Anthropic Acquires Vercept to Supercharge Claude's Computer Use Capabilities

A Strategic Acquisition in the Agent Race

On February 25, 2026, Anthropic announced its acquisition of Vercept, a Seattle-based AI startup specializing in computer-use agent technology. The deal brings Vercept's nine-person team, including co-founders Kiana Ehsani, Luca Weihs, and Ross Girshick, into Anthropic's workforce. Vercept's external products will be shut down by March 25, 2026.

The acquisition is Anthropic's latest move to strengthen Claude's ability to operate software autonomously, a capability that has become central to the competitive landscape of AI assistants. Vercept's technology for understanding screen interfaces and translating natural language into on-screen actions directly addresses the technical challenges Anthropic is working to solve.

What Vercept Built

Vercept developed tools for complex agentic tasks, most notably Vy, a computer-use agent that could operate a remote Apple MacBook in the cloud. The company's approach combined a proprietary model built to understand screen interfaces with the ability to map natural-language instructions to on-screen actions.

The Vy agent could complete tasks inside applications the same way a human with a laptop would: clicking buttons, filling forms, navigating menus, switching between applications, and handling multi-step workflows. This is fundamentally different from API-based automation. Rather than requiring applications to expose programmatic interfaces, Vy interacted with the visual layer that human users see.

This visual-interaction approach solves a practical problem. Most enterprise software does not have comprehensive APIs. Employees interact with these tools through graphical interfaces, and automating those interactions requires an AI system that can see and understand screens.

The Team Behind Vercept

Vercept's founding team brings significant technical credentials:

• Kiana Ehsani (CEO): AI researcher with expertise in embodied AI and perception
• Luca Weihs: Background in AI systems that interact with physical and digital environments
• Ross Girshick: A prominent computer vision researcher, previously at Meta AI Research (FAIR) and Microsoft Research

The startup had raised a total of $50 million from an impressive roster of angel investors, including former Google CEO Eric Schmidt, Google DeepMind chief scientist Jeff Dean, Cruise founder Kyle Vogt, and Dropbox co-founder Arash Ferdowsi.

Notably, one of Vercept's co-founders was recruited by Meta prior to the acquisition, which reportedly complicated the startup's trajectory and influenced the decision to join Anthropic.

Claude's Computer Use Progress: From 15% to 72.5%

The most striking data point in the announcement is the improvement in Claude's computer use performance. On OSWorld, a widely used evaluation benchmark for AI computer use, Anthropic's Sonnet models progressed from under 15% in late 2024, when computer use was first released as a feature, to 72.5% today with Claude Sonnet 4.6.

The 72.5% score represents near-human-level performance on tasks like spreadsheet navigation, multi-tab web form completion, and application switching. Anthropic describes this as demonstrating that Claude can now handle the kinds of multi-step tasks within live applications that would be impossible to solve through code alone.

The Vercept acquisition is expected to accelerate this trajectory further. The team's expertise in screen perception and interaction mapping directly targets the remaining 27.5% gap.

Strategic Context: The Agent Arms Race

The acquisition arrives at a moment when AI companies are aggressively investing in agent capabilities. The ability for an AI to operate software autonomously, rather than simply answering questions about it, is widely viewed as the next major capability frontier.

Anthropic is not the only company pursuing this direction:

• OpenAI has been developing computer use capabilities within its GPT models and through its Codex product line
• Google launched Antigravity, an AI development tool that incorporates agent-like features
• Microsoft integrated Copilot agents across its Office suite
• Startups like Adept AI and Multion have raised significant funding for computer-use agent technology

The Vercept acquisition gives Anthropic a dedicated team with proven expertise in exactly this domain, acquired rather than built from scratch.

What Changes for Claude Users

For existing Claude users, the Vercept acquisition signals continued investment in two areas:

Enterprise Automation

Claude Cowork, Anthropic's enterprise product, already connects to tools like Google Drive, Gmail, DocuSign, and FactSet. Enhanced computer use capabilities mean Claude can interact with applications that lack API integrations, expanding the range of tasks it can automate.

For example, an HR professional could ask Claude to navigate a benefits administration system, fill out enrollment forms for a new hire, and confirm the submission, all through the visual interface rather than requiring the benefits platform to offer an API.

Developer Tools

Claude Code, Anthropic's developer-focused product, benefits from improved computer use through better interaction with IDEs, terminals, and development tools. The ability to navigate complex development environments visually complements Claude's existing code generation capabilities.

Acquisition Pattern: Building Through Acqui-hires

The Vercept deal follows Anthropic's earlier acquisition of Bun, suggesting a pattern of targeted acqui-hires to fill specific technical gaps. Rather than building every capability internally, Anthropic is identifying teams with demonstrated expertise and folding them into the organization.

This approach has trade-offs. It is faster than internal development, but requires successful team integration. The nine-person Vercept team is small enough to integrate smoothly, but the loss of their independent product (Vy) means existing Vercept customers will need alternatives.

Financial and Market Implications

Vercept had raised $50 million in funding. While the acquisition price was not disclosed, the caliber of the angel investors and the strategic value of the team suggest a premium return for investors. For Anthropic, which recently closed a $30 billion Series G at a $380 billion valuation, the acquisition cost is a rounding error relative to its resources.

The timing also coincides with growing speculation about an Anthropic IPO. Strengthening product capabilities through acquisitions before a potential public offering is a standard strategic playbook.

Risks and Limitations

Computer use as a product category faces several challenges:

• Reliability: Even at 72.5% on benchmarks, a roughly 1-in-4 failure rate on complex tasks limits deployment in high-stakes workflows
• Security: An AI that can see and interact with any screen-based application raises data access and privacy concerns
• Latency: Visual interaction is inherently slower than API calls, creating performance trade-offs
• Trust: Enterprise customers need high confidence that autonomous agents will not take incorrect actions with real consequences

Conclusion

Anthropic's acquisition of Vercept is a focused investment in one of the most technically demanding capabilities in AI: teaching machines to use software the way humans do. The 72.5% OSWorld score demonstrates substantial progress, and Vercept's team brings the specific expertise needed to close the remaining gap. For the AI industry, the deal underscores that computer use has moved from a research curiosity to a strategic priority. The companies that solve this problem first will define the next generation of enterprise AI automation.