--- title: "Cloud Sessions and Dispatch" description: "Work that survives your laptop closing — and the /teleport handoff." order: 25 duration: "15 min" chapter: "06-delegation-not-prompting" type: lesson --- ## Work that follows you between devices Remote Control keeps the work on your machine and gives your phone a window into it. There is a second model: the work runs on Anthropic's infrastructure and *any* device is a window into it. Open **claude.ai/code** in a browser — or the Code tab in the Claude mobile app — and start a session against one of your repositories. The session runs in a cloud sandbox. Close the laptop, and it keeps working. Check it from your phone in line at the bank. From the terminal, `claude --cloud "fix the failing invoice test"` kicks off the same kind of session without leaving your shell. The piece that makes this a system rather than a gimmick is **handoff**: | You are... | You want to... | How | |-----------|----------------|-----| | In the cloud session | Finish the work locally, in your editor | `/teleport` — pulls the branch and the full conversation into your terminal | | At your terminal | Push a long task somewhere it survives your laptop closing | `claude --cloud "task"` | | On your phone | Start something now, polish it at your desk later | Start in the mobile Code tab, teleport it down when you sit down | | In Slack | Turn a thread into work | Mention @Claude — it spawns a session with the thread as context | The rule of thumb: **Remote Control when the work needs your machine** (your files, your local MCP servers, your credentials). **Cloud sessions when the work just needs to get done** and you want it to survive your laptop going to sleep. Cornwall Market's QuickBooks skill lives on Sarah's machine, so she uses Remote Control for bookkeeping — but when she wants the store's website copy updated, she fires a cloud session from her phone and never opens the laptop at all. ## The orchestrator and the alternative There is a third way to work from your phone, and it splits the difference: **Dispatch**. It lives in Cowork (the Claude Desktop app you used in Chapter 05). Message a task from the Claude app on your phone, and Dispatch spawns a session on your paired desktop at home — Cowork for knowledge work, Claude Code for development — with full access to your local files, connectors, and skills. Your phone is the steering wheel; your desktop is the engine. Pavel Huron ran a 48-hour experiment that captures the pattern perfectly:
Pavel's 48-Hour Dispatch Experiment

Morning coffee: Dictated a competitor analysis task. Dispatch spawned a Cowork session and started researching.

Walking the dog: Voice-messaged a code review request. Dispatch spawned a Code session and began reviewing.

Jump arena with kids: Checked in on both tasks between trampolines. Both were progressing. Gave a couple of course corrections.

Result: 25 minutes of direction, 3+ hours of execution.

The ratio is the point. Dispatch does not make you faster at doing work. It lets you spend 25 minutes directing and get 3 hours of output. That is not a productivity improvement — it is a category change. OpenClaw (formerly Moltbot, formerly Clawdbot) offers a similar vision from the open-source side — an autonomous agent you direct from WhatsApp, Telegram, or Signal. The comparison is inevitable. > "After using it a bit, Claude Cowork Dispatch covers 90% of what I was trying to use OpenClaw for, but feels far less likely to upload my entire drive to a malware site." > — Ethan Mollick Mollick's quote gets at the fundamental difference: the security model. With Dispatch, your files stay on your machine, sessions run sandboxed, and you approve destructive actions in real time. With OpenClaw, you are wiring an open-source agent directly into your accounts and machine yourself — enormously capable, and exactly as safe as your own configuration. Having alternatives is healthy — the space is evolving fast, and competition makes all the tools better. But for most users, the security model tips the balance. ## Four capabilities in one chapter Step back and look at what you have now that you did not have at the start of this chapter: 1. **Claude Code** — a terminal agent that sees your filesystem and runs commands 2. **An active skill** — your organizational knowledge, operational in every Claude Code session 3. **Remote Control** — your phone as a window into your full local environment 4. **Cloud sessions** — work that runs on Anthropic's infrastructure and hands off between devices with `/teleport` These things compound. The skill encodes your knowledge (Chapter 04). Claude Code makes that knowledge operational (this chapter). Remote Control means you do not need to be at your desk to use it. And the distillation pipeline is live — every conversation correction can flow back into CLAUDE.md or the skill itself, making the system smarter with each interaction.
Try This

From your phone via Remote Control, give Claude a real task. Not a test — something from your actual work this week. Cornwall Market: Sarah, standing in the stockroom, sends a message from her phone: "Summarize this week's invoices from Chen's Produce and flag anything unusual." Claude reads the files, applies the categorization skill, and responds with a summary — all while Sarah is counting inventory.

Notice how the relationship feels different from typing into a chat window. You are not operating a tool. You are delegating to a system that knows your rules, sees your data, and acts on your behalf.

In the next chapter, you will connect your actual data via MCP — turning Claude Code from a system that knows your rules into a system that can act on your live data. Then you will set up routines and channels to make the system always-on. The delegation is just beginning. ## Check your understanding Answer in your own words — write it down before opening the key. Your tutor grades against the criteria and generates fresh variants on retries. **Q1.** State the decision rule for Remote Control vs cloud sessions, give one task for each from your own work, and name what breaks if you choose wrong in each direction. **Q2.** What does /teleport move, and describe the specific situation where it is exactly the right tool. **Q3.** Why must corrections be persisted the moment they happen ("update the skill") — what is the cost of a correction that lives only in the conversation?
Answer key — attempt every question first ## Answer key ### Q1 **Model answer:** Remote Control when the work needs YOUR machine — local files, local MCP servers, your credentials; cloud sessions when the work just needs to get done and should survive your laptop closing. Wrong toward cloud: the job cannot reach the local data it needs. Wrong toward Remote Control: the laptop sleeps and the work dies with it. **Pass criteria:** correct rule with the machine-dependency criterion; both failure directions named ### Q2 **Model answer:** /teleport pulls a cloud session down to your terminal — the branch AND the full conversation context. Right moment: you started work on your phone or in the browser, it progressed, and now you are at your desk wanting to finish properly in your editor with full local tools. **Pass criteria:** names both the code and the conversation moving; scenario is cloud-to-local continuation ### Q3 **Model answer:** An unpersisted correction evaporates at session end, so the same mistake returns tomorrow — and every session after. Persisting it (skill or CLAUDE.md) converts a one-time fix into permanent system knowledge; that flow of corrections into the knowledge base is the distillation pipeline actually operating. **Pass criteria:** names session-scoped memory loss; repeated-mistake cost; persistence as the pipeline mechanism
**Next:** [Connect Your Data](/mastering-claude/07-the-always-on-system/26-connect-your-data/)