Prentis, AI Lab Backed by Hoffman and Pincus, Seeks $100M at $1B Valuation

Prentis, a newly launched AI research lab focused on computer-use models, is in talks to raise $100 million at a $1 billion valuation as it races to commercialize AI agents that can operate computers the way office workers do. The startup, co-founded by Reid Hoffman, Marc Pincus and serial entrepreneur Ritankar Das, says it already has up to $50 million in contracted business and is targeting workflow automation in industries such as healthcare, manufacturing and trade compliance.

The fundraising talks underscore how quickly investor attention has shifted toward a fast-emerging category of AI systems designed not just to generate text, but to click, type and move through real software environments. If Prentis can validate its claims, it could become one of the more prominent players in a crowded battle to build practical AI agents for everyday business tasks.

What is Prentis and why does it matter?

Prentis is an early-stage AI lab building models that learn how people use computers in real work settings, with the goal of automating repetitive office workflows end to end. The company’s pitch is straightforward: instead of relying on humans to collect files, switch between software systems and complete routine administrative steps, its agents would do the work directly.

That ambition matters because computer-use AI has become one of the most closely watched frontiers in the industry. Major labs are pursuing the same prize: software that can interact with interfaces, complete tasks and reduce the need for human intervention in back-office operations. The category could eventually touch insurance, logistics, customer operations, healthcare administration and many other sectors where workers spend hours on repetitive digital processes.

How much is Prentis trying to raise?

Prentis is reportedly seeking $100 million in new funding at a $1 billion valuation, according to people familiar with the discussions. The proposed round would place the company among the better-funded newcomers in the computer-use AI space, especially for one that only launched in April.

Those same people said Prentis has already signed contracts worth as much as $50 million with several customers. Investor materials seen by TechCrunch suggest the company expects to reach an annualized run rate of roughly $75 million by the third quarter of this year, although the deck also notes that the numbers are performance-based, subject to execution and tied to savings rather than booked revenue.

How does Prentis say its technology works?

Prentis says it trains models on how office workers navigate routine digital tasks across documents, browser windows and enterprise systems. The aim is to build agents that can observe a workflow, identify the necessary steps and carry them out inside the same software tools humans already use.

According to the company’s materials, that could include handling insurance claims, resolving customs duty refund exceptions and managing other paperwork-heavy processes that typically require employees to search across multiple systems and manually verify information.

The company’s central argument is that this approach can cut labor time and reduce costs in settings where the work is repetitive but still too messy for traditional automation. Rather than hard-coding narrow rules for each process, Prentis is betting on models that can generalize across interfaces and adapt to new tasks.

Why focus on computer use instead of coding?

Prentis believes automating ordinary office tasks will become a larger opportunity than coding assistants. That view reflects a growing conviction inside parts of the AI industry that the biggest near-term gains may come not from helping engineers write software faster, but from replacing the large volume of repetitive administrative work that still runs on spreadsheets, portals and legacy enterprise applications.

In practice, that means the company is targeting the messy middle of modern business operations: work that is digital, repetitive and valuable, but not standardized enough to be automated by simple scripts. If successful, that market could be enormous.

What does Prentis claim about model performance?

Prentis says its Hive-32B model outperforms leading rivals, including OpenAI’s GPT-5.4 and Anthropic’s Claude Opus 4.6, on two computer-use benchmarks: WindowsAgentArena and ScreenSpot-v2. The first measures how well a model can complete tasks inside real Windows applications; the second evaluates whether it can correctly locate on-screen controls.

The company also argues that its system is far cheaper to operate than frontier models because it relies on a much smaller architecture. In its pitch, Prentis claims the cost per task is about 10 times lower than the large frontier APIs being used by competitors. If true, that would be a major advantage for enterprise deployment, where cost efficiency often determines whether a tool scales beyond pilots.

TechCrunch said it has not independently verified the benchmark claims, and that caveat matters. In a young market like computer-use AI, benchmark wins can help drive attention and fundraising, but real-world reliability, security and latency often matter more to enterprise buyers than a single leaderboard result.

Prentis’s own materials frame its edge as a combination of smaller model size and lower task cost, with the company arguing that this makes the product more practical for everyday workflows than more expensive frontier systems.

Why is the market getting crowded so fast?

The short answer is that nearly every major AI lab now sees computer-use agents as a strategic prize. These systems promise to bridge the gap between a model that can talk intelligently and one that can actually perform work inside software applications.

That overlap has drawn intense competition from some of the biggest names in the field. Anthropic, OpenAI and Mira Murati’s Thinking Machines are also building for computer-use applications, according to one source familiar with the landscape. Anthropic has already made a direct move in the category by acquiring Vercept, a Seattle startup that worked on computer-use technology, then integrating the team and shutting down its standalone product.

For Prentis, the competitive pressure means that product quality alone may not be enough. To win, it will need to prove it can deliver reliable automation, integrate into enterprise systems, and keep operating costs low enough to make deployment broadly attractive.

Who is behind Prentis?

Prentis is led by Ritankar Das, a young founder with an unusually broad track record across research, startups and venture-backed health technology. Das is also the founder of Titan, a holding company that builds and runs AI businesses.

Das, now 31, was once UC Berkeley’s youngest University Medalist in more than a century. He graduated at 18 with dual degrees in bioengineering and chemical biology, then went on to earn a master’s degree in biomedical engineering at Oxford. He later began an AI PhD program at Cambridge, where he was a Gates Cambridge Scholar, before leaving to start Titan in 2014.

