Claude beyond the terminal: three paths

1. Remote Control

The simplest option.

Claude still runs locally on your computer, but you can control the same session from the web or your phone.

In a running Claude Code session, run:

/remote-control

Flow:

phone / web
-> Remote Control
-> same Claude Code session
-> local repository

You do not move the repository.

You do not create a new workflow.

You simply stop being tied to the terminal.

You can start a larger task:

investigate failing CI
-> fix
-> verification

walk away from the computer, and later enter the same session from your phone.

If Claude needs clarification, you can answer.

If you see it went in the wrong direction, you can redirect it.

If it finished a stage, you can give the next instruction.

So you still have:

developer <-> agent

but the interaction is no longer tied to the terminal.

2. Claude Code Cloud Session

Remote Control still has one dependency:

Claude performs the work on your computer.

If you want to remove that limitation as well, you can run Claude Code as a cloud session.

phone / web
-> Cloud Session
-> remote environment
-> repository

You take exactly the same kind of task that worked locally before:

review PR

or:

investigate failing CI
-> fix
-> verification

and run it remotely.

Your laptop is no longer the execution environment.

But it is still an interactive session.

You can return to it from phone or web, see what is happening and keep talking to Claude:

agent works
-> you check result
-> you clarify
-> agent continues

So cloud session removes the dependency on the laptop, but does not remove the human from the interaction.

It is still collaboration:

developer <-> agent

only the agent works remotely.

3. Managed Agents

Managed Agents solve a slightly different problem.

It is no longer about:

I want to talk to Claude more conveniently.

It is about:

I have a well-defined task and I want the agent to execute it as part of the SDLC.

Example:

CI failed
-> Managed Agent
-> investigate
-> fix
-> verification
-> result

Or:

PR opened
-> Managed Agent
-> review
-> findings

Here you are not rebuilding Claude's whole way of working.

The repository is still the source of what we prepared earlier:

repo
+-- CLAUDE.md
+-- .claude/
|   +-- skills/
|   +-- agents/
+-- MCP config
+-- code

Managed Agents mainly give you a managed runtime where this setup can run without a local developer session.

Anthropic provides a CLI for this:

ant

which you use to create and deploy a Managed Agent and the environment it runs in.

The most important change looks like this:

before:

developer
-> runs Claude

now:

SDLC event
-> runs Managed Agent

The agent receives the task and performs the agent loop itself:

task
-> Claude decides what it needs
-> tool call
-> result
-> next decision
-> next tool
-> ...
-> task finished

You do not have to guide it through every step.

This is an important difference from the previous two variants.

Remote Control and Cloud Session are still primarily interactive:

developer
<-> Claude

Managed Agent can act as part of the SDLC:

SDLC event
-> Managed Agent
-> agent loop
-> result

Implement this today

Do not build three new workflows.

Take one you already have.

For example:

investigate failing CI

Remote Control

Run it locally and take over the session from your phone with:

/remote-control

Cloud Session

Run the same task as a remote Claude Code session.

Managed Agent

If the task is already well-defined and repeatable, run the same use case as a Managed Agent and connect it to a specific SDLC event.

Level complete

The level is complete if you can run at least one earlier workflow without sitting at the local terminal, remotely or as an automatically triggered Managed Agent.

At earlier levels, we improved the way Claude performs work.

Here we also change where that work runs and how much continuous developer involvement it requires.

Materials

• Remote Control in Claude Code
• Claude Code cloud sessions
• Managed Agents quickstart