OpenAI is moving its Codex coding agent from the desktop into developers’ pockets, intensifying a battle over how — and where — AI writes software.

On May 13, CEO Sam Altman hinted at broader ambitions for ChatGPT, asking on X, “would you call it a superapp?”Sam Altman (sama)¹ The next day, OpenAI announced that Codex, its AI coding tool that can write code and control apps on a user’s computer, would be accessible from the ChatGPT mobile app on iOS and Android.¹

Early coverage emphasized how the update lets users “monitor and manage their development workflows remotely,” seeing Codex live environments on any device where it’s running and working across threads, reviewing outputs, approving commands, and starting new tasks from their phones.² OpenAI stressed that files, credentials and permissions stay on the local machine while screenshots, terminal output and test results stream back to mobile in real time.¹

Business publications framed the feature as a quality-of-life upgrade for “open-laptop walkers” who leave machines running Codex in public spaces; now they can “remotely control their Codex AI coding app while on the go” rather than physically carrying a live laptop everywhere.³

Strategically, Axios noted that mobile access is part of a broader push to make Codex “cheaper and easier to use” as OpenAI competes with Anthropic’s Claude Code, including subsidies like two free months of Codex usage for companies that switch over.⁴ Anthropic already offers a similar mobile-friendly “remote control” for Claude Code, underscoring a pattern in which Anthropic raises prices amid demand while OpenAI cuts or offsets costs to grab market share.⁴

OpenAI leaders pitched the shift as a step toward ubiquitous agents. President Greg Brockman called it “a huge step forward for universal usage of agents”² and later said that “using codex from the ChatGPT app is such a freeing experience,” highlighting how developers can now “build things from your phone” and “develop from anywhere, anytime.”³ At the same time, analysts warn that approving autonomous agents from a small screen while multitasking could increase the risk of errors, even as AI labs race to make that workflow the new normal.⁴