Two days after cutting off 135,000 OpenClaw users from Claude subscriptions, Anthropic just tipped its hand on what comes next. And it is exactly what you think.

In an exclusive interview with Semafor, Anthropic Chief Commercial Officer Paul Smith confirmed that customers are asking the company to build its own version of OpenClaw. When pressed on whether Anthropic would deliver, Smith did not deny it. "Without speculating too much on our product roadmap, it evolved pretty quickly," he said.

Semafor reporter Reed Albergotti described his takeaway bluntly: "It was more a matter of when, not if."

The Agent Platform Arms Race

The timeline here is worth laying out because it shows how fast this race is accelerating. OpenAI acquired OpenClaw in February 2026 and turned it into the default agent platform for ChatGPT users. Nvidia responded by announcing NemoClaw, marketing it as the security-enhanced enterprise alternative. Now Anthropic is publicly confirming demand for its own version.

Anthropic already has a prototype of sorts. Claude Cowork, the companys desktop agent product, recently added a feature called Dispatch that lets users message Claude from their phone while it works autonomously on their desktop. Smith pointed to this as evidence that Anthropic is already moving in the direction customers want. "Think of the features that people found useful: the ability to have a whole bunch of agents working on my desktop but remote control from my mobile device," he said.

Kill the Subsidy, Then Launch the Replacement

Here is the strategic play that nobody is saying out loud. Anthropic just killed flat-rate access to Claude for third-party agent tools. Now it is confirming it will build its own first-party agent platform. The message to developers is clear: stop building on someone elses platform. Ours is coming, and it will be the only one that works natively with Claude.

This is the classic platform move. Cut off the ecosystem that grew organically, then offer your own version that you control end to end. Apple did it with third-party apps. Google did it with Chrome extensions. Now Anthropic is doing it with AI agents.

The Peloton Problem

Smith offered one surprisingly honest moment about the limits of all this. When asked whether Anthropics enterprise customers can keep up with the pace of AI change, he used an analogy: "Rolling out broad-based AI to an entire employee base is like putting a Peloton in everyones sitting room. It does not mean everyones going to start to get fit."

That is the most honest thing any AI executive has said this year. The tools are getting smarter. The companies selling them know most buyers will never use them properly. And they are building the next product anyway.

First reported by Semafor. Additional context from Axios reporting on Anthropics leaked source code.