On March 18, Anthropic launched Claude Cowork Dispatch — a feature that lets Claude remotely control your desktop, coordinate across tabs, and dispatch work to subagents. The timing was no accident: it dropped the same day as a Latent Space podcast episode with Felix Rieseberg discussing the technical architecture. Multiple observers, from Simon Willison to Ethan Mollick, reached the same conclusion: this is Anthropic's OpenClaw play.
What Claude Cowork Dispatch actually is
At its core, Cowork Dispatch is an Electron-based remote control layer that sits inside the Claude desktop app. It gives Claude the ability to see your screen, navigate your browser, click buttons, type into fields, and coordinate multiple parallel tasks — all from within the Claude interface. The "dispatch" part refers to subagent coordination: Claude can break a complex request into sub-tasks and farm them out.
The technical choices are revealing. Anthropic built on Electron rather than a native platform, which means sandboxing is baked into the Chromium security model. Felix Rieseberg discussed this extensively on the podcast — the decision to use Electron wasn't about developer convenience but about having battle-tested sandboxing primitives at the OS level. Every agent action runs inside Chromium's process isolation.
Unlike OpenClaw, which is a background daemon that runs continuously and can be triggered by cron jobs, heartbeats, and scheduled tasks, Cowork Dispatch is session-scoped. It lives inside the Claude desktop app window. When you close the app, the agent stops. OpenClaw's gateway model gives it persistence that desktop-embedded agents don't have yet.
The comparison everyone is making
The immediate reaction from Simon Willison and Ethan Mollick — both prominent voices in the AI space — framed Cowork Dispatch as Anthropic's answer to OpenClaw. This is notable because Anthropic famously "fumbled" (as the Latent Space piece put it) the Clawdbot relationship. OpenClaw became the dominant local agent framework partly because Anthropic didn't have a competing product.
Now they do. But the architectural differences matter. OpenClaw is infrastructure — a daemon that orchestrates sessions, manages memory, schedules tasks, and provides a skill/plugin ecosystem. Cowork Dispatch is an application layer — a user-facing feature inside Claude Desktop. They solve overlapping but different parts of the stack.
Jensen Huang reportedly said at GTC that "every company needs an OpenClaw strategy." Cowork Dispatch is Anthropic's strategy. Whether it's the right one depends on whether you believe agents should be infrastructure (always on, deeply integrated) or applications (open when needed, close when done).
What this means for the agent ecosystem
Three things shift with this launch:
1. Desktop agent control is now table stakes. If Anthropic ships it, Google and OpenAI will follow (or already are). The "AI that can use your computer" category just went from experimental to production-grade from a major lab. Expect this to accelerate browser-use and desktop-use tooling across the ecosystem.
2. The harness wars get real. OpenClaw, Cowork Dispatch, Claude Code, Codex — these are all competing to be the orchestration layer for AI agents. The winner won't necessarily be the one with the best model but the one with the best execution environment, memory, and coordination primitives. OpenClaw's advantage here is that it's model-agnostic and runs as infrastructure rather than being tied to a single provider's desktop app.
3. Subagent coordination is the next battleground. Both Cowork Dispatch and OpenClaw now support spawning subagents. The interesting question is interoperability — can an OpenClaw subagent talk to a Cowork Dispatch session? Not yet. But as the ecosystem matures, the platforms that play well with others will win.
For PromptEngines: what to watch
We run OpenClaw as our agent infrastructure at PromptEngines. Cowork Dispatch doesn't change that — OpenClaw's daemon model, skill system, and cron scheduling are purpose-built for what we need. But we should watch two things:
First, Cowork Dispatch's remote control primitives. If Anthropic open-sources the Electron sandboxing and browser control layers, those could become useful components for OpenClaw skills or complementary tools. The desktop control space is evolving fast, and having Anthropic invest in it benefits everyone.
Second, the competitive pressure on OpenClaw's feature set. Anthropic shipping subagents and dispatch means OpenClaw needs to keep innovating on its unique advantages: model-agnostic execution, persistent memory (ByteRover), cron scheduling, and the skill ecosystem. The gap that made OpenClaw the default choice is narrowing, not because OpenClaw got worse, but because the alternatives got better.
The bottom line
Claude Cowork Dispatch is a significant launch. It validates everything the agent ecosystem has been building toward — that AI agents should be able to control your computer, coordinate complex tasks, and work autonomously. Anthropic is late to this party, but they arrived with a well-engineered product.
For those of us already running agent infrastructure: the question isn't whether to switch (we shouldn't — yet), but whether to start building bridges. The agent ecosystem is big enough for multiple orchestration layers. The platforms that connect will win.
Sources: Latent Space AINews · Latent Space Podcast: Felix Rieseberg · Simon Willison · Ethan Mollick