Reid Hoffman Backs Prentis, an AI Lab Chasing a $1B Valuation

A new AI research lab called Prentis — co-founded by Reid Hoffman, Mark Pincus, and serial entrepreneur Ritankar Das — is reportedly in talks to raise $100 million at a $1 billion valuation. The company launched in April and is focused on a specific, unglamorous problem: getting AI to handle the repetitive computer tasks that fill most office workers' days.

What Prentis Actually Does

Rather than chasing general-purpose AI or coding assistants, Prentis is training models to watch how people navigate real workflows — clicking through documents, switching between systems, filling out forms — and then automating those steps. The practical targets are things like processing insurance claims or sorting through customs duty paperwork, tasks that currently require a human to spend time hunting down information across multiple tools.

The lab's flagship model, Hive-32B, is a relatively compact 32-billion-parameter system. According to investor materials obtained by TechCrunch, Prentis claims Hive-32B outperforms OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6 on two computer-use benchmarks — WindowsAgentArena, which tests task completion on live Windows applications, and ScreenSpot-v2, which measures whether a model can correctly identify on-screen interface elements. TechCrunch notes it hasn't independently verified those claims.

The cost angle is the other part of the pitch. Prentis says running Hive-32B costs roughly ten times less per task than frontier-model APIs, which matters a lot if you're trying to deploy AI across thousands of routine workflows rather than occasional high-stakes queries.

Early Revenue and Some Caveats

The company says it has signed contracts worth up to $50 million with several customers, including a healthcare management organization and a pair of manufacturers. Its pitch deck projects an annualized run rate of around $75 million by Q3 of this year — though that figure is based on a contracted fee equal to 20% of realized savings, not recognized revenue, and is described as "performance-dependent and subject to final execution." In other words, those numbers reflect what Prentis expects to earn if the technology actually delivers the efficiency gains it promises.

A Crowded Market

Prentis is entering a space that's getting competitive fast. Anthropic and OpenAI are both building computer-use agents of their own, and Mira Murati's Thinking Machines Lab is working in the same direction. Anthropic moved aggressively earlier this year by acquiring Vercept, a Seattle startup focused on computer use, folding in its founders and shutting down the product. It's worth noting that Reid Hoffman has previously been skeptical about some AI ventures, so his decision to back this one is a signal he sees a real market here.

The underlying bet — that automating everyday office tasks will eventually be a bigger AI use case than writing code — isn't unreasonable, but it's far from proven. Karpathy has made a similar argument about AI shifting from a tool you query to one that works alongside you, which aligns with what Prentis is building toward.

The People Behind It

The CEO, Ritankar Das, is 31 and has an unusual background: UC Berkeley's youngest University Medalist in over a century, Oxford master's degree, and a Gates Cambridge Scholar who dropped out of an AI PhD program at Cambridge to start Titan, a holding company that builds and operates AI businesses. Titan's portfolio includes Tala Health (a $100M seed round last year) and Forta Health (a $55M raise in 2024). Das runs Prentis out of that same structure.

Hoffman and Pincus are co-founders in a more advisory capacity. Hoffman has been in full founder mode lately, recently stepping off Microsoft's board to focus on Manas AI, an AI drug-discovery startup. Pincus runs Reinvent Capital, with Hoffman as a senior adviser.

Prentis has already hired more than 25 people, including researchers from OpenAI, Google DeepMind, Meta, Tencent, and Alibaba. Whether a leaner, cheaper model focused on mundane workflows can carve out space against better-funded competitors is the real question — but the early contract numbers suggest there's at least some demand to test that theory.