The $1 Billion Startup Teaching AI to Do Your Desk Job
Prentis, a four-month-old lab backed by LinkedIn co-founder Reid Hoffman, is in talks to raise $100 million. It claims its AI can handle insurance claims and customs paperwork cheaper than any rival. Here is what ordinary workers should know.
Key points
- Prentis, an AI startup launched in April 2025, is in talks to raise $100 million at a $1 billion valuation.
- The company has signed contracts worth up to $50 million with customers in healthcare management, clothing retail and manufacturing.
- Its pitch deck projects an annualised revenue run rate of $75 million by Q3 2025, though those figures are performance-dependent, not recognised revenue.
- Prentis claims its Hive-32B model beats OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6 on two computer-use benchmarks; TechCrunch, which broke the story, has not independently verified those results.
- Co-founders include LinkedIn co-founder Reid Hoffman and Zynga founder Marc Pincus alongside CEO Ritankar Das, 31.
A four-month-old AI startup wants to sit at your desk and handle the repetitive bits of your job. Prentis, first reported by TechCrunch, is training what it calls computer-use models: AI that watches a screen and operates software the same way a person would, minus the frustration.
What does Prentis actually do?
It builds AI agents, meaning software that carries out multi-step tasks without human prompting, aimed at white-collar workflows. Processing an insurance claim, chasing a customs duty refund, pulling data from several disconnected systems without a person hunting through paperwork: those are the targets.
Prentis charges a fee equal to 20 percent of whatever savings the AI delivers. No savings, no bill. That model has already produced contracts worth up to $50 million with clients in healthcare management, clothing retail and manufacturing, according to people familiar with the discussions.
The pitch deck forecasts a $75 million annualised run rate by Q3 2025. Those are estimated figures tied to contracted fees, not money already collected, and the company's own small print flags they depend on the AI actually performing.
Who is behind it, and where is the money coming from?
CEO Ritankar Das, 31, graduated UC Berkeley at 18 with a double major in bioengineering and chemical biology, then earned a master's at Oxford before dropping out of a Cambridge AI PhD programme to found Titan, a holding company that builds AI businesses using profits from its own exits rather than outside investors.
Titan's portfolio includes Tala Health, a virtual care provider that raised a $100 million seed round in 2024, and Forta Health, an autism-care startup that raised $55 million from Insight Partners the same year. Titan-founded disease-prediction company Dascena was acquired by CirrusDx in 2022.
For Hoffman and Pincus, Prentis is a side project. Hoffman recently stepped back from Microsoft's board to focus on Manas AI, a drug-discovery startup; he also co-founded Inflection AI before Microsoft absorbed most of that team in 2024. Pincus runs Reinvent Capital.
Should office workers be worried?
Possibly, but not yet. The market's crowded. Anthropic, Mira Murati's Thinking Machines Lab and OpenAI are all building competing computer-use agents. Anthropic moved directly into the category by acquiring Seattle startup Vercept earlier this year and folding its founders into its own team.
Prentis claims its Hive-32B model, a 32-billion-parameter AI (the number roughly indicates how much information the model has absorbed), beats GPT-5.4 and Claude Opus 4.6 on two standard computer-use benchmarks. It also claims roughly ten times lower cost per task than those rivals. Neither claim has been independently verified. We looked at the job-displacement evidence more closely in our 23 July piece on what the numbers actually show about AI and employment.
Survivorship bias is real here. Prentis is four months old, its revenue figures are projections, and the workers it's targeting were promised similar automation a decade ago by robotic process automation tools that never quite delivered.
If your job involves repeating the same clicks across the same software systems every day, watch this space carefully. The tools are genuinely getting better. Start documenting exactly what you do and why: the humans who understand a workflow deeply will be the ones managing the AI doing it, not the ones replaced by it.