Das has described Titan as a modern version of an old-fashioned holding company, one that funds itself through exits from its own businesses rather than depending on traditional outside limited partners. The model has already produced a small portfolio of healthcare-focused AI and technology companies.

What else has Titan built?

Titan’s portfolio provides context for the kind of company-building Das favors. Its businesses have included:

• Tala Health, an AI-powered virtual care provider that raised a $100 million seed round last year;
• Forta Health, an autism care startup that raised $55 million in 2024 led by Insight Partners;
• Dascena, a disease-prediction company acquired by CirrusDx in 2022.

That history suggests Prentis is not a random side bet, but part of a broader pattern: building data-intensive software businesses around healthcare and operational automation, then scaling them with a mix of technical ambition and commercial discipline.

What role are Reid Hoffman and Marc Pincus playing?

Hoffman and Pincus bring signaling power, operating experience and investor credibility to the project. Both are well-known Silicon Valley figures with deep ties to startup formation, product scaling and venture capital networks.

Hoffman, the LinkedIn co-founder and a partner at Greylock, recently said he was stepping away from Microsoft’s board after nearly a decade to focus more fully on founder work through Manas AI, a drug-discovery startup he is backing. He was also an early OpenAI investor and previously co-founded Inflection AI with Mustafa Suleyman before Microsoft absorbed much of that team in 2024.

Pincus, who founded Zynga, now runs the investment firm Reinvent Capital and has Hoffman serving as a senior adviser. He also published a memoir last month, adding another layer to his high-profile return to the startup spotlight.

For Prentis, the presence of Hoffman and Pincus likely helps on two fronts: fundraising and customer confidence. In a market where buyers may still be uncertain about how to deploy autonomous software safely, well-known backers can speed early trust.

Hoffman has recently described himself as shifting back into a more hands-on founder posture, while Pincus has moved into investing and advisory work after building Zynga into a major consumer-tech company.

What customers is Prentis targeting?

Prentis appears to be aiming squarely at enterprises with large volumes of repetitive administrative work, especially in sectors where workflow errors are expensive and labor-intensive. The examples in the source material point to healthcare administration, manufacturing operations and customs-related paperwork.

That focus makes strategic sense. These are places where a small increase in automation can produce meaningful savings, especially if the AI can handle exceptions instead of just standard cases. The startup’s pitch suggests it wants to sell not as a general-purpose chatbot, but as a highly specialized digital worker.

In theory, the customer value proposition is easy to understand:

1. Reduce manual data entry and repetitive navigation between systems.
2. Lower the labor cost of repetitive exceptions handling.
3. Improve throughput without forcing companies to replace their existing software stack.

In reality, adoption will depend on whether the product can handle failure cases reliably and whether enterprises trust the system enough to let it act without constant human oversight.

Why the valuation is drawing attention

A $1 billion valuation for a company launched only months ago is notable even by AI standards, but it reflects a broader market trend. Investors are paying up for startups that show even early signs of traction in high-stakes AI infrastructure and agent software, particularly when those startups are backed by prominent names.

Prentis also benefits from the narrative shape of the current AI wave. The industry has moved from models that answer questions to models that can call tools, browse interfaces and perform actions. That shift has created a new crop of companies trying to become the “operating layer” for AI-driven work.

If Prentis can validate its benchmark claims and convert performance-based contracts into durable revenue, a premium valuation may appear more justified. If not, the company could face the same challenge many AI startups do: impressive demos and investor buzz that prove harder to translate into stable enterprise adoption.

How does Prentis fit into the broader AI agent race?

Prentis fits into one of the most important transitions in artificial intelligence: the shift from assistance to action. Instead of simply generating text or summarizing data, agent systems are designed to carry out steps in the software stack and complete tasks with limited human input.

That transition could reshape productivity software, business process outsourcing and enterprise automation. It could also raise new questions about auditability, permissioning, error recovery and security. A system that can act inside enterprise software is more powerful than a chatbot, but also riskier if it makes mistakes or is manipulated.

Companies competing in this space will need to answer several practical questions:

• Can the AI work across different interfaces without custom retraining?
• Can enterprises monitor and approve its actions?
• Can the system handle edge cases, log decisions and recover from errors?
• Can it do all of that at a cost that makes broad deployment worthwhile?

Prentis says yes on cost and capability. The market will now decide whether the claims hold up outside a pitch deck.

Timeline of the company’s rapid rise

Prentis has moved unusually fast from launch to funding talks, customer contracts and public comparison with leading AI labs. The pace reflects both the urgency of the market and the fundraising momentum surrounding agentic AI.

What happens next?

The near-term question is whether Prentis can close its funding round and prove that its early customer contracts turn into meaningful revenue. If the company can show that its agents really do save time and reduce operating costs, the startup could become a serious competitor in enterprise automation.

But the bar is high. Computer-use AI is still an emerging category, and the largest incumbents in the field have far more resources, broader model research and existing enterprise relationships. That means Prentis will need more than a good benchmark result. It will need reliability, defensible economics and enough product specificity to stand out in a crowded race.

For now, the story is less about one startup alone and more about where AI is heading next. The industry’s focus has shifted from answering questions to doing work. Prentis is one of the latest companies trying to turn that idea into a business.