OpenAI put Codex in the ChatGPT mobile app today:

Codex is now in the ChatGPT mobile app

I had to read that twice because it is not the “Codex on mobile” I already had in my head.

The old mobile thing was cloud Codex wired to GitHub projects. Useful, but a separate world from whatever I had open on my desk. The new one connects to a machine where Codex is already running, a laptop, a Mac mini, a devbox, or a managed remote box, and reflects the live state back to my phone: active threads, approvals, plugins, project context. I can start a turn on desktop, approve the next step from a coffee shop, or nudge the task from bed without standing up.

That last part is uncomfortably specific because that has been my actual problem for months.

I have been running Remodex, an open-source remote-control app for Codex. Its README calls it a “local-first open-source bridge + iOS app.” It does a lot:

  • end-to-end encrypted pairing between iPhone and Mac
  • steering an active Codex run without restarting it
  • follow-up prompts mid-turn
  • notifications when a turn finishes or gets stuck
  • git actions from the phone
  • reasoning controls
  • live streaming while Codex runs on the Mac
  • shared thread history with the Mac

I built it from source and ran it myself for security and privacy reasons. When it worked, it was great.

When it did not, it was annoying in the way only a peer-to-peer bridge can be annoying. I had to keep a local daemon up. Sometimes I would walk a block from my desk, the connection would time out, and I would be left with exactly as much remote control as I deserved: none. The daemon also liked to silently try to reconnect in the background, forever. None of this was a disaster. Just enough small paper cuts that I noticed them every time.

Some of that is fair. Remodex was a small project filling a gap that should not have existed. It had to invent a bridge because the platform owner had not built one yet.

Now the platform owner has.

Claude Code already had Remote Control, and it is the closest comparison. The Claude docs say Remote Control runs on your machine, so local MCP servers, tools, and project configuration stay available. Right idea. In day-to-day use it still felt like a first-generation remote window into a local session. The docs spell out a few practical limits too: one remote session per interactive process, the local process has to keep running, and a network drop of roughly ten minutes can time the session out.

Codex mobile reads more like an official product growing around the local agent, not a remote window bolted to the side. Approvals, plugins, project context, active threads, dual streaming to desktop and phone, these are not extras. They are the workflow.

So I probably do not need Remodex anymore.

I feel weirdly sad about that.

Someone spent months building a real tool, probably with the same hope every small developer has: there is a painful gap, I can solve it, maybe that becomes a product. Then one launch from the platform owner absorbs the gap.

This is good for me as a user. I want the official thing to be excellent.

It is rough for the people building in the empty space around these platforms. The empty space can disappear overnight.

Figma made it less theoretical. Anthropic launched Claude Design on April 17, 2026 as a research preview for prototypes, slides, one-pagers, and other visual work. Figma closed down 6.89% that day, from $20.32 to $18.92. The Motley Fool later said Figma finished April down 16%, with Claude Design as one of the pressure points investors reacted to.

I am not claiming Claude Design replaces Figma. Markets overreact. Designers also do a lot more than spit out a pretty prototype from a prompt, a sentence I am annoyed to be writing in 2026.

The fear just arrives faster now. A frontier model company can step into your category with distribution, compute, a familiar chat surface, and enough ambition to make investors reprice the target before most users have even tried the new thing.

Today I felt that with Remodex. Tomorrow it could be Msty. Or your app. Or some tiny tool that is actually good, built by someone who spotted a real gap before the platform owner did.

I do not have a clean rule. “Do not build near platforms” is useless advice, almost every useful piece of software lives near some platform.

So a narrower one: if a feature only exists because the provider has not built the obvious missing button yet, treat it as temporary. Use it. Maybe even sell it. Just do not confuse it with a moat.

The thing has to keep being useful after the obvious button ships.

I should probably tape that to my monitor. Then watch Codex read it from my phone.