Within a single month, Anthropic shipped two distinct ways to control Claude from your smartphone. First came Remote Control for Claude Code, then Dispatch (now officially called “Répartition” in French-language builds) for Cowork. Two products. Same problem. Very different philosophies. And hovering in the background: OpenClaw, the open-source agent with 250,000 GitHub stars that lets you control your entire machine via WhatsApp or Telegram — with essentially zero guardrails.
The question worth asking isn’t just what these tools do. It’s what their simultaneous arrival signals about where AI-assisted computing is heading — and who is going to control it.
Three Tools, One Goal
All three solutions start from the same premise: there are moments when you need to direct your AI agent and you’re not sitting at your desk. The approaches, however, couldn’t be more different.
Remote Control creates an encrypted tunnel between your phone and a running Claude Code session on your PC. Your terminal stays open and active on the desktop; your phone simply becomes the input device. Think of it as a secure remote keyboard — what you type on your phone appears in your Claude Code session in real time. Setup is deliberately simple: launch Claude Code, run the remote control command, scan the QR code on your phone, and you’re live inside the Claude app.
Dispatch operates at a higher abstraction level. Rather than tunneling into a terminal session, it turns your phone into a command bridge that reaches across your entire desktop environment — including every application connected to Cowork. Send a message from the Claude mobile app, and Claude executes the task on your PC using whatever tools and connectors are available. It can even keep your computer from going to sleep while working. Dispatch is also easier to activate: the QR code appears immediately from within Cowork, and pairing takes seconds.
OpenClaw operates in a different category entirely. This open-source agent gets unrestricted access to your full operating system, controlled through WhatsApp or Telegram. The use cases demonstrated are genuinely impressive — and real. But within weeks of launch, over 40,000 machines were found to be exposed, drawing significant attention from cybersecurity analysts.
Anthropic didn’t ship two similar products by accident. Remote Control targets developers already inside the Claude Code ecosystem. Dispatch targets a much broader audience through Cowork’s more universal, non-developer-facing design. Two tools, two audiences, one platform strategy.
Remote Control: Honest Limitations
Remote Control works well for what it is — a convenient way to continue a Claude Code session when you’ve stepped away from your desk but remain in your general environment. Real-world testing confirms it’s fast and responsive. But the constraints are significant enough to shape how you’ll actually use it.
Only one device can connect to a session at a time. Connections time out after 10 minutes of inactivity. The terminal running Claude Code must stay open on your desktop — this is not a background daemon. And availability is restricted to Pro and Max plan subscribers. Perhaps most importantly, it does not support third-party Claude providers, meaning it only works within Anthropic’s own infrastructure.
Remote Control is best understood as a convenience feature for short-range use — reviewing and directing an active coding session from another room, not remotely managing your PC from a different city.
Dispatch: More Polished, Still Early
Dispatch offers a noticeably smoother user experience. The mobile pairing is faster, the interface more intuitive, and the integration with Cowork’s 38 application connectors meaningfully expands what you can delegate. Tasks sent from the phone can trigger sub-agents that run on the desktop, visible in real time inside the Cowork application. The ability to keep the desktop awake remotely is a practical addition that Remote Control lacks.
That said, Dispatch is explicitly flagged as a research-mode feature — and it shows. In testing, it occasionally loses track of which operating system it’s working on, and can conflate skills available in the web interface versus those installed locally. For browser-level actions, it requests explicit permission to operate Chrome without further confirmation — a level of access that’s mild compared to OpenClaw, but worth understanding before enabling.
| Feature | Remote Control | Dispatch | OpenClaw |
|---|---|---|---|
| Base platform | Claude Code | Cowork | Open source / any OS |
| Control interface | Claude mobile app | Claude mobile app | WhatsApp / Telegram |
| System access level | Terminal session only | Desktop + 38 connectors | Full OS |
| Security model | Encrypted tunnel, sandboxed | Sandboxed, permission-gated | Unrestricted (Wild West) |
| Target audience | Developers | General / non-coders | Power users / tinkerers |
| Maturity | Stable, limited scope | Research / early access | Mature but risky |
The Real Divide: Controlled Garden vs Open Frontier
The contrast between Anthropic’s approach and OpenClaw maps onto a broader philosophical split in how AI agents should operate. On one side: predictability, sandboxing, defined permissions, and a clear security boundary. Anthropic is investing heavily in this model — both Remote Control and Dispatch are built around it. You get a walled garden with reliable, auditable behavior.
On the other: OpenClaw’s full-OS access delivers capabilities that are genuinely remarkable. The use cases shared publicly aren’t fabricated — what you can accomplish with unrestricted agent access is extraordinary. But the exposure risk is real. Forty thousand vulnerable machines within the first weeks of availability is not an abstract concern.
OpenClaw’s power comes precisely from its lack of constraints — and so does its danger. For most users, the risks significantly outweigh the marginal gains over a well-configured Dispatch or Claude Code setup. The hidden treasure exists, but so do the traps.
Where This Is All Going
Today’s landscape is fragmented: AI agents live inside terminals, desktop applications, and open-source projects that speak entirely different languages. The trajectory over the next 18 months points toward consolidation — a unified command layer where mobile and desktop experiences are seamlessly connected, with a security model baked in from the start rather than bolted on later.
Anthropic is positioning itself to own that convergence. Remote Control and Dispatch are early moves in what is shaping up to be a significant race for the AI mobile control interface. Neither product is finished. Both signal intent.
If you’re a developer already in Claude Code, Remote Control is a useful convenience for in-range work. If you want to delegate desktop tasks from your phone with broader application support, Dispatch is the more capable — if rougher — option. Either way, the mobile AI control experience is being built right now, and what ships in the next year will look very different from what’s available today.
