Prentis, a new artificial intelligence research laboratory focused on computer-use models, is in advanced talks to raise $100 million at a $1 billion valuation, according to two people familiar with the discussions. The startup was co-founded by serial entrepreneur Ritankar Das, LinkedIn co-founder Reid Hoffman, and Zynga founder Mark Pincus.
Launched in April, Prentis is training models to learn how office workers navigate routine workflows across documents and systems. The company's ultimate goal is to build AI agents that can take control of a computer to automate those tasks. The new funding round would give the company the resources to scale its model development and commercial deployment.
Prentis is part of a rapidly expanding field known as computer-use AI, where models interact with graphical user interfaces exactly as a human would: moving cursors, clicking buttons, typing text, and switching between applications. The startup believes that automating everyday office tasks will soon overtake coding as AI's biggest commercial use case. With a smaller, cheaper model, it claims to be far more economical to deploy across the long tail of enterprise workflows.
The company has already signed contracts worth up to $50 million with several customers, including a healthcare management service organization, a manufacturer, and goods and clothing manufacturers. Investor materials examined by reporters project an estimated $75 million annualized run rate by the third quarter of this year. The company's pitch deck cautioned that these figures reflect estimated annualized value based on a contracted fee equal to 20 percent of savings realized, not recognized revenue, and are performance-dependent and subject to final execution. That fee structure suggests Prentis is confident enough in its automation capabilities to tie its revenue directly to measurable customer outcomes.
Prentis says its proprietary model, Hive-32B, outperforms much larger and better-funded rivals on two industry-relevant benchmarks. On WindowsAgentArena, which measures end-to-end task completion in real Windows applications, and ScreenSpot-v2, which tests a model's ability to identify the correct on-screen control, Hive-32B reportedly surpasses OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6. The company further claims that its model achieves these results with roughly ten times lower cost per task than frontier API models, making it more practical for organizations that need to automate thousands of routine activities without incurring massive inference bills. These performance claims have not been independently verified by outside evaluators.
If accurate, the benchmark results suggest that specialized models trained for narrow enterprise workflows can compete effectively with general-purpose frontier systems while operating at a fraction of the cost. That could make AI-driven office automation accessible to mid-sized companies that have been priced out of the market for today's most advanced AI models. It also raises the possibility that the future of enterprise AI may not be dominated by giant all-purpose systems, but by lean, task-specific models that are optimized for actual business processes.
The market for computer-use agents is already crowded. Anthropic, OpenAI, and Mira Murati's Thinking Machines Lab are all working on similar capabilities, according to one source familiar with the industry. Anthropic has been particularly active in acquiring talent: it purchased the Seattle-based computer-use startup Vercept earlier this year, absorbed its founders, and shut down the startup's product. That move underscores the strategic importance these models have gained within major AI laboratories and the urgency they feel to dominate this emerging category.
For Prentis, the competitive landscape presents both pressure and validation. On one hand, the company must move quickly to secure enterprise contracts and prove its technology in real-world environments. On the other hand, the presence of deep-pocketed incumbents confirms that computer-use agents are expected to be a major source of revenue for the AI industry in the coming years. Prentis's focus on practical office workflows could give it an edge over larger labs that are still exploring the boundaries of general-purpose assistants.
Prentis is the latest venture from Ritankar Das, a 31-year-old entrepreneur who previously founded Titan, a holding company that builds and operates AI businesses. Das was UC Berkeley's youngest University Medalist in more than a century, having graduated at 18 with a double major in bioengineering and chemical biology. He went on to earn a master's in biomedical engineering at Oxford before beginning a PhD in artificial intelligence at Cambridge as a Gates Cambridge Scholar. He left the PhD program in 2014 to found Titan.
Das has characterized Titan as an intentional throwback to the old-fashioned holding-company model exemplified by Berkshire Hathaway. Rather than relying on outside limited partners for funding, Titan has been financed by the proceeds of its own exits. One of Titan's portfolio companies, disease prediction startup Dascena, was acquired by CirrusDx in 2022. Other Titan-launched businesses include Tala Health, an AI-powered virtual care provider that raised a $100 million seed round last year, and Forta Health, an autism care startup that raised $55 million led by Insight Partners in 2024. These ventures reveal Das's pattern of building AI-first companies in regulated and high-value verticals.
