not words

built

for

actions,

actions,

built for

actions,

actions,

Foundation models

Foundation models

not words

We’re a research lab building the best foundation

We’re a research lab building the best foundation

models in the world that act on computers.

They will automate anything.

We're a research lab building the next generation of foundation models that act on computers and the real world.

Words built the last wave of AI. Actions will build the next. Computers will finally use computers.

Words built the last wave of AI. Actions will build the next. Computers will finally use computers.

Words built the last wave of AI. Actions will build the next. Computers will finally use computers.

Today’s foundation models can generate words, but still cannot reliably operate a computer, where most real economic value will be unlocked. We’re building models that learn from human behaviour at scale, powered by the largest dataset of computer interactions ever collected.

Our models will execute tasks autonomously with reliability and speed, unlocking entirely new classes of products the world hasn’t yet imagined.

Our models will execute tasks autonomously with reliability and speed, unlocking entirely new classes of products the world hasn’t yet imagined.

Today's models owe their success to unfathomable quantities of pretraining text generated by billions of humans over the course of four decades of internet use. This scale of data has not existed in the domain of computer use - until now.

In pursuit of

In pursuit of

Universal Agents

Universal Agents

Today’s foundation models can generate words, but still cannot reliably operate a computer, where most real economic value will be unlocked. We’re building models that learn from human behaviour at scale, powered by the largest dataset of computer interactions ever collected.

Our models will execute tasks autonomously with reliability and speed, unlocking entirely new classes of products the world hasn’t yet imagined.

Our models will execute tasks autonomously with reliability and speed, unlocking entirely new classes of products the world hasn’t yet imagined.

Today's models owe their success to unfathomable quantities of pretraining text generated by billions of humans over the course of four decades of internet use. This scale of data has not existed in the domain of computer use - until now.

In pursuit of

Universal Agents

Our Philosophy

Tomorrow's data scale defines today's infrastructure. The behaviour data we’re collecting will soon become much larger than any text-based training corpus in the world. We're building infrastructure to handle this from day one. Our systems are designed for the long haul, and our team is comprised of ruthless performance engineers to get us there.

Great products drive great research, not the other way around. Computer use models are still in their infancy: today they fail too often and move too slowly to automate most real work. The biggest product unlocks will come as the models mature, and the only path there is through building widely-used products that surface utility at scale and create the feedback loop research needs.

Models that take seconds to act aren't models that act. The best computer use frameworks today take up to tens of seconds per action. That's unusable for real work. We're targeting sub-second latency and we’re not afraid to ditch transformers to get there.

About us

We’re researchers who build. We scaled reinforcement learning infrastructure at Cohere and shipped the Command models that set new state-of-the-art.

We move fast, stay flat, and do whatever it takes to solve hard problems. We’re backed by Y Combinator and are hiring exceptional engineers who want to push the frontier of computer-use models. Leave us a note at hiring@inductionlabs.com if that sounds like you.

Born in SF, backed by Y Combinator

induction labs

All rights reserved © 2025

Born in SF, backed by Y Combinator

induction labs

All rights reserved © 2025

Born in SF, backed by Y Combinator

induction labs

All rights reserved © 2025