Reid Hoffman and Mark Pincus — the minds behind LinkedIn and Zynga respectively — are backing a new AI research lab called Prentis that wants to teach machines to do the thing most office workers spend most of their day doing: clicking through documents, filling forms, and navigating software systems that nobody fully understands.
Prentis, co-founded with serial entrepreneur Ritankar Das, is in talks to raise $100 million at a $1 billion valuation, according to TechCrunch’s reporting. The startup has already signed contracts worth up to $50 million with healthcare, manufacturing, and clothing companies, and its investor materials project $75 million in annualised run rate by Q3.
🔍 THE BOTTOM LINE
The pitch is simple: coding was AI’s first big use case, but the bigger market is the billions of office workers navigating spreadsheets, CRM systems, and customs forms every day. Prentis is building a 32-billion-parameter model called Hive-32B that claims to outperform OpenAI’s GPT-5.4 and Anthropic’s Claude Opus 4.6 on two computer-use benchmarks — at roughly one-tenth the cost per task. If the benchmarks hold, it’s a serious shot across the bow of every frontier lab racing to control the desktop.
What Prentis Actually Does
Launched in April 2026, Prentis trains models to learn how office workers navigate routine workflows across documents and systems. The goal is AI agents that can control computers to automate those tasks — handling insurance claims, processing customs duty refund exceptions, and eliminating the paperwork hunt that consumes hours of white-collar labour.
The startup’s pitch deck claims its Hive-32B model outperforms rivals on WindowsAgentArena, which measures end-to-end task completion on real Windows applications, and ScreenSpot-v2, which tests a model’s ability to locate the right on-screen control. Prentis says its edge comes from running a much smaller, cheaper model — roughly 10 times lower cost per task than frontier APIs.
TechCrunch hasn’t independently verified those benchmark results, and the startup’s own pitch deck notes that its revenue figures are “performance-dependent and subject to final execution.” But the model size alone — 32B parameters versus the multi-trillion-parameter frontier models — tells you what Prentis is betting on: that specialised, smaller models can win the office-automation race by being cheap enough to deploy across everyday workflows.
The Founders Behind It
This is a lab with serious pedigree. Ritankar Das, the CEO, is 31 and was UC Berkeley’s youngest University Medalist in more than a century, graduating at 18 with a double major in bioengineering and chemical biology before earning a master’s at Oxford as a Gates Cambridge Scholar. He dropped out of an AI PhD program at Cambridge to found Titan, a holding company that builds and operates AI companies — including Tala Health, which raised a $100 million seed round last year, and Forta Health, an autism care startup that raised $55 million from Insight Partners.
Hoffman needs little introduction. The LinkedIn co-founder and Greylock partner was an early OpenAI investor and co-founded Inflection AI with Mustafa Suleyman before Microsoft absorbed most of that team in 2024. He stepped down from Microsoft’s board last month to go “founder mode” on Manas AI, a drug-discovery startup. Pincus, the Zynga founder, now runs Reinvent Capital with Hoffman as a senior adviser.
Why Computer Use Is the Most Crowded Race in AI
Prentis is entering one of the most competitive spaces in AI. Anthropic, OpenAI, and Mira Murati’s Thinking Machines Lab are all building computer-use agents. Anthropic has also been acquiring talent directly — it bought the Seattle computer-use startup Vercept earlier this year, folding in its founders and shutting down its product.
Google embedded computer control directly into Gemini 3.5 Flash in June, letting developers build agents that see, click, and type across browsers and desktops — a move we covered in our earlier reporting on Gemini’s computer-use launch. Perplexity launched “Computer” as a multi-model agentic platform. The pattern is clear: every frontier lab sees the office desktop as the next platform war.
The reason is simple economics. Coding is a $200 billion market. Office automation — the people navigating documents, systems, and processes across every industry — is measured in trillions. Prentis’s pitch that automating routine computer tasks will “outpace coding as AI’s biggest use case” is the same thesis driving every major lab’s computer-use push.
The Smaller-Model Gambit
Prentis’s contrarian bet is on model size. While OpenAI, Anthropic, and Google are racing toward ever-larger frontier models, Prentis is running a 32B-parameter model and claiming it beats the giants on specific computer-use tasks. The claim is that for the narrow domain of navigating office software, a specialised smaller model is more economical to deploy — 10x cheaper per task — and good enough.
This echoes the broader industry debate about whether specialised models will eat the general-purpose frontier models’ lunch in specific verticals. We’ve seen this play out in coding (where smaller coding-focused models compete with frontier models on cost-per-token) and in enterprise AI (where fine-tuned domain models outperform generalists on specific tasks). Computer use may be the next vertical where this dynamic plays out.
Prentis has already hired more than 25 employees, including researchers who previously worked at OpenAI, Google DeepMind, Meta, Tencent, and Alibaba. The talent density is real. The question is whether a 32B model can actually hold its own against GPT-5.4 and Claude Opus 4.6 on the messy, unpredictable reality of real office workflows — not just benchmarks.
NZ Angle
For New Zealand’s economy — where a large proportion of businesses are SMEs dealing with exactly the kind of document-heavy, system-hopping work Prentis is targeting — computer-use agents could be transformative. Customs processing, IRD filings, ACC claims, and the endless inter-agency paperwork that dominates Kiwi office life are precisely the workflows Prentis is building for. If smaller, cheaper models like Hive-32B deliver on the cost-per-task promise, the barrier to adoption drops significantly for companies that can’t afford frontier API pricing at scale.
The broader implication for NZ’s tech workforce is the same one we’ve been tracking: if AI agents can navigate office systems autonomously, the administrative roles that dominate white-collar employment face real pressure. We covered this dynamic in our reporting on AI-driven layoffs and the pattern is accelerating, not slowing.
❓ FAQ
Is Prentis’s $1 billion valuation justified? The valuation is based on $50M in signed contracts and a projected $75M annualised run rate by Q3 — though those figures reflect estimated annualised value based on a contracted fee equal to 20% of savings realised, not recognised revenue. The numbers are promising but performance-dependent.
How does Hive-32B compare to frontier models? Prentis claims it outperforms GPT-5.4 and Claude Opus 4.6 on WindowsAgentArena and ScreenSpot-v2 benchmarks at roughly 10x lower cost per task. These claims haven’t been independently verified.
What does Reid Hoffman’s involvement mean? Hoffman was an early OpenAI investor and co-founded Inflection AI. His involvement signals that serious tech money sees computer-use agents as the next major AI frontier — and that the smaller-model approach is worth backing.
When will Prentis’s products be available? The startup is already operating with signed customers in healthcare, manufacturing, and clothing. No general availability timeline has been announced — it’s enterprise-focused for now.
🔍 THE BOTTOM LINE
Prentis is entering the most crowded race in AI with a contrarian bet: that a 32B model can beat trillion-parameter frontier models on office automation at a tenth of the cost. The pedigree of its founders and the $50M in signed contracts suggest it’s not just another startup. But the computer-use space is where Anthropic, OpenAI, Google, and Thinking Machines Lab are all converging — and the winner won’t be decided by benchmarks, but by who can actually make agents work reliably in the messy, unpredictable world of real office software.
📰 Sources
- TechCrunch — Prentis, new AI lab co-founded by Reid Hoffman, Mark Pincus in talks to raise $100M
- Prentis (prentis.ai)
- Fortune — Tala Health raises $100 million seed round
- Anthropic — Vercept acquisition
