Anthropic acquires Vercept to boost Claude’s ability to operate software like humans

Anthropic said on 25 February it has acquired Vercept to strengthen the ability of its Claude models to perform complex tasks directly inside software applications, as competition intensifies to build AI systems that can act autonomously across digital workflows.

The deal underscores a growing push among AI developers to move beyond chatbots toward “computer-using” agents that can navigate live apps, handle multi-step work and coordinate across tools much like human employees.

Anthropic did not disclose the financial terms of the deal.

“People are using Claude for increasingly complex work, writing and running code across entire repositories, synthesising research from dozens of sources, and managing workflows that span multiple tools and teams,” Anthropic said in a statement.

Computer-use capabilities allow Claude to operate software interfaces rather than rely solely on code execution or text responses. “Computer use enables Claude to do all of that inside live applications, the way a person at a keyboard would,” the company said. “That means Claude can take on multi-step tasks in live applications and solve problems impossible with code alone.”

Vercept’s technology focuses on enabling AI systems to perceive and interact with graphical interfaces,  a key technical hurdle for building autonomous digital agents.

“Making AI genuinely useful for completing complex tasks requires solving hard perception and interaction problems,” Anthropic said, describing Vercept’s core approach.

The startup’s founders, Kiana Ehsani, Luca Weihs and Ross Girshick, and their team will join Anthropic, while Vercept will discontinue its external products in the coming weeks.

“The Vercept team … have spent years thinking carefully about how AI systems can see and act within the same software humans use every day,” Anthropic said, adding that the expertise “maps directly onto some of the hardest problems we're working on.”

The acquisition comes shortly after Anthropic released Claude Sonnet 4.6, which the company said significantly improves the model’s ability to complete tasks inside operating systems and web environments.

On OSWorld, a benchmark for AI computer-use performance, Anthropic said its Sonnet models improved from “under 15% in late 2024 … to 72.5% today,” approaching human-level capability on tasks such as navigating spreadsheets and filling out web forms across browser tabs.

Vercept is the latest in a series of acquisitions by Anthropic aimed at accelerating its technical roadmap. The company said it targets teams “whose technical ambitions match ours” and whose work aligns with its emphasis on safety.