Anthropic called Claude Code Channels an open plugin mechanism. Weeks later it cut off subscription access for OpenClaw, the exact use case the feature was designed for. The reason: a single OpenClaw agent burns $1,000 to $5,000 a day.
image from grok
Anthropic launched Claude Code Channels in March 2026 as an 'open plugin mechanism' but restricted third-party agent harnesses less than a month later on April 4, citing compute economics—autonomous agents through harnesses like OpenClaw could consume $1,000-5,000/day in API costs versus flat subscription rates of $20-200/month. OpenClaw creator Peter Steinberger (who joined rival OpenAI in February) accused Anthropic of copying features before locking out open-source competition, a timing that benefited OpenAI's positioning as the developer-friendly alternative. Anthropic offered one-time credits and refunds but acknowledged the structural price gap cannot be closed under subscription models.
Earlier in March 2026, Anthropic announced Claude Code Channels: a feature letting developers control its Claude Code AI agent from Telegram and Discord. The company's blog post called it "an open plugin mechanism." On April 4, Anthropic sent an email to subscribers saying it would no longer allow Claude subscriptions to power third-party agent harnesses, including OpenClaw. The gap between "open" and "locked out" was less than a month.
The official reason is compute economics. A single OpenClaw agent running for one day could burn $1,000 to $5,000 in API costs, VentureBeat reported, with one observer calling it "the pace of a company watching its margin evaporate in real time." Anthropic's subscriptions (Pro at $20/month, Max at $100 to $200/month) were never priced for that usage pattern. Boris Cherny, Head of Claude Code, acknowledged on X that "subscriptions weren't built for the usage patterns of these third-party tools" and that "third party services are not optimized" for the prompt cache efficiency Anthropic's first-party tools use.
The company is offering a one-time credit equal to the monthly subscription price, redeemable until April 17, plus up to 30% off pre-purchased extra usage bundles. Cherny said Anthropic is still offering full refunds. But the underlying price gap doesn't close: a power user running autonomous agents via a third-party harness now pays market rate per token, not a flat subscription fee.
Peter Steinberger, the creator of OpenClaw, isn't buying the capacity argument. "Funny how timings match up," he posted on X. "First they copy some popular features into their closed harness, then they lock out open source." The features Steinberger cited (messaging agents through external services like Discord and Telegram) arrived in Claude Code Channels weeks before the API lockdown. Steinberger and investor Dave Morin attempted to negotiate a delay, and secured one week; the April 4 cutoff came as announced.
The timing is peculiar beyond the messaging contradiction. Steinberger joined Anthropic's rival OpenAI in February 2026. OpenClaw continues as an open-source project with support from OpenAI, TechCrunch reported. Anthropic restricted third-party harness access while OpenAI was positioning itself as the more developer-friendly alternative.
For the average Claude.ai user, nothing changed. For the power users who built the autonomous agent workflow ecosystem (the ones who ran OpenClaw instances continuously, integrated them with internal tools, automated code review and deployment pipelines) the economics shifted suddenly and with short notice. One OpenClaw user, cited on X, said switching to the recommended API billing would make their instances "far too expensive to make it worth using" and they'd likely migrate to a different model provider.
What Anthropic built with Claude Code Channels was technically an MCP (Model Context Protocol) integration, the same protocol OpenClaw uses. The company marketed it as openness. The practical effect of the April 4 policy is narrower: subscribers can still use Claude models via the API at market rates, but the flat-rate subscription window for third-party harnesses is closed. The distinction between "we support MCP" and "we support third-party harnesses consuming our subscription tier" is the line Anthropic is drawing.
The broader pattern is a recalibration. As agentic AI tools proliferate, the flat-rate subscription model has a structural problem: unlimited usage for $20/month doesn't survive a user who deploys a continuous agent workflow. Anthropic isn't alone in confronting this. The question is how companies communicate the transition. Anthropic's email cited capacity management and engineering constraints. Steinberger's interpretation is that the constraints arrived conveniently after Anthropic shipped its own competing integration.
What comes next is predictable. Some OpenClaw users will pay the higher rates, some will migrate to competitors, and some will recalibrate their usage to stay within the new billing model. The developer community that built workflows around flat-rate Claude access is now a test case for what happens when that subsidy ends.
Story entered the newsroom
Research completed — 7 sources registered. Anthropic cut off third-party harnesses including OpenClaw from Claude subscription auth on April 4 2026 at 12pm PT. The core contradiction: Anthropic
Draft (636 words)
Reporter revised draft (635 words)
Reporter revised draft (631 words)
Reporter revised draft (630 words)
Reporter revised draft based on fact-check feedback (627 words)
Approved for publication
Published (630 words)
Get the best frontier systems analysis delivered weekly. No spam, no fluff.
Artificial Intelligence · 1h 16m ago · 4 min read
Artificial Intelligence · 2h 8m ago · 3 min read
