Anthropic acquires Vercept

Anthropic is buying Vercept, a Seattle startup that built a “computer-use” agent capable of operating a remote Mac in the cloud, and will shut the product down on March 25. The acquisition, reported by TechCrunch, folds Vercept’s team into Anthropic as large model makers race to turn agent demos into production software that clicks buttons, fills forms, and executes transactions on a user’s behalf.

Vercept’s pitch was straightforward: instead of asking a model for advice, let it take the actions. Its product Vy could control a hosted Apple MacBook, a design that makes the agent feel like a familiar employee—opening browsers, navigating interfaces, and completing multi-step tasks. That also makes it a new kind of intermediary. If an agent mis-clicks, is tricked by a malicious prompt, or routes a payment to the wrong destination, the failure isn’t a bad answer on a chat screen; it is an executed action with a timestamp.

The deal arrives with visible strain in the agent ecosystem. Vercept raised a reported $50 million, according to CEO Kiana Ehsani, and had a high-profile investor list including Seth Bannon’s A12, plus angel backers such as Eric Schmidt and Jeff Dean, TechCrunch notes. Yet one co-founder, Oren Etzioni, publicly described the outcome as “throwing in the towel,” even while acknowledging a positive return, after the company gave customers 30 days to leave the platform. In other words: even well-funded agent startups can end up as talent acquisitions once the cost of distribution, compliance, and enterprise sales begins to dominate the engineering story.

The internal politics of the acquisition were unusually public. TechCrunch describes a dispute between Etzioni and Bannon in a LinkedIn thread, with accusations over hiring choices and responsibility for the outcome. The argument matters less than what it reveals: in agent software, the hard part quickly becomes everything around the model—reliability, security, customer support, and liability allocation. Those are the domains where incumbents with scale can price smaller entrants out.

Anthropic has been assembling the pieces for this shift. The company bought coding agent engine Bun in December to scale Claude Code, and now adds a team focused on more complex “computer-use” behavior. The pattern is consistent: build or buy the scaffolding that turns a model into a controllable worker.

As agents move into payments, procurement, and account access, the next market is likely to be “agent security”: audit logs, permissioning, fraud detection, and dispute resolution designed for software that acts like a user. Vercept’s product will disappear in a month, but the problem it tried to solve—delegating real actions to software—will not.

On March 25, customers who let Vy click for them will have to migrate. Anthropic will keep the engineers.