As users increasingly rely on Claude for complex, multi-step workflows, from writing and running code across entire repositories to synthesizing research and managing cross-team tasks, Anthropic is doubling down on computer-use capabilities.
The company announced it has acquired Vercept, a startup focused on enabling AI systems to perceive and interact with software applications the way humans do.
Computer use allows Claude to operate inside live applications, navigating interfaces, completing forms, editing documents, and switching between tools, rather than relying solely on code execution. This unlocks multi-step problem-solving inside real-world software environments, making it possible to tackle tasks that previously required direct human interaction.
Why Vercept
Vercept was founded on the idea that making AI genuinely useful for complex task completion requires solving difficult perception and interaction challenges. Its co-founders Kiana Ehsani, Luca Weihs, and Ross Girshick have spent years researching how AI systems can “see” and act within everyday software environments.
Anthropic said Vercept will wind down its standalone product in the coming weeks, with the team joining Anthropic to push forward the frontier of AI-driven computer use.
Rapid Progress in Benchmarks
The acquisition follows the recent release of Claude Sonnet 4.6, which showed significant gains in computer-use performance. On OSWorld, a widely used benchmark for evaluating AI systems’ ability to operate software, Anthropic’s Sonnet models improved from under 15% performance in late 2024 when computer use first launched, to 72.5% today.
Claude Sonnet 4.6 is now approaching human-level performance on tasks such as navigating complex spreadsheets, managing browser tabs, and completing web forms across applications.
A Broader Expansion Strategy
Vercept joins a growing list of specialized teams acquired by Anthropic, including Bun. The company says it prioritizes teams whose technical ambition aligns with its own and whose work advances core capabilities while maintaining strong safety standards.
As AI systems move from chat interfaces to agents capable of operating full software environments, the race to master computer use is emerging as a critical next frontier, and Anthropic is clearly positioning Claude at the center of that shift.
This expansion comes alongside other recent initiatives, such as Anthropic’s updated Responsible Scaling Policy, which introduces a Frontier Safety Roadmap, regular Risk Reports, and a clearer separation between company commitments and industry recommendations, further emphasizing transparency and accountability in the deployment of advanced AI systems.
