YouTube Summaries

← All summaries

OpenAI Merges Codex into ChatGPT

2026-07-11 Sat ⏱ 26 min t3dotgg

Theo reacts to OpenAI's overhaul of the Codex app, which has been merged into a unified ChatGPT interface (with a new "co-work"-style "work" mode alongside "codex" and "chat" sub-modes). While he credits the update with real technical improvements — better computer-use, faster performance, improved mobile reliability, unified plugins, and a Claude-Pro-handoff-style workflow — he argues OpenAI made a serious branding and strategic mistake by folding the beloved, dev-focused "Codex" identity into the broader ChatGPT consumer product. He traces Codex's rise, argues its appeal came precisely from not being ChatGPT or Claude Code, connects the move to OpenAI's apps-org turmoil (Fidji Simo's departure for health reasons), and closes by explaining why this reinforced his decision to build and maintain the open-source Codex alternative, T3 Chat/Code.

# Setup: Loving Codex, Then Losing It

Theo explains his long-standing affection for the Codex app (distinct from the many other things OpenAI has called "Codex": a 2022 code model, a 2025 CLI, a model variant, then the desktop app). He liked it enough to build his own open-source clone (T3 Code) as insurance against regressions. On the day of the video, OpenAI rebranded/merged the Codex app into a new "ChatGPT" overhaul — updating Codex actually installed ChatGPT with a Codex-shaped icon. He was tipped off in advance by OpenAI contacts hoping he'd help soften the reception, but felt that effort failed.

# What Changed: The New ChatGPT/Codex/Work Structure

The former three-pane structure (Home/Chat, Work/"co-work", Code) has been consolidated: "co-work" merged into the main app alongside "chat" and "codex" modes. Codex is now a mode inside ChatGPT rather than its own app; "work" largely mirrors codex functionality without adding much distinct value yet. Notably, "chat" — the original core ChatGPT feature — is relegated to a popup rather than a primary tab, which Theo predicts will annoy longtime ChatGPT users as much as the Codex change annoys developers. Project/folder setup also got worse (no longer able to attach multiple folders per project).

# What's Actually Good About the Update

Real improvements: unified plugins (PDFs, spreadsheets, presentations, Gmail, Slack, Notion, Drive), a "sites" beta competing with Lovable, major computer-use upgrades (multitab, enterprise auth, speed), in-app PR review/editing, Pro-to-Codex handoff, a persistent app icon option for Codex loyalists, notably reduced system resource/heat usage (suggesting it was already OpenAI's internal daily driver), better settings/model-picker UX, more reliable mobile support, improved documentation/SEO, programmatic tool calling, and better SSH video rendering.

# Why Codex's Brand Mattered

Theo argues Codex succeeded specifically because it wasn't ChatGPT or Claude Code — a distinct, developer-first brand with strong community identity (merch, meetups, viral goodwill). Codex's naming was already muddled across models/CLI/app, but the app itself started winning huge growth (5x/month), triggering Super Bowl ad spend, precisely as it began outperforming ChatGPT's growth trajectory.

# Why OpenAI Did It — and Why It Was a Mistake

He speculates this reflects internal pressure to redirect Codex's success toward ChatGPT (OpenAI's primary revenue driver) and org restructuring around apps/dev tooling, referencing Fidji Simo's (CEO of Applications) departure for health reasons as a symptom of deeper struggles aligning research, apps, chat, and dev-tool teams. He contends this was a strategic miscalculation: Codex users specifically didn't want ChatGPT or Claude Code, and folding Codex in strips away the developer identity and community allegiance that made it thrive — replacing "you gotta try Codex" with the far less compelling "you gotta try the Codex mode in the new ChatGPT app."

# Fallout and Personal Reaction

The change saps motivation both for users evangelizing Codex and for the team who joined OpenAI specifically to build it — Codex is now just a toggle. He worries developer-focused priorities will get deprioritized as ChatGPT features dominate the UI, though the underlying open-source app-server/CLI (which powers third-party tools like his T3 Code) hasn't been taken away yet. He closes reaffirming gratitude for having built T3 Code as an open-source hedge, since he now has no desire to use the merged experience, calling it "a sad way to kill a brand" well-loved in the developer community.