OpenAI Unifies ChatGPT & Codex — Start Your Exit Plan

OpenAI announced on May 16 that it is collapsing ChatGPT, Codex, and its developer API into a single product organization under co-founder Greg Brockman — three days before Google I/O opened. Thibault Sottiaux, the engineer who built Codex into one of OpenAI’s fastest-growing products, now leads core product and platform across consumer, enterprise, and developer surfaces. This is not a product launch. It is an org-chart restructuring that tells you exactly where OpenAI’s roadmap pressure will land: on the agentic consumer super-app, at the cost of everything else.

TL;DR

• What: OpenAI merged ChatGPT, Codex, and the developer API into one team under Greg Brockman
• Why it matters: Every API deprecation, pricing change, and breaking update now flows from a team optimizing for a consumer super-app and an $852B IPO
• Risk: Developers building on OpenAI’s fragmented surfaces are exposed to IPO-cycle prioritization, not product-cycle stability
• Action: Treat OpenAI’s developer API as a managed-exit dependency — put a migration path on your roadmap today

What Happened

Greg Brockman wrote in an internal memo that OpenAI will “invest in a single agentic platform and merge ChatGPT and Codex into one unified agentic experience for all.” That language is doing a lot of work. “Single” means the three separate bets developers were tracking — the chat surface, the coding agent, and the raw API — are now one team with one mandate. Sottiaux, previously leading Codex alone, now owns the combined platform across consumer, enterprise, and developer.

The consolidation coincides with aggressive cuts to OpenAI’s non-core products. Sora, the video generation app, has been shut down entirely — it consumed vast computing resources relative to its revenue. Adult mode for ChatGPT was shelved. OpenAI for Science appears deprioritized. The internal framing, dating back to Sam Altman’s December “code red,” is explicit: anything not tied to the agentic platform roadmap is at risk.

Reports indicate the team is building a unified desktop experience that merges Codex, ChatGPT, and the Atlas research tool into a single agentic interface. No launch date has been announced, but this is where Sottiaux’s engineering resources are going. OpenAI is internally describing this as a “super app.”

If you are building on the Codex CLI, the Agents SDK, or the raw completions API as separate stable surfaces, those surfaces are now one team’s problem. Separate teams meant separate advocates. One team means one priority stack — and you are not the consumer paying $20/month.

Why This Matters

The timing is the tell. OpenAI announced the restructuring three days before Google I/O, where Gemini’s agentic coding tools were the headliner. The competitive pressure driving this consolidation is visible in the traffic numbers: ChatGPT’s share of global AI web traffic dropped from 86.7% to 64.5% in the twelve months ending January 2026. Gemini surged from 5.7% to 21.5% in the same period. Those are not rounding errors. That is a 22-point decline in market share at a company preparing to go public at an $852 billion valuation.

The competitive picture on the coding-agent side is equally uncomfortable for OpenAI. Cursor reached $2 billion in annualized revenue by February 2026 and is reportedly in talks to raise at a $50 billion valuation. Claude Code was generating over $2.5 billion in run-rate revenue by the same month. Both are building their agent platforms as purpose-built developer tools, not consumer super-apps that happen to also serve an API.

That distinction matters. When OpenAI’s developer API was its own team with its own leadership, API developers had at least one advocate whose job was to keep the API stable, well-documented, and competitive. Under the new structure, the API is a surface of a consumer product. The person deciding whether to deprecate an endpoint or change pricing semantics is the same person deciding how the ChatGPT desktop app should feel. These are fundamentally different optimization pressures.

The financial context makes this worse, not better. OpenAI projects $14 billion in losses for 2026 on roughly $25 billion in annualized revenue, with profitability not expected before the 2030s. A unified org is cleaner for an IPO filing — one product line, one revenue story, one executive owner. But what is clean for the S-1 is not clean for developers building on the platform. IPO-cycle decisions optimize for revenue growth narrative and margin story, not for API stability guarantees. If a breaking change to the API also improves the consumer product’s metrics, the consumer product wins. Every time.

The migration question is not “should I leave OpenAI?” It is “can I leave OpenAI in 30 days if I need to?” If the answer is no, your architecture has a single point of failure you did not choose.

There is also the Musk v. Altman trial to consider. Jury selection concluded in late April at Oakland federal court, with closing arguments on the liability phase wrapping up by mid-May and deliberations set to begin around May 19. Brockman is simultaneously leading OpenAI’s product strategy and serving as a central figure in a lawsuit that could unwind the nonprofit-to-profit conversion on which the entire IPO depends. Neither Anthropic nor Google has an analogous distraction pulling executive attention away from product decisions. This is not a scandal concern — it is an execution risk. The person running your platform’s future is in a courtroom.

The Take

I would be treating OpenAI’s developer API as a managed-exit dependency right now, not a foundation to build on. That sounds harsh — OpenAI’s models are still excellent, the API still works, and millions of developers still ship on it every day. But the structural incentives have shifted. Three separate teams meant three separate roadmaps, three separate advocates, and three separate reasons to keep the developer surface stable. One team means one priority stack, and that priority stack is a consumer super-app targeting an IPO at nearly a trillion dollars.

The practical move is not panic migration. It is abstraction. If you are on the Codex CLI or the Agents SDK, put a migration path on your roadmap today. That means an LLM routing layer — LiteLLM, your own adapter, whatever gives you the ability to swap providers in hours, not weeks. It means keeping your agentic logic decoupled from OpenAI-specific SDK primitives. It means accepting that the cost of portability is a small tax now, but the cost of lock-in could be an emergency rewrite later.

Cursor and Claude Code are building purpose-built developer tools with developer-first business models. OpenAI is building a consumer platform that happens to serve developers. When those incentives diverge — and they will — you want to be on the side of the table where the vendor needs you, not where the vendor needs your metrics for a roadshow slide.

The org chart does not lie. Read it accordingly.