OpenAI Superapp: ChatGPT Merges 3 Apps After 75% Agent Loss

Three products, one desktop app, and a public confession that OpenAI spread itself too thin. ChatGPT Agent mode — once celebrated as the future of AI automation — lost 75% of its paying users in just months. Now new Applications CEO Fidji Simo is cleaning house, and the trigger was Anthropic's Claude eating into OpenAI's business customers.

The Wall Street Journal first reported on March 19, 2026 that OpenAI will merge ChatGPT, the Codex coding platform (a standalone coding tool launched in February 2026), and the Atlas browser into a single "superapp" for macOS and Windows. If you use any of these tools today, your desktop experience is about to change significantly.

ChatGPT Agent Mode: The 75% Collapse That Triggered Everything

When OpenAI launched ChatGPT Agent mode — a feature that lets AI autonomously browse the web, run code, and complete multi-step tasks without you clicking every step — it looked like a hit. At peak, 4 million weekly active paying subscribers were using it. That number then fell below 1 million: a 75% drop in just months.

The culprit wasn't bugs or pricing. According to The Decoder's internal reporting, users simply didn't understand what Agent mode was for. They opened it, weren't sure what task to give it, and left. This is a product-market fit failure (when a product's capability doesn't match a clear need in users' minds) in textbook form.

• Peak users: 4 million weekly active paying subscribers
• After decline: Under 1 million weekly active paying subscribers
• Drop rate: 75% attrition in months post-launch
• Root cause: Users didn't know what the feature was actually for
• Compounding issues: Speed problems, reliability failures, and cybersecurity concerns also hurt adoption

This is the mess Simo inherited. Her diagnosis was blunt: "We realised we were spreading our efforts across too many apps and stacks, and that we need to simplify our efforts. That fragmentation has been slowing us down and making it harder to hit the quality bar we want."

Three Apps Becoming One: The OpenAI Superapp Blueprint

Over just five months, OpenAI launched three separate desktop products in rapid succession: Atlas (an AI-native browser with a built-in ChatGPT sidebar and browser memory, launched October 21, 2025), then Codex (a standalone coding assistant, launched February 2, 2026). All while maintaining ChatGPT as the primary product. The result was three overlapping apps with no clear hierarchy.

All three are now being merged. The internal rollout plan, per reported memos, goes in stages:

1. Stage 1: Codex gets expanded productivity tools beyond just coding
2. Stage 2: ChatGPT is folded into Codex as the unified interface
3. Stage 3: Atlas browser is absorbed last, completing the superapp

The finished product will be described as "agentic" — meaning the AI can act autonomously (on its own, without you clicking every step) to handle tasks on your computer: writing code, analyzing data, browsing the web, and managing files, all from one desktop interface. Target platforms are macOS and Windows. No official launch date has been set — the timeline is described as "over the coming months."

One important carve-out: the mobile ChatGPT app (iOS and Android) stays as a separate standalone application. The superapp strategy is desktop-first, aimed squarely at developers, analysts, and enterprise workers — exactly the users Anthropic has been winning.

The Anthropic Effect: How Claude Code Triggered OpenAI's Pivot

Simo explicitly named Anthropic's enterprise momentum as a "wake-up call." Two specific products reportedly spooked OpenAI's leadership into action:

• Claude Code — an AI coding agent (a tool that can independently write, test, and fix software without a human guiding every step) that has converted enterprise software teams who previously relied on ChatGPT
• Claude Cowork — Anthropic's enterprise collaboration product, targeting the business teams that OpenAI needs as paying customers to sustain its valuation

While OpenAI chased consumer buzz with Sora (AI video generation), partnered with designer Jony Ive on an AI hardware device, and launched Atlas as a standalone browser, Anthropic kept its focus narrow: coding tools and enterprise customers. It worked. And now OpenAI is reorganizing to fight on exactly that turf.

Simo's internal memo said it plainly: "We cannot miss this moment because we are distracted by side quests." The side quests — Sora, the Ive hardware project, standalone Atlas — are being deprioritized. The main quest is enterprise productivity through a unified, coding-first desktop app.

What OpenAI Is Shelving to Focus on AI Automation

Consolidation always has casualties. Products that received significant investment and public attention are now being wound down or put on pause:

• Sora: OpenAI's AI video generation tool — a major PR headline at launch — is being deprioritized internally
• Standalone Atlas browser: Launched just 5 months ago as its own app, now being absorbed and discontinued as an independent product
• Jony Ive hardware partnership: The collaboration with Apple's legendary former design chief on a dedicated AI device is being scaled back

Sam Altman and Mark Chen (OpenAI's Head of Research) are evaluating which other product lines to cut. President Greg Brockman is temporarily co-leading the restructuring alongside Simo. The honest read: OpenAI launched too many products too fast, and is now paying the organizational cost of that velocity.

Should You Be Excited — Or Cautious?

The superapp concept has real appeal. Instead of switching between a chat window, a separate browser, and a code editor, everything lives in one AI-powered desktop environment. For developers, analysts, writers, and business users, that kind of unified workflow eliