Anthropic acquires Vercept, a computer-use startup

Anthropic has acquired Seattle startup Vercept, which specialized in agentic tools for computer control. The deal landed at an awkward moment: shortly before it closed, Meta managed to poach one of the company's founders. The acquisition went ahead anyway, and it signals clearly where Anthropic intends to head in the coming years.

The race for so-called "computer use" — the ability of AI agents to work with application interfaces on their own, the way a living person does — began among leading tech companies back in 2024. OpenAI introduced Operator, which can navigate a browser and fill in forms. Google built comparable capabilities into the Gemini ecosystem. Anthropic did not stay on the sidelines: in autumn 2024 it announced its own computer use feature for Claude 3.5 Sonnet, which let the agent move the cursor, click elements and type text. But in maturity of implementation, Anthropic trailed its competitors, and the Vercept purchase is meant to fix exactly that.

The Seattle startup bet on something most players in the market only talk about: building agents that can work inside real applications under conditions as close as possible to a user's everyday practice. Vercept's key product is an AI agent that does not merely move around web pages but also interacts with desktop software: corporate applications, utilities and complex interfaces where standard API integrations are simply unavailable. In essence, it is an agent that "sees" the screen and "acts" with its hands, just like an employee at a laptop. The gap between a polished demo and a reliable production-grade solution is huge, and Vercept, by all appearances, has gone considerably further down that road than many others.

Meta's attempt to poach one of Vercept's founders before the deal closed says a lot. The largest tech corporations have long competed not only for finished products but also for specific people with rare expertise. Agentic systems for computer control demand a specific mix of knowledge: computer vision, handling of user interfaces, planning of multistep tasks and dependable recovery from errors. Few specialists with that background exist, and the fact that Meta deliberately hunted the Vercept team further confirms the strategic value of this area.

For Anthropic, the acquisition is not just an addition to its patent portfolio or headcount. The company has been steadily building infrastructure for an agentic future: the Model Context Protocol, expanded tool access, and long-horizon planning capabilities in Claude. Computer use remains the hardest and, at the same time, the most attractive frontier in that work. If an agent can not only answer questions but also carry out multistep tasks in a real software environment, it stops being a chatbot and becomes a full-fledged digital employee. That promise drives the interest of enterprise customers willing to pay for automating routine processes.

The industry as a whole is moving toward a model in which AI agents will not assist humans with suggestions but take over specific tasks from start to finish. That changes the reliability requirements: an agent that makes mistakes when filling in a form or loses context midway through a task is not merely useless, it is dangerous in a corporate setting. That is why the technologies Vercept developed, tools for precise, stable interaction with applications, are critically important for any company aiming for leadership in the agentic segment.

The acquisition of Vercept is a deliberate bet by Anthropic that the next major round of AI competition will be played out not in the space of language models but in the space of actions. Whoever first teaches its agent to work with the real world reliably and at scale will gain an advantage that will be extremely hard to close. By all appearances, Anthropic understands this very well.

*Meta is designated an extremist organization and banned in Russia.