TypeSafe AI exits stealth with $40M to build AI for use by software
SiliconANGLE Paul Gillin ● Covered by 4 sources
TypeSafe AI came out of stealth with $40M and a model built to live inside software. It aims to make AI output predictable, not chatty, which is the whole point.
Based on reporting by SiliconANGLE, Paul Gillin — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
TypeSafe AI has stepped out of stealth with $40 million in seed funding and a very specific pitch: stop using AI as a talker and start using it as a decision engine inside software. The San Francisco startup was founded in 2024 by Diogo Almeida, a former OpenAI researcher who worked on reinforcement learning from human feedback, InstructGPT, ChatGPT and GPT-4. DCVC led the round, and Forbes reported a $200 million valuation from someone familiar with the deal.
The company’s first model, Jev, is built to take structured questions and return structured answers. Think yes-or-no probabilities, list selections or scores on a set scale. Each result comes with probabilities and a confidence measure, so developers can decide when a workflow should move forward automatically, ask for more data or hand off to a person.
That is TypeSafe’s core bet: the very thing that makes large language models good at pleasing humans makes them awkward in production software. They can sound sure while being wrong, shift methods from one request to the next and force teams to add human review where consistency matters. Jev is meant to be a more predictable component, one that can sit alongside deterministic code instead of replacing it.
TypeSafe says the model uses a method it calls Reinforcement Learning for Calibrated Decisions and can generate hundreds of decisions in parallel from a single prompt. The company says that setup lets Jev return results in less than 100 milliseconds and makes it up to 100 times faster and less expensive than other frontier models. On its website, TypeSafe lists a cost of 39 cents per 1,000 workflows, versus $3.31 for OpenAI’s Gpt-5.6 Luna and R19.49 for Anthropic’s Claude Haiku 4.5, while TypeSafe’s own tests claimed Jev was nearly 194 times faster and about 445 times cheaper. Those figures haven’t been independently verified and will vary by workload, network location and comparison method.
The startup sees its biggest market in high-volume business work: service requests, invoices, security alerts and AI-agent outputs. In those cases, Jev could act as a control layer while a language model writes the prose. It’s a neat idea, and probably a more honest one than asking chatbots to cosplay as software infrastructure.
My take — AI-written commentary, not fact-checked reporting
This is the rare AI pitch that sounds like it has met production software before. The industry has spent years teaching models to be charming; now it is finally paying for one that can say “I’m 72% sure” and let the code do the rest. That’s not sexy, which is usually how you know it might be useful.
Read more about this at: SiliconANGLE