Common questions
What do you mean by an AI compiler?
Today, developers have to adapt prompts, tools, workflows, and reasoning settings to the model underneath. We want people to specify what they want an agent to do, while the compiler figures out how to make the underlying intelligence do it.
Why does AI need a compiler?
Before compilers, software had to be written around the hardware it ran on. AI is in a similar state today: every model behaves differently and changes to models, versions, or reasoning settings can change what works. A compiler separates what the developer wants from the intelligence used to execute it.
What are you building today?
We're starting with model migrations. Give us an agent that works on one model and we adapt its prompts + other traits so it works on another.
Why are model migrations part of a compiler?
A migration is the simplest version of the compiler problem: preserve what the developer wants while changing the intelligence underneath. From there, the same problem expands to continuously deciding how one agent - and eventually fleets of agents - should execute human intent.
What enables Propensity Labs to do this?
We systematically study how models behave and which interventions change that behavior. Every model, environment, and intervention adds to a dataset that makes the next problem easier.
Where does this ultimately go?
We believe AI agents will eventually do essentially all economic work. At that point, the fundamental programming problem is no longer how to perform the work. It is how humans specify what they want and reliably turn that into machine execution. We want Propensity to be that compiler.
Why a Public Benefit Corporation?
So we stay independent and can study every lab's models without being tied to any of them. We are committed to reducing the risks from advanced AI systems.
Can I try out your product or collaborate?
Yes! Reach us at info@propensitylabs.ai.