Anthropic Swallows Vercept in a Bet That AI's Next Frontier Is Learning to See for Desktop Agents

By
Tomorrow Capital
1 min read

The acquisition of a nine-person Seattle startup signals a broader war for the desktop—and raises hard questions about how long any edge in it will last.

On February 25, Anthropic declared it had acquired Vercept, the Seattle-based AI startup behind a desktop agent called Vy, folding its team into the company's effort to make Claude not just a conversationalist, but a worker—something that can open a spreadsheet, fill out a web form, navigate a browser tab, and do it all the way a human at a keyboard would.

Behind the press release, the deal is a story about perception—literal and strategic.


Vercept was founded by Kiana Ehsani, Luca Weihs, and Ross Girshick, all alumni of the Allen Institute for AI. Their thesis was deceptively simple: AI agents fail at computer tasks not because they lack intelligence, but because they can't adequately see. Most systems today parse the underlying DOM structure of a webpage—reading raw code rather than experiencing the interface the way a human does. Vy was different. It was vision-based, observing the screen as rendered pixels, identifying buttons and fields by sight rather than by scraping source code.

That distinction matters enormously in practice. Real-world software is messy. Enterprise applications, legacy tools, CAPTCHA walls, and dynamically rendered interfaces routinely defeat DOM-parsing agents. Vy was engineered to confront that messiness directly.

"VyUI's understanding of UI surpasses the models themselves," observed one technology analyst close to the deal. "That's the core value of this acquisition. Claude's desktop product has always been limited by how well it could read the screen. Vercept fills that gap directly."


Anthropic's own numbers underscore the urgency. When the company first released computer use capabilities in late 2024, Claude scored under 15% on OSWorld—a standard benchmark evaluating AI performance on complex computer tasks. With the launch of Claude Sonnet 4.6, that figure jumped to 72.5%, approaching what researchers consider human-level performance on tasks like navigating spreadsheets and completing multi-tab web forms. The Vercept team is now inside Anthropic, tasked with closing the remaining gap.

The deal follows Anthropic's acquisition of Bun, a coding agent, late last year. A pattern is forming: Anthropic is assembling the components of a fully autonomous digital worker, piece by deliberate piece.


The human drama around the deal is less tidy. Oren Etzioni—a co-founder of the Allen Institute and an early Vercept backer—went public with his frustration, calling the quick shutdown of Vy "throwing in the towel" for a company that had built genuine traction in just over a year. His criticism sparked a public exchange with lead investor Seth Bannon of A12, who led the $16 million seed round. Angels including Eric Schmidt, Jeff Dean, Kyle Vogt, and Arash Ferdowsi had also backed the startup, which raised approximately $50 million in total.

CEO Ehsani pushed back on Etzioni's framing. Joining Anthropic, she argued, wasn't retreat—it was acceleration. Vy will shut down entirely around March 25, with users directed to Claude as an alternative.

UiPath, whose robotic process automation business now faces a more capable AI-native competitor, saw its shares fall 3 to 4 percent following the announcement.


And yet, for all the strategic logic of the move, a sober question lingers at its edges.

The moat around vision-based UI understanding may be thinner than Anthropic is paying for. The technical approach—training AI to interpret rendered screens rather than underlying code—is not proprietary. Google, ByteDance, and Alibaba are each investing heavily in desktop agent capabilities, with resources that dwarf what a nine-person startup could marshal. The talent and the institutional knowledge Vercept carries are real assets. The technology itself, however, may face capable alternatives within months.

Anthropic has made a logical move. Whether it is a durable one depends on what the company builds next—and how fast the rest of the world's AI giants are moving down the same road.

Vy goes dark in thirty days. The race it was built to win is only beginning.

You May Also Like

This article is submitted by our user under the News Submission Rules and Guidelines. The cover photo is computer generated art for illustrative purposes only; not indicative of factual content. If you believe this article infringes upon copyright rights, please do not hesitate to report it by sending an email to us. Your vigilance and cooperation are invaluable in helping us maintain a respectful and legally compliant community.

Subscribe to our Newsletter

Get the latest in enterprise business and tech with exclusive peeks at our new offerings

We use cookies on our website to enable certain functions, to provide more relevant information to you and to optimize your experience on our website. Further information can be found in our Privacy Policy and our Terms of Service . Mandatory information can be found in the legal notice