Banning OpenClaw and Launching Its Own Agent Platform: Anthropic’s Infrastructure Ambitions Come to Light
TechFlow Selected TechFlow Selected
Banning OpenClaw and Launching Its Own Agent Platform: Anthropic’s Infrastructure Ambitions Come to Light
Whoever controls the routing and composition logic of Agents makes the underlying models swappable.
Navigating Web3 tides with focused insightsAuthor: Claude, TechFlow
TechFlow Introduction: This week, Anthropic executed a two-pronged strategy: on April 4, it cut off subscription-based access for 135,000 OpenClaw instances; on April 8, it launched Claude Managed Agents—a cloud-hosted agent orchestration service.
Together, these moves signal a clear strategic pivot—from selling model APIs to selling agent runtime infrastructure. With its Annual Recurring Revenue (ARR) having just surpassed $30 billion, Anthropic is leveraging pricing power and platform lock-in effects to redefine the rules of the AI agent landscape.
Within a single week, Anthropic completed two decisive actions—so aligned in intent that interpretation is nearly unnecessary.
On April 4, Anthropic officially revoked Claude Pro and Claude Max subscribers’ ability to use their subscription quotas via third-party agent frameworks such as OpenClaw. Overnight, 135,000 active instances were forced to shift either to pay-per-use billing or API-based pricing. Four days later, on April 8, Anthropic launched the public beta of Claude Managed Agents—a full-stack, cloud-hosted infrastructure offering sandboxed execution, state management, and multi-agent coordination.
One door closes while another opens. The open-source community’s outrage is understandable—but from a business logic standpoint, both moves serve the same goal: Anthropic no longer wants to be merely a model supplier. It aims to become the foundational infrastructure platform for the agent era.
Banning OpenClaw: The $20 All-You-Can-Eat Buffet Ends
OpenClaw’s popularity needs no introduction.
Previously, users ran agents using Claude’s $20 monthly subscription quota. But the economics didn’t add up: a single heavy user could consume $1,000–$5,000 worth of compute per day—placing unsustainable strain on Anthropic’s infrastructure.
According to VentureBeat, Boris Cherny, Head of Claude Code at Anthropic, announced the change on X (formerly Twitter), stating that subscription plans “were never designed for usage patterns involving third-party tools,” and that the company must “prioritize customers using our own products and APIs.”
The timeline adds further nuance.
In January, Anthropic filed a trademark opposition against Clawdbot. On February 14, Steinberger announced he was joining OpenAI—and Sam Altman publicly welcomed him. On February 20, Anthropic updated its Terms of Service to explicitly prohibit using subscription OAuth tokens with third-party tools. On April 3, Semafor reported that Anthropic was building its own OpenClaw competitor—and Chief Commercial Officer Paul Smith acknowledged that customers had “been asking us to build this” for some time. On April 4, the policy went into full effect.
Steinberger’s response was blunt: “First copy the most popular open-source features into your closed internal tool—then lock open source out.” He and investor Dave Morin attempted negotiations with Anthropic but secured only a one-week delay in enforcement.
Anthropic offered two transitional measures: a one-time credit equal to the monthly subscription fee, and up to a 30% discount on pre-purchased additional usage bundles. Yet for heavy users, switching from flat-rate monthly billing to usage-based pricing could increase costs by up to 50-fold.
Managed Agents: From Selling Models to Selling Runtime Infrastructure
In the same week it banned OpenClaw, Anthropic unveiled its alternative.
On April 8, Claude Managed Agents entered public beta. According to Anthropic’s engineering blog, the service draws inspiration from operating system abstraction principles: it decomposes agents into three independently replaceable components—session (conversation logs), harness (the invocation loop), and sandbox (code execution environment). These components are fully decoupled, so failure in any one does not affect the others.
The engineering blog explains why this architecture is necessary. Early versions bundled all components into a single container—making it a “pet”: once it crashed, the entire session was lost, and debugging couldn’t access user data.
With decoupling, containers become “cattle”—crashed ones are simply replaced, and the harness restores state from session logs to resume execution.
Pricing-wise, Managed Agents adds a $0.08 fee per session-hour (billed per millisecond) on top of standard API token charges. Idle wait time is not billed. Web searches triggered by agents cost $10 per thousand queries.
According to SiliconANGLE, early adopters include Notion, Rakuten, Asana, and Sentry. Asana has embedded agents into its project management workflows, creating an “AI teammate” capable of autonomously claiming tasks and drafting deliverables; Sentry has paired its existing debugging agent with a Claude-powered patch-generation agent—cutting the cycle from bug detection to pull request submission from months down to weeks.
Two features are currently in research preview: first, agents can spawn child agents when handling complex tasks; second, agents gain self-assessment capabilities—developers define success criteria, and Claude iterates autonomously until those criteria are met.
The Platform Economics Behind the Two Moves
Viewing both actions side-by-side reveals a crystal-clear commercial logic.
Anthropic’s ARR has just surpassed $30 billion. Per The Information, this represents more than a threefold increase from roughly $9 billion at the end of 2025—and over 1,000 enterprise customers now spend more than $1 million annually.
Separately, Claude Code contributes over $2.5 billion in annualized revenue. At this scale, allowing 135,000 OpenClaw instances to consume thousands of dollars in compute per month—for just $20—was financially unsustainable.
Yet pure cost control alone doesn’t fully explain the timing of Managed Agents’ launch.
Angela Jiang, Anthropic’s Head of Platform Products, stated in an interview that a gap remains between the capabilities of Anthropic’s models and how enterprises actually use them. Managed Agents aims to enable enterprises to deploy “teams of Claude agents” to handle real-world workloads.
This is a classic platform lock-in strategy. Once enterprises run their agents atop Anthropic’s managed infrastructure, data pipelines, monitoring configurations, and permission systems become deeply embedded in daily operations—making migration prohibitively costly.
For a company valued at $380 billion and actively considering an IPO, such stickiness holds far greater value than simple API call fees.
Multiple analysts and prominent social media figures have long argued that “the real battlefield for AI lies at the orchestration layer.” Whoever controls the routing and composition logic of agents renders the underlying models interchangeable.
OpenClaw already supports seamless switching among Claude, GPT-4o, Gemini, and other models. After being forced off fixed-rate plans, some of its 135,000 users will inevitably migrate to local models or alternative providers.
Google took similar action in February, prohibiting third-party tools from borrowing Gemini CLI’s OAuth credentials. Taken together, these developments signal AI’s industry-wide shift—from “model competition” to “platform competition.”
The subscription-based, unlimited-usage model is ending across the board. Usage-based pricing and infrastructure bundling are becoming the new normal.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
深潮TechFlow
