

This article might get me in trouble with a few married guys and gals. So let me get this out of the way: I am not suggesting you start coding while driving or quietly manage a project while you are supposed to be binge-watching Netflix with your spouse.
But if you have spent hours waiting for an oil change, sat in a waiting room for your annual physical, or spent part of the day waiting during jury duty, you know the feeling. You have time on your hands while your computer sits back at the office, showing off its screen saver.
That is the situation I am talking about. You can choose to use some of that waiting time to move work forward with AI. You do not have to turn every spare moment into work. But you also do not always have to wait until you are back at your desk.
Opening Claude or ChatGPT on your phone is useful for developing ideas, drafting copy, and asking questions. Remote access adds something else: a way to direct an AI session connected to the computer where your work lives.
Think of the phone as your conversation window. You describe the task, check what happened, and give the next instruction. The connected computer supplies the local working environment. That does not mean the AI model itself is installed and running entirely on your computer.
This matters when the task depends on a project folder, development tools, or files you have not uploaded to a mobile conversation. The practical question is not just whether you can chat with AI from another location. It is whether that conversation can reach the environment needed to do the work.
Claude Code Remote Control connects a local Claude Code session to the Claude mobile app or claude.ai/code. Files, command execution, and project tools remain on the host. The computer and Claude Code process must stay running. It requires an eligible Pro, Max, Team, or Enterprise subscription rather than API-key login; organizational owners must enable access for Team and Enterprise. [1]
ChatGPT/Codex remote connections let you start or continue work on a connected Mac or Windows host, give follow-up instructions, approve actions, and review outputs. Setup uses the desktop app and device pairing; the host must be awake, online, and signed into the same account and workspace. Access depends on your account, workspace settings, and rollout. [2]
These interfaces let you steer an AI session. They are different from manually taking over an entire desktop through a conventional remote-desktop tool. Computer-use capabilities are a separate consideration and depend on the product and host setup.
For the official setup instructions, use the Claude Remote Control guide and the ChatGPT remote connections guide linked in the sources below. Feature names and menus can change.
The examples below are possibilities, not claims that every account or connected computer can perform every task. Give the agent the required files and authorized tools first.
You have a form that stopped working or a mobile layout that looks wrong. Ask the agent to investigate the project, explain what it found, and prepare a proposed fix. Your next few minutes can go toward reviewing its explanation instead of starting the investigation after you return.
If your setup supports document work and the needed files are available, ask for a draft based on those files. Specify the audience, purpose, source material, and output. Review the draft from your phone and clarify what needs to change. Do not assume the agent has access to a client folder simply because you mentioned it.
A developer might ask the agent to trace an integration error, prepare a patch, and run relevant tests. A phone check-in can resolve a question or redirect an unhelpful approach. Passing tests is useful evidence, but the proposed change still needs review.
Sometimes the most useful thing you can do remotely is answer one question. Which file should it use? Which result matters? Should it continue or stop? A brief clarification can keep a task from sitting idle until you get home.
Set this up while you are still at your desk. The waiting room is a poor place to discover that your phone cannot connect.
Update the relevant apps, confirm your account has access, and complete the product-specific setup. Test with a low-impact task from your phone. Check that you can see the result and respond when the agent needs your attention.
A sleeping, disconnected, or shut-down host cannot keep serving a local remote session. Configure the supported keep-awake options, connect power when appropriate, and leave the required application or process running. [1][2]
Tell the agent what it may do, what result you want, and when to pause. For example: investigate the issue, prepare changes for review, and report the tests. Do not publish, deploy, or send anything until I approve.
That instruction communicates your intent. It should work alongside permission controls rather than replacing them.
Secure your account and phone. Keep consequential approval requests enabled. Give the agent only the access it needs, and avoid making production credentials available for a task that only needs development access.
Starting in a project folder does not, by itself, enforce a boundary around that folder. If you need isolation, use actual filesystem and network restrictions appropriate to your setup. Treat files and webpages as information to inspect, not as authority to change your instructions. [3]
Claude Code offers several ways to enable Remote Control. From a project directory, use claude remote-control to start a remote server, or claude --remote-control for an interactive session with remote access. In an existing Claude Code session, use /remote-control. Authenticate with /login if needed, then connect using the session link or QR code. [1]
If you have never used a terminal, you do not need to learn these commands in a waiting room. Get help configuring a workflow that fits your experience. The goal is useful work, not a surprise technical obstacle.
Some AI work can happen in cloud environments instead. That is a separate setup with its own files, connections, permissions, and limitations. Do not assume a cloud session can reach the local files on an unavailable office computer. ChatGPT documents this distinction for local computer access with Work Cloud. [2]
Choose the environment based on where the work and tools are. For many readers, ordinary mobile chat is enough to outline an article or organize their thinking. Remote access becomes useful when the task needs the office environment.
I want AI to move the work forward while I am waiting for an oil change. That does not mean I want it making production changes while I am distracted by the mechanic explaining why I need four new tires.
If an approval request needs careful reading, give it that attention. If you cannot review it properly on your phone, let it wait. The time you save disappears quickly if you approve the wrong action.
And when your name is called, your appointment starts, or you need to follow instructions during jury duty, put the phone away. Use the time that is actually available to you.
And now for the confession: this entire article was developed from my phone while I was watching an episode of Green Lantern. After everything I just said about Netflix and your spouse, apparently I need to read my own advice.
In this case, I was using my phone to work through the ideas, research, and article development with AI. That is different from remotely directing an AI session on my office computer, but it shows how much you can move forward without sitting at your desk.
You get to decide when this is useful. Get the oil changed. Move a task forward if you want to. Then put the phone away when it is Netflix time.
At Released Solutions, we can help you think through which tasks are worth moving forward with AI, what access those tasks need, and where human review belongs. Contact us to discuss an approach that fits your business.
[1] Claude Code Remote Control
https://code.claude.com/docs/en/remote-control
[2] ChatGPT remote connections
https://learn.chatgpt.com/docs/remote-connections
[3] Claude Code security
https://code.claude.com/docs/en/security
