Prentis AI Lab: Hoffman and Pincus-backed startup targets enterprise automation with AI agents

Prentis, a new AI research lab focused on computer-use models, is positioning itself as a contender in enterprise workflow automation. Co-founded by serial entrepreneur Ritankar Das, LinkedIn co-founder Reid Hoffman, and Zynga founder Mark Pincus, the lab is in talks to raise $100 million at a $1 billion valuation, according to TechCrunch. Prentis aims to build AI agents that learn how office workers navigate documents and systems, then automate those tasks.

What happened

Launched in April, Prentis claims contracts worth up to $50 million with several customers, including a healthcare management organization and manufacturers. The startup projects an estimated $75 million annualized run rate by Q3, based on a fee equal to 20% of savings realized. However, the pitch deck notes these figures are "performance-dependent and subject to final execution."

Prentis also claims its Hive-32B model outperforms OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6 on two computer-use benchmarks: WindowsAgentArena (end-to-end task completion on Windows apps) and ScreenSpot-v2 (on-screen control location). The company argues its edge comes from running a smaller, cheaper model, claiming roughly 10 times lower cost per task than frontier APIs. TechCrunch has not independently verified these benchmark results.

Why AI builders should care

The story signals a growing market push toward computer-use AI agents as a major use case beyond coding. Prentis is entering a crowded field that includes Anthropic, OpenAI, and Mira Murati's Thinking Machines Lab. For builders, the key takeaway is the emphasis on cost-per-task efficiency. If validated, Prentis's approach could influence how teams deploy AI agents in enterprise IT environments, favoring smaller models optimized for specific workflows over general-purpose frontier models.

Practical implications

Prentis's model suggests a shift toward task-specific, lower-cost inference for office automation. Builders evaluating computer-use agents should watch for independent benchmarks and real-world deployment results. The 20% savings-fee model is an unusual monetization approach that ties revenue directly to customer outcomes, which could align incentives but also introduces risk if savings are hard to measure.

Caveats

Many of Prentis's claims rely on investor materials and have not been independently verified. The $50 million in contracts and $75 million run rate are based on estimated annualized value, not recognized revenue, and are subject to final execution. Benchmark comparisons against GPT-5.4 and Claude Opus 4.6 are self-reported. The company has not responded to requests for comment. Builders should treat these figures as directional until third-party validation emerges.