Fixing Codex Remote Control: A Career Networking Lesson in AI Collaboration

When Your AI Tool Breaks, Let It Diagnose Itself

I've spent the last few years building a career around product development, and lately that means working with AI tools like Codex. But last week, I hit a wall: I couldn't get Codex remote control to connect my Mac to my phone. I tried everything—re-logging in, restarting, checking my account, even upgrading the app. Nothing worked. The error message was always the same: "Unable to enable remote control. Please try again."

Then it hit me: why not let ChatGPT, the very tool that was failing, figure out its own problem? That decision turned into a fascinating journey that taught me something valuable about career networking in the AI age.

The Setup: A Digital Employee That Follows You

My goal was simple. I wanted to run Codex on my Mac, have it work on code, then walk away and still monitor progress from my phone. I could send new commands, check on tasks, and keep the development moving even when I wasn't at my desk. For anyone who relies on AI coding agents, this is a game-changer. But the setup was anything but simple.

The phone side was easy. I opened ChatGPT, went to Settings, then Remote Control, and added a new connection. It asked me to confirm I was using the same account and workspace on both devices. I was, so that wasn't the issue. The real problem was on the Mac.

The Red Text That Wouldn't Go Away

On my Mac, I went to Settings, then Connections, then "Control This Mac." I clicked Allow, and instead of pairing, I got that dreaded red text. I tried restarting, logging out, re-pairing the phone—nothing helped. My first thought was that my account didn't have the right permissions, or maybe my personal workspace didn't support this feature. But after digging, I ruled both out.

Updating Codex Didn't Help

I noticed the desktop app had an update available. I was on version 26.803.61601, and there was a newer one, 26.810.50856. I installed it, hoping that would fix whatever bug was causing this. It didn't. The red text was still there, mocking me.

Looking at the Logs: The Clue That Changed Everything

At this point, I was ready to give up. But then I remembered that Codex keeps logs on macOS. I found them at ~/Library/Logs/com.openai.codex/YYYY/MM/DD. After searching for anything related to remote control, I found a line that caught my eye:

method=remoteControl/enable errorCode=null refresh_remote_control_started refresh_remote_control_completed nextConnectionCount=0 previousConnectionCount=0 creationFailureCount=0

This was telling. The remote control module was starting up fine, but it never actually made a connection. The error code was null, and the connection count stayed at zero. That meant the problem wasn't with the toggle itself—it was something deeper, probably in the network path.

Testing the Network: Direct vs. Proxy

Here's the thing: my Mac needs a proxy to reach ChatGPT. I checked my system proxy settings and found that HTTP, HTTPS, and SOCKS were all set to 127.0.0.1 with specific ports. That explained why ChatGPT worked fine and Codex could write code—they were using the proxy. But maybe the remote control feature wasn't.

To test this, I ran two quick experiments. First, I tried a direct curl to chatgpt.com with a 10-second timeout. It failed with "Failed to connect to chatgpt.com port 443: Timeout was reached." Then I tried the same curl but specified the HTTP proxy with --proxy http://127.0.0.1:33210. That time, I got HTTP/1.1 200 Connection established immediately. The difference was night and day. Direct connection: dead. Proxy: alive.

The Fix: Injecting Proxy Variables into Codex

That experiment gave me a strong hypothesis: Codex's remote control backend wasn't inheriting the system proxy. To test it, I quit Codex completely, then launched it from the terminal with the proxy environment variables set:

export HTTP_PROXY=http://127.0.0.1:33210
export HTTPS_PROXY=http://127.0.0.1:33210
export ALL_PROXY=socks5://127.0.0.1:33211
open -a "Codex"

I clicked Allow again, and this time, it worked. My phone connected to my Mac's Codex. The fix was confirmed. The system proxy was fine, the main app could use it, but the remote control's background connection wasn't picking it up automatically.

Making It Permanent with a Launcher

But typing those export commands every time was annoying. So I created a small AppleScript app that waits 8 seconds (to let my proxy software start), then launches Codex with those environment variables. I used osacompile to build it, then added it to my login items and kept it in the Dock. Now, whenever I need to restart Codex, I just close it and click the launcher. It's clean, reversible, and doesn't touch any system settings.

A New Career Networking Skill: Human-AI Collaboration

This whole experience taught me something that goes beyond fixing a technical glitch. It's about how we network with our own tools. In the past, if an app failed, I'd search forums or ask a colleague. But now, with AI assistants that are deeply integrated into our workflows, we have a new option: let the AI diagnose itself.

ChatGPT didn't get it right the first time. It made wrong assumptions, had to backtrack, and revised its approach based on my feedback. But because I kept feeding it real-time information—screenshots, logs, curl results—it kept narrowing down the possibilities until we found the root cause. That's not a one-shot Q&A; it's a collaborative debugging session.

This is a career skill that's becoming more valuable by the day. Being able to articulate a problem clearly, run experiments, and work with an AI to iterate toward a solution is like having a superpower. It's not about knowing all the answers; it's about knowing how to ask the right questions and how to use the tools you have.

So, the next time your AI tool misbehaves, don't just Google the error. Let the AI look at its own logs. You might be surprised at what it can find when you work together.

How to Tell If This Is Your Problem

If you're stuck with the same "Unable to enable remote control" error, here's how to know if it's a proxy issue:

• Your ChatGPT and Codex apps work fine, but remote control fails.
• You're using the same account and workspace on both devices.
• You've tried restarting, re-pairing, and updating.
• You need a proxy to access ChatGPT in the first place.

Then check your system proxy with scutil --proxy, and run the two curl tests I did. If direct fails and proxy works, that's your culprit.

The Bigger Picture

This whole episode reinforced my belief in the power of the "AI super individual." You don't need a big team or expensive infrastructure to solve complex problems. You just need to know how to define the problem, how to feed the AI the right information, and how to take responsibility for the outcome. That's the new career networking—not just connecting with people, but connecting with your AI tools in a way that multiplies your capabilities.

And honestly, the best part isn't that I can now control Codex from my phone. It's that I've learned a repeatable process for troubleshooting any AI tool. That's a skill that will pay off for years to come.