Ivan Lee is the co-founder and CEO of Datasaur, which builds private AI for enterprise: open weights models deployed inside a company's own servers, so sensitive data never leaves. He studied computer science, trained as an engineer and has been a product manager his whole career, building AI products at Yahoo and at Apple, where he worked on Siri. Datasaur is his second startup, and he has been at it for eight years. He joined us in Episode 56 (October 23, 2026).
Datasaur sets up AI inside the client's own infrastructure. Ivan said none of the sensitive data, whether clinical patient data or Social Security numbers, is sent to Anthropic or OpenAI (02:12). It deploys open weights models straight onto the servers, together with the context, tooling and everything else the agents need to work.
Most clients are in healthcare, finance, legal and the public sector. To the client, an engagement feels like white glove consulting, a model Ivan called "AI native services" (15:14). The client describes a workflow, and Datasaur recommends a model and explains how much work it will take. Sometimes that is as simple as connecting the model to an internal Dropbox or OneDrive.
Behind that sits what Ivan called a proprietary model factory (16:01). It takes the requirements, workflow, benchmarks and historic data, suggests the best model, and handles fine tuning and the engineering around the model. It is built privacy first.
Pair that alongside the fact that there's been a whole new wave of open weight models this year, and suddenly, there is both motive as well as opportunity to build your own AI and turn that into a strategic asset.
— Ivan Lee, Episode 56
We could pause all model development now for the next decade, and we would still be gathering a lot of improvements across the enterprise just using these open weights models alone.
— Ivan Lee, Episode 56
Jordan asked about the Jevons paradox. Ivan's answer was token math (11:11). Agents use far more tokens, which he linked to the GPU glut earlier this year, when Anthropic had to slow down its services.
Your average chatbot query is 200 to 2,000 tokens. Your average agentic task is 3,000,000 tokens.
— Ivan Lee, Episode 56
In the news segment (17:18), we covered Meta Muse passing 10 million users and Anthropic's Duet. Ivan, who worked on Siri, likes personal agents. Still, he thinks the enterprise wins first. Clients already have operational playbooks written to train the next human employee, while consumers all want different airports and seats.
I think agents are just universally a better fit for the enterprise than they are for the consumer world at this point.
— Ivan Lee, Episode 56
His own favorite build (22:05) is an MCP server for his Whoop bracelet, connected to Claude. It knows how he slept, tracks his activity and gives him a training plan each day: gym, run or rest day.
For more on the hardware and token costs behind running models, see our AI infrastructure page, plus AI in healthcare and AI in finance.
Ivan Lee is the co-founder and CEO of Datasaur. He studied computer science, trained as an engineer and worked as a product manager his whole career, building AI products at Yahoo and Apple, where he worked on Siri. Datasaur is his second startup, and he has been working on it for eight years.
Datasaur builds private AI for enterprise. It deploys open weights models onto a client's own servers, along with the context and tooling the agents need, so data like clinical patient records or Social Security numbers is never sent to Anthropic or OpenAI.
Ivan said Datasaur largely works with healthcare, finance, legal and public sector organizations, the teams that care most about privacy and security.
Ivan said yes for most work. On his benchmark slide, every open weights model cleared Claude 4.5, the point where he says agents started working, and he called frontier models overkill for tasks like summarizing emails or recapping meeting notes.
Ivan put the average chatbot query at 200 to 2,000 tokens and the average agentic task at 3,000,000 tokens, over three orders of magnitude more.
