On April 4 at 12 PM PT, Anthropic cut off Claude Pro and Max subscription coverage for OpenClaw and other third-party harnesses. OpenClaw still works with Claude, but now users have to pay through Extra Usage bundles or bring their own Anthropic API key.

Anthropic is accused of “copy and close”… copy the features of OpenClaw and then prevent them from being used in the same way Claude Code is.

Claude Code and Claude Cowork, Anthropic's own first-party products, remain covered under the subscription. The cheap path stayed open for Anthropic's tools and closed for everyone else's.

That is why people are angry.

Anthropic's public case is straightforward.

OpenClaw-style usage is not normal chat usage. A standard Claude subscriber asks a question in a browser tab, gets an answer, and moves on. An OpenClaw user can run an autonomous agent all day. That agent might browse the web continuously, watch email and calendars, execute code, send messages, maintain long-running memory, and schedule recurring tasks.

From Anthropic's perspective, that is not a normal subscription workload. It is an always-on compute job attached to a flat monthly price.

They also argue that first-party tools like Claude Code and Claude Cowork are far more optimized for prompt caching. That matters because prompt caching reduces repeated compute costs. If the same context can be reused, the model does less expensive work. Anthropic's complaint is that OpenClaw and similar third-party harnesses were much worse at taking advantage of those efficiencies, which made them disproportionately expensive to serve.

Boris Cherny, head of Claude Code, even said he personally submitted pull requests to improve OpenClaw's prompt caching behavior before this policy change went live.

In the most charitable reading, Anthropic is not attacking open source. It is unbundling a use case that its subscription pricing model never really covered.

That is the clean version.

The problem is not that Anthropic's explanation is impossible.

The problem is the optics.

If Anthropic had said, "We cannot economically support autonomous agent workloads inside flat-rate subscriptions anymore, so everyone has to move to metered billing," people would still be upset, but the logic would at least be neutral.

That is not what they did.

Instead, they said Claude subscription coverage still applies to Anthropic's own products, and no longer applies to third-party harnesses like OpenClaw.

That instantly changes the story from an infrastructure issue into an ecosystem issue.

The exact same model family remains available under the subscription inside Anthropic's own harnesses. But if an open-source project built the workflow users actually want, the economics just changed overnight.

That is why Peter Steinberger, OpenClaw's creator, framed the move as copy-then-close. His criticism is not just that Anthropic cut off OpenClaw. It is that OpenClaw helped popularize a certain style of agent workflow, Anthropic copied many of those ideas into Claude Code, and then Anthropic shut down the cheap subscription path that made OpenClaw viable for a huge slice of users.

Whether you think that is malicious or just inevitable, the sequence is brutal.

A lot of people still think the story is that Anthropic banned OpenClaw.

That is too shallow.

The real thing Anthropic banned was the old economics.

OpenClaw became powerful not just because it worked, but because it was affordable. Hobbyists, builders, and power users could attach Claude subscriptions to long-running agents and get something close to a real digital worker for a predictable monthly price.

That made the whole category explode.

Now that predictability is gone.

Users are being pushed into one of three buckets:

Some reports estimate the effective cost increase for heavy users could be anywhere from 5x to 50x, depending on how they were using OpenClaw before.

That is not a tweak. That is a platform reset.

And it lands hardest on exactly the people who did the most to make Claude the default model inside the agent ecosystem.

The people who built tutorials, workflows, Telegram bots, Discord setups, cron automations, and courses around OpenClaw are the same people who now have to explain why the cost model just broke.

Anthropic did not merely change a pricing rule. It invalidated an entire layer of community assumptions.

This is where the backlash gets sharper.

The confirmed tactics are already enough to tell the story.

Anthropic used a product-class billing cutoff. It preserved the first-party subscription path for Claude Code and Claude Cowork. It moved third-party harnesses to overages and API usage. It softened the blow with one-time credits and discounted bundles. And it rolled the policy out in a way that made clear OpenClaw was just the first target, not the last.

That alone is enough for critics to say Anthropic solved a real engineering problem in the most strategically self-serving way possible.

Then there is the disputed layer.

Some users claim Anthropic was not just classifying traffic by product or account, but was potentially scanning prompts or metadata to identify OpenClaw-style usage. Others claim requests that failed before began working once Extra Usage was enabled. Those receipts are not fully settled in public reporting yet, so they should be treated carefully.

But the reason the allegation spread so quickly is simple: it fits the mood of the moment.

After the Mythos leak, the Claude Code source leak, and the DMCA mess that wiped out 8,100 GitHub repos, a lot of people were already primed to interpret every Anthropic move as more strategic than innocent.

So even if the prompt-scanning claim ends up overstated, the deeper accusation is already cemented.

Anthropic is building a moat around Claude Code.

And OpenClaw is one of the things outside that moat.

To be fair, the company is not making an obviously absurd argument.

If a $20 or $200 flat-rate subscription is being used to power something that looks more like a full-time agent than a chat assistant, the math eventually breaks. Either the service gets slower, the quality drops, or the pricing model changes.

Anthropic is choosing to make the heavy-use case metered instead of letting it quietly eat the economics of the whole system.

That is a rational business move.

The best argument against Anthropic is even simpler.

They did not make the expensive use case metered for everyone.

They made it metered for the open ecosystem while preserving the subscription path for their own closed harness.

That is why this is not being read as neutral capacity management.

It is being read as the moment Anthropic chose its side.

Not open-source agents. Not third-party builders. Not the weird experimental ecosystem that helped make Claude feel like the frontier model for real agent workflows.

Anthropic chose Claude Code.

And by doing that, it changed the agent ecosystem overnight.

stay tuned…

Wes “why can’t we all just get along” Roth

PS here’s my video of all of this: