The unexpected death of Codex
TL;DR
Codex is now just a mode inside ChatGPT, not a standalone app, and the rebrand erases a developer-focused identity that was growing 5x monthly.
Developers liked Codex because it wasn't ChatGPT, and the merger removes the very thing that attracted them in the first place.
The unified app has real improvements including better performance, cooler laptop thermals, unified plugins, and improved computer use with multitab browsing.
Fidji Simo's departure as CEO of Applications highlights organizational struggles in OpenAI's product structure, with health issues forcing her into an advisory role.
Theo's open-source T3 Code clone exists specifically because he didn't trust OpenAI not to ruin Codex, a hedge that paid off.
The Breakdown
A eulogy for Codex
Theo opens by declaring his affection for the Codex app, which he loved enough to build an open-source clone called T3 Code as a fallback. The Codex name had already become confusing, referring to OpenAI's first code model, a CLI tool, a specialized model variant, and finally a desktop app. Despite the branding chaos, the app itself had won over developers, including Theo.
The rebrand that erased Codex
OpenAI merged Codex into ChatGPT with the new 'ChatGPT Work' overhaul. When Theo clicked update on his Codex app, it transformed into ChatGPT. The only remaining trace of Codex is a small label and a buried setting to switch the dock icon back. ChatGPT now has three modes: Chat (demoted to a popup), Work (for non-dev tasks), and Codex (for code).
The actual improvements buried in the merge
Theo acknowledges real improvements: unified plugins for Gmail, Slack, and Notion; better computer use with multitab browsing and enterprise auth; a new Sites feature; and handoffs between Chat and Codex modes. Performance is noticeably better, his laptop runs cooler, and the model picker got a cleaner UI. These gains suggest OpenAI has been using this unified app internally while the standalone Codex app stagnated.
How Codex's brand got diluted over time
Codex branding was already a mess. The name referred to a model, a CLI, and an app, with no clear identity. Compare that to Claude Code, which everyone understands as the CLI. OpenAI started fixing this by killing the Codex-specific models and folding their capabilities into the main models, which was the right call. The app was the next step, and it had started building real brand equity with 5x monthly growth and even a Super Bowl ad.
Why developers loved Codex for what it wasn't
The problem is that Codex succeeded because it wasn't ChatGPT. Developers chose it specifically because it was built for them, by devs, with their needs in mind. People wore Codex merch, organized meetups, and switched from Claude Code. That allegiance is hard to rebuild when Codex becomes 'the Codex mode in the ChatGPT desktop app.' Theo's friend Rero nailed it: the pitch went from 'you have to try Codex' to a mouthful no one will say.
The organizational chaos behind the decision
Fidji Simo, the CEO of Applications at OpenAI, is moving to an advisory role due to serious health issues. Her departure isn't acrimonious, Sam Altman publicly wished her well, but it highlights organizational struggles in bridging research, apps, and developer tools. Theo suspects internal team shuffling and the desire to put great engineers on the cash-cow ChatGPT product drove the decision more than user needs.
T3 Code as the open-source insurance policy
Theo built T3 Code as a hedge against exactly this kind of scenario. He didn't trust OpenAI not to screw up Codex eventually, whether through performance issues, unwanted UI changes, or killing it entirely. Now he's grateful to have an open-source alternative where he can pin versions, modify behavior, and avoid features he doesn't want. About a quarter of T3 Code users are already running forks or patched versions.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.