TypeSafe, an artificial intelligence startup led by former OpenAI researcher Diogo Almeida, emerged from stealth on September 17 with a $40 million seed round. The financing was led by DCVC, with general partner James Hardiman backing the deal. Almeida founded the company with Erik Gafni and Sasha Sheng to build models designed for software systems rather than for people typing into a chat window. The size of the round is large for a company only now stepping into public view.
Almeida helped develop reinforcement learning from human feedback, a technique central to ChatGPT, during his time at OpenAI. TypeSafe describes its work as machine-native, or composable, artificial intelligence, and frames itself as a research lab rather than a maker of consumer chatbots. The founders argue that conventional models cause problems in production because their outputs can be unpredictable, which makes them awkward to embed inside automated software pipelines that need consistent results.
The company’s first product is called Jev, a name drawn from the Jevons Paradox. TypeSafe says the system runs with latency below 100 milliseconds and can be far cheaper and faster than comparable models. Features include parallel processing of outputs and calibrated confidence scoring, which the team says lets software make autonomous decisions with a clearer sense of how reliable each answer is before acting on it.
“The reason why LLMs work as well as they do is that they’ve been optimized for human-preference, which was a big part of my work at OpenAI,” Almeida said. He argues that the same design choice keeps people tied into the loop and limits how far automated systems can run on their own without checking back with a human at each step.
TypeSafe opened an early-access waitlist for developers and said it would use the seed money to expand its team and prepare the product for wider release. The round adds to a run of well-funded AI labs founded by researchers who have left the largest model developers. Investors are betting that models tuned for machines, rather than conversation, will find demand as companies push more work onto autonomous software over the coming years.
