Update, March 19: References to Anthropic’s OAuth token deprecation have been updated to reflect that the rollout has been inconsistent and no formal announcement has been made.

In late January 2026, Anthropic sent a polite but firm email to Peter Steinberger, the Austrian developer behind a project called Clawdbot. The message: your name is too close to our trademark. Please change it.

Steinberger complied. By 3:38 a.m. ET on January 27, he’d renamed the project to Moltbot, then again to OpenClaw after bots swiped his original GitHub handle. The rebrand didn’t slow anything down. OpenClaw hit 100,000 GitHub stars, got acquired by OpenAI, and was declared “definitely the next ChatGPT” by Jensen Huang at GTC 2026.

Seven weeks after the cease-and-desist, Anthropic shipped Dispatch — a feature inside Claude Cowork that lets users assign tasks to Claude from their phone while the desktop app executes them autonomously. As Latent Space put it: “Everyone was waiting for OpenAI to unleash its computer-control agent — Anthropic dropped it first.”

The irony is structural. Anthropic’s trademark complaint catalyzed the viral cycle that made OpenClaw a household name. Now Anthropic has built a product that occupies the same functional space. And in between those two events, it moved to deprecate Claude Pro and Max OAuth tokens from third-party tools — including OpenClaw — pushing thousands of developers to reconsider their model stack entirely.

What Dispatch Actually Does

Dispatch is a research preview inside Claude Cowork, available first to Claude Max subscribers and rolling out to Pro users. The setup: keep the Claude Desktop app running on a Mac or Windows machine. Open the Claude mobile app, navigate to the persistent Cowork thread, and type a task — summarize an inbox, pull data from Notion, generate a report.

Claude on the desktop takes over. It reads local files, uses configured connectors and plugins, browses the web, builds spreadsheets, and writes documents. The user doesn’t see intermediate steps. When the work is done, the result appears in the same mobile chat thread.

The functional overlap with OpenClaw is hard to miss. Both systems let a user delegate real computer tasks to an AI agent that executes them autonomously. Both operate across messaging surfaces. Both handle multi-step workflows without babysitting. The key differences: Dispatch runs exclusively through Anthropic’s closed ecosystem (Claude Desktop + Claude Mobile), while OpenClaw is model-agnostic, open-source, and runs on whatever hardware the user chooses.

Simon Willison compared Dispatch favorably to OpenClaw. Ethan Mollick did the same. The consensus in developer communities: this is Anthropic’s answer to the agent wave it inadvertently created.

The Token Lockout That Backfired

Dispatch didn’t arrive in isolation. It landed after Anthropic spent weeks tightening the screws on third-party access to Claude.

In January 2026, thousands of OpenClaw users running Claude Pro or Max subscriptions via OAuth tokens found their setups dead overnight. No warning email. No grace period. Requests returned 403 errors and a “Terms of Service violation” notice. Anthropic had decided that piping a Max subscription through anything other than Claude Code was a bannable offense.

The compliance documentation followed on February 20, making it explicit: OAuth tokens from Free, Pro, or Max accounts are banned in any third-party tool, including Anthropic’s own Agent SDK. The Hacker News thread drew hundreds of comments from developers who’d built production workflows around Claude-powered OpenClaw instances.

Anthropic’s apparent strategy: force users into its own ecosystem — Claude Desktop, Claude Code, and now Dispatch — rather than letting third-party tools ride on consumer-tier subscriptions.

$200 a Month Became $15

The community response was swift and structural. Rather than capitulating, developers rebuilt.

One widely circulated Medium guide documented replacing a $200/month Claude Max-powered OpenClaw deployment with two $5/month VPS instances on Hostinger and Contabo, running Kimi K2.5 as the primary model and MiniMax M2.5 as a cheap fallback. Total cost: approximately $15 per month. The author reported that the new setup replaced half their n8n automation workflows in the process.

The pattern extends beyond one guide. OpenClaw’s architecture is fundamentally model-agnostic — it routes messages to whichever LLM the user selects. When Anthropic cut off access, users switched to Kimi, Mistral, GPT-5.4, and local models running on consumer hardware. The platform’s value proposition — an open-source operating system for personal AI agents — turned out to be structurally independent of any single provider.

This dynamic mirrors a pattern from cloud computing’s early days: when a vendor locks down access, open-source alternatives absorb the demand. AWS restricting Elasticsearch led to OpenSearch. Docker’s licensing changes pushed users to Podman. Anthropic’s token lockout is accelerating the model-agnostic agent stack.

The Three-Move Sequence

Viewed together, Anthropic’s actions over the past eight weeks form a coherent strategy:

Move 1: Eliminate the trademark association. The cease-and-desist to Steinberger in January removed the “Clawd” naming that linked the viral open-source project to Anthropic’s brand. This was defensible trademark protection. It also backfired spectacularly — the rename drama generated a second wave of press coverage and accelerated the project’s growth from 9,000 to 100,000+ GitHub stars.