For Reid Hoffman and Mark Pincus, Prentis is something of a side project. Hoffman, a LinkedIn co-founder and Greylock partner, said last month that he was stepping down from Microsoft's board after nearly a decade to go into founder mode on Manas AI, an AI drug-discovery startup he is also backing. Hoffman was an early investor in OpenAI and co-founded Inflection AI with Mustafa Suleyman before Microsoft absorbed most of that team in 2024. His history of identifying major technology shifts lends credibility to Prentis's thesis that computer-use agents will become a core enterprise tool.
Pincus, best known for founding Zynga, currently runs the investment firm Reinvent Capital, where Hoffman serves as senior adviser. Pincus also published a memoir, Life at the Speed of Play, last month. His background in building consumer platforms that scaled to hundreds of millions of users could help Prentis design its products around clean user experiences and easy onboarding, even in enterprise environments.
According to its website, Prentis has already hired more than 25 employees, including researchers who previously worked at OpenAI, Google DeepMind, Meta, Tencent, and Alibaba. The level of talent the startup has attracted is notable for a company that has remained relatively quiet since its founding. It suggests the founders' reputations and the promise of long-term equity upside have been enough to pull senior researchers away from some of the most prestigious AI laboratories in the world.
The broader implications of computer-use models are significant. If Prentis and its competitors succeed, many of the mundane tasks that consume office workers' days—processing invoices, retrieving records, filling out forms, reconciling accounts, and updating databases—could be largely automated. That transformation would not only create cost savings but would also reshape the nature of white-collar work. Many jobs could shift toward supervising AI agents, handling exceptions, and focusing on higher-level judgment rather than repetitive data entry.
Companies that deploy these systems are being asked to trust AI with sensitive internal data and system-level access. Prentis's decision to charge a fee based on 20 percent of realized savings aligns its incentives with customer outcomes, but it also means the company only earns money when demonstrable automation occurs. That performance-based pricing model could become a template for other enterprise AI vendors, especially those seeking to convince skeptical finance departments that generative AI delivers measurable return on investment.
The claimed $75 million annualized run rate by the third quarter is striking for a company that only launched in April. Even with the caveats attached to the figures, it suggests early enterprise demand for AI agents is robust. If the new funding round closes successfully, Prentis will need to move quickly to expand its model capabilities, hire more researchers, and grow its sales pipeline beyond the small number of initial customers. The capital would also allow the company to invest in the infrastructure needed to process large volumes of enterprise data and to serve customers across different industries.
The involvement of experienced founders and investors gives Prentis a distinct advantage in navigating the complex path from research prototype to sustainable company. Hoffman's record of spotting platform shifts, Pincus's experience building viral consumer networks, and Das's scientific and operational track record create a leadership team capable of addressing both product design and business execution. Few early-stage AI startups can point to a founder lineup with as many notable successes.
At the same time, the competitive environment remains intense. Major AI labs have enormous compute resources, established enterprise sales channels, and deep relationships with the world's largest companies. Anthropic's acquisition of Vercept is evidence that even the biggest players are preparing for a long fight in the computer-use space. Prentis's lean, specialized approach may serve as an effective counterweight. By focusing on specific workflows and optimizing for cost, it can serve customers that need reliable results without the expense of massive general-purpose models.
Observers are also waiting to see whether Prentis can maintain its benchmark performance as it scales to more diverse environments. Small, specialized models sometimes excel on narrow test sets but struggle when confronted with the chaos of real-world software versions, unusual interface layouts, and unpredictable user behavior. Prentis's defense is that its model is specifically designed for the common, repetitive tasks found in most offices, where the range of actions is limited and fairly predictable. That focus may allow the model to maintain high accuracy in production settings.
The fundraising talks remain ongoing, and final terms could change. A $1 billion valuation after less than a year of operation would be a remarkable achievement, reflecting both the strength of the founding team and the intense investor appetite for AI companies that promise to automate real work. The ultimate test will be whether Prentis can convert its early contracts and benchmark claims into a durable business. That will depend on its ability to deliver consistent savings to customers, fend off competition from much larger labs, and keep its technology ahead of an industry moving at breakneck speed.
Source: TechCrunch News