Anthropic isn’t only coming out with a flurry of releases of plugins and products, but it’s also acquiring companies to quickly grow its capabilities.
Anthropic announced today that it has acquired Vercept.ai, a Seattle-based startup specializing in vision-based computer automation. The deal brings Vercept’s nine-person team — including co-founders Kiana Ehsani, Luca Weihs, and Ross Girshick — under Anthropic’s roof to accelerate the development of Claude’s computer use capabilities. Vercept will wind down its external product in the coming weeks as the team integrates into Anthropic.
What Vercept Brings to the Table
Vercept was founded on a straightforward but technically demanding premise: making AI truly useful for complex tasks requires solving hard problems in how machines perceive and interact with software. The team, which includes alumni from the Allen Institute for AI (Ai2), built Vy — a Mac application that lets users control their computers through natural language commands. Rather than relying on APIs, connectors, or pre-written scripts like traditional robotic process automation (RPA) tools such as UiPath, Vy used computer vision AI to “see” and interpret screens the same way a human would, enabling flexible automation of everything from invoice management to form filling with a single command.
The startup attracted serious attention from the tech investment community. In June 2025, Vercept closed a $16 million seed round led by Fifty Years, with participation from a notable roster of backers including former Google CEO Eric Schmidt, Google DeepMind’s Jeff Dean, Dropbox co-founder Arash Ferdowsi, and Cruise founder Kyle Vogt. The company had also demonstrated impressive technical results, claiming 92% accuracy on computer automation benchmarks in partnership with Together AI — a striking figure compared to OpenAI’s reported 18.3% on similar tasks — alongside 5x performance gains and 30% cost savings.
Powering the Next Generation of Claude’s Computer Use
For Anthropic, the acquisition is a direct investment in one of its most strategically significant capabilities. Claude’s computer use feature — which allows the model to navigate live applications, run code across repositories, fill out forms, and complete multi-step workflows the way a human at a keyboard would — has seen dramatic improvement over the past year. On OSWorld, a widely used benchmark for AI computer use, Anthropic’s Sonnet models climbed from under 15% accuracy in late 2024 when the feature first launched to 72.5% with the latest Claude Sonnet 4.6 release. The model is now approaching human-level performance on tasks like navigating complex spreadsheets and completing web forms across multiple browser tabs.
The Vercept team’s deep expertise in visual perception and human-computer interaction maps directly onto the hardest remaining problems in that domain. As users increasingly rely on Claude for sophisticated, multi-tool workflows, the ability for the model to reliably see and act within software environments becomes a core competitive differentiator.
A Pattern of Strategic Acquisitions
This is not Anthropic’s first acquisition. The company also previously acquired Bun, signaling a broader strategy of bringing in specialized engineering teams to accelerate development rather than building every capability from scratch. For a company competing against well-resourced rivals like OpenAI, Google DeepMind, and Microsoft, targeted acquisitions offer a way to rapidly close capability gaps and secure talent in a fiercely competitive hiring market.
Vercept’s founders bring particularly strong credentials. CEO Kiana Ehsani is a robotics expert, and co-founder Ross Girshick is a pioneer in vision AI with prior experience at Meta and Ai2. Their work on how AI systems can see and act within the same software humans use every day is precisely the kind of foundational expertise Anthropic needs as it pushes Claude toward more autonomous, agentic operation.
The acquisition underscores a broader industry trend: as the frontier of AI moves from generating text to taking real-world actions, the companies that crack reliable, general-purpose computer use will hold a significant advantage. Anthropic is making clear it intends to be among them.