TypeSafe AI, the San Francisco startup founded by former OpenAI researcher Diogo Almeida, exited stealth on September 15, 2026 with a $40 million seed round led by DCVC. The company is pitching a new class of “machine-native” AI: composable models designed to slot into software as a primitive rather than converse with humans.
A Model For Software, Not Chat
The startup’s flagship, Jev, promises frontier-level semantic judgment at sub-100 millisecond latency, streams hundreds of parallel outputs from a single prompt and returns calibrated confidence scores so developers can build autonomous decisioning around it. TypeSafe claims Jev runs roughly 100x faster and cheaper than the leading frontier chat models when used for narrow, high-volume tasks like classifying support tickets, moderating content or filtering trades.
“Most intelligence should eventually live inside software, running quietly in the background,” Almeida said, framing Jev as an alternative to routing every decision through a giant conversational model. Co-founders Erik Gafni and Sasha Sheng lead product and research.
DCVC Bets On Predictability Over Personality
James Hardiman, the DCVC partner who led the round, described TypeSafe’s wedge as turning “increasingly capable models into technology developers can reliably build into products at scale.” The pitch echoes DCVC’s broader deep-tech thesis of underwriting infrastructure that gets cheaper and more measurable over time.
Riding The Composable-AI Wave
The round lands as investors chase infrastructure plays that turn generative models into reliable software components. TypeSafe joins recent AI-native raises such as Skild AI, Salesforce’s Koa reasoning model and Cornelis Networks, all racing to bring frontier-class intelligence into the software stack without the frontier price tag.
Reporting based on coverage from Business Wire, Yahoo Finance, Dealroom and Finsmes.