Anthropic acquired the startup Vercept on February 25th. And this isn’t a footnote – it’s a statement.
What Vercept Built
Vercept is based in Seattle and focused on a problem many underestimate: How do you teach an AI to operate software the way a human does? Not through APIs. Not through code. But by seeing, clicking, and navigating.
Their product was called Vy – a cloud-based agent that could remotely control complete macOS environments. Filling out spreadsheets, switching between browser tabs, completing multi-step tasks in real applications. Sounds like science fiction? It already works.
Why Anthropic Is Making This Move
The timing is no coincidence. Claude has made massive progress in Computer Use recently – from under 15% to a 72.5% success rate on the OSWorld benchmark. That’s close to human level. But “close” isn’t enough when you want to build agents that work truly reliably.
The Vercept team – led by CEO Kiana Ehsani, Luca Weihs, and Ross Girshick – has spent years researching exactly these problems. They’re moving to Anthropic entirely. Not all founders, though: Oren Etzioni and Matt Deitke have already landed at Meta’s Superintelligence Lab.
The Numbers
Vercept had raised a total of $50 million in funding. Both sides are silent about the purchase price. What we know: Vy will be shut down in 30 days. The team will be fully integrated into Anthropic’s Computer Use research.
Why UiPath Is Getting Nervous
The market reacted immediately. UiPath – the RPA giant whose business model is built on software automation – saw its stock drop after the announcement. Understandable: if Claude learns to operate software directly, you might not need separate RPA tools anymore.
What This Means for You
Computer Use is the next major evolutionary step for AI assistants. Today, Claude writes code. Tomorrow, Claude operates your computer. And with the Vercept team on board, that “tomorrow” just got significantly closer.
It’s Anthropic’s second acquisition in three months – after Bun last year. A clear signal: Anthropic isn’t just building models, but assembling entire teams to bring Computer Use to scale.
Sources: Anthropic Blog, TechCrunch, GeekWire