on top of. A Cambrian explosion of software emerged that nobody could have designed top-down.
Intelligence today is like databases before SQL: powerful, but every use is bespoke. Once a smart decision becomes as dependable and invokable as a database query, builders will stack them the same way. The intelligence revolution will be like the early internet: unplanned, distributed grassroots efforts built by builders for builders and owned by all. ⁴
Safety is a precondition for composability. You let a component run unattended if it's reliable; you only build on top of it if it's trustworthy. It takes trust to bury a dependency five layers deep in a system. People will only do so if they can inspect it, test it, and constrain it piece by piece, the way we've always engineered dependable software. That's how the composable path enables an ecosystem of small, legible primitives to safely evolve into trustworthy, complex systems.
The people who built databases didn't imagine Google. The people who built internet protocols didn't envision Stripe. They made lower-level capabilities so dependable that they could not only run in the background, but also be layered on top of. A Cambrian explosion of software emerged that nobody could have designed top-down.
Intelligence today is like databases before SQL: powerful, but every use is bespoke. Once a smart decision becomes as dependable and invokable as a database query, builders will stack them the same way. The intelligence revolution will be like the early internet: unplanned, distributed grassroots efforts built by builders for builders and owned by all. ⁴
Safety is a precondition for composability. You'll let a component run unattended if it's reliable; you'll only build on top of it if it's trustworthy. It takes trust to bury a dependency five layers deep in a system. People will only do so if they can inspect it, test it, and constrain it piece by piece, the way we've always engineered dependable software. That's how the composable path enables an ecosystem of small, legible primitives to safely evolve into trustworthy, complex systems.
A new shape of intelligence
We're building the paradigm shift beyond AGI.
Step 1
Ship the shape of machine-native composable AI with the highest possible intelligence-per-dollar.
Step 2
Make our AI reliable enough to transform the economy via real automation.
Step 3
Empower the world with higher-level intelligence abstractions that are stable enough to compose and layer upon, for a collaborative, emergent future.
As we say at TypeSafe:
We're building prod, not God.
1
There are many definitions of this, but our favorite is global TFP growth reaching 3% within five years and holding at that level for ten (unprecedented in economic history, but achievable if AI's gains diffuse across the real economy).
2
RLHF, the algorithm almost all of current AI is trained with, directly optimizes for human preference. Modern “reasoning” models are trained with RLVR as well, which optimizes for programmatically evaluatable tasks (i.e., benchmarks).
3
This is the dream of neuro-symbolic AI: neural networks for perception paired with symbolic logic for reasoning. Sometimes cheekily summarized as “smart if-statements.”
4
On top of rock solid infrastructure built by people who care A LOT. 🥹