Move 2: Restrict third-party access to Claude on consumer plans. By moving to deprecate OAuth tokens in January/February, Anthropic began closing the pipeline that let OpenClaw users run Claude through their own infrastructure. The rollout has been inconsistent — some users retained access while others lost it overnight — but the direction is clear: channel users toward Anthropic’s own interfaces and enterprise API pricing.

Move 3: Ship a first-party alternative. Dispatch gives Claude users the autonomous task execution they were getting through OpenClaw, but locked inside Anthropic’s closed-loop ecosystem — Claude Desktop on the machine, Claude Mobile as the remote control, Anthropic’s servers doing the inference.

Each move is individually rational. The sequence creates an uncomfortable narrative: Anthropic identified the most exciting use case in the AI agent space, tried to suppress the open-source version, locked users out of the underlying model, and then shipped its own product that does the same thing.

Why Dispatch Can’t Replace OpenClaw

Dispatch has a clean UX and the benefit of deep integration with Claude’s capabilities. But it has constraints that make it a complement, not a replacement, for OpenClaw:

Single-provider lock-in. Dispatch runs on Claude only. If Anthropic raises prices, changes rate limits, or suffers downtime, users have no fallback. OpenClaw users running Kimi K2.5 as a primary and MiniMax M2.5 as a backup demonstrated exactly this resilience during the January lockout.

Closed-source platform dependency. OpenClaw’s codebase is public, forkable, and extensible. Dispatch is a feature inside a proprietary app. Users can’t inspect the execution pipeline, add custom tools, or modify the agent’s behavior beyond what Anthropic’s interface exposes.

Desktop-tethered. Dispatch requires the Claude Desktop app to stay running on a physical machine. OpenClaw runs on VPS instances, Raspberry Pis, Mac Minis, or any server — headless, 24/7, without a desktop GUI.

Single thread. The current Dispatch preview supports one persistent conversation thread. OpenClaw users routinely run agents across WhatsApp, Telegram, Discord, Slack, and email simultaneously.

These aren’t temporary limitations of a research preview. They reflect a fundamental architectural difference between a closed-ecosystem feature and an open-source platform.

The Ecosystem Has Already Forked

Anthropic’s moves are happening against a backdrop of rapid ecosystem fragmentation. At GTC 2026, NVIDIA launched NemoClaw, an enterprise-grade OpenClaw stack with policy enforcement and network guardrails. A minimalist alternative called Claw-Family shipped v0.1.4, positioning itself as OpenClaw with “99% fewer lines of code.” Meta launched Manus “My Computer” as its own desktop agent play.

The agent layer is fragmenting exactly the way operating systems did in the 1990s: a dominant open-source platform (OpenClaw/Linux), enterprise distributions (NemoClaw/Red Hat), minimalist forks (Claw-Family/Alpine), and proprietary alternatives from major vendors (Dispatch/Windows). Anthropic is building Windows. The community has already chosen Linux.

What This Means for the Agent Market

The Anthropic-OpenClaw arc clarifies three dynamics that will shape the agent market through 2026:

Model providers can’t capture the agent layer. Anthropic tried: trademark pressure, token lockouts, first-party product launch. The community routed around every move. The agent orchestration layer — the part that connects LLMs to real-world tools and workflows — is decoupling from the inference layer. Whoever provides the model matters less than whoever controls the agent runtime.

Open-source momentum compounds. OpenClaw’s 100,000+ GitHub stars, OpenAI acquisition, and Jensen Huang’s endorsement all happened after Anthropic’s intervention. Each attempt to constrain the project added energy to it. Open-source agent platforms now benefit from the same network effects that made Linux unstoppable in server computing.

The $15/month stack changes the economics. When the community demonstrated that a fully functional AI agent setup costs less than a Netflix subscription, it removed the pricing moat that companies like Anthropic rely on. Enterprise customers will still pay premium prices for SLAs, compliance, and support. But the baseline — a personal AI agent that handles email, calendar, messaging, and automation — is now effectively free.

Anthropic’s Dispatch is a good product solving a real problem. But it arrived in a market that Anthropic’s own actions helped create, serving users it spent weeks trying to lock out, competing against an open-source alternative it tried to rename out of existence.

The cease-and-desist letter that started this whole sequence may turn out to be the most expensive piece of legal correspondence in Anthropic’s history — not because of any lawsuit, but because of what it set in motion.

Sources: Quasa.io, Latent Space, Medium/@rentierdigital, CNET, CNBC/Mad Money, NVIDIA GTC Blog, dev.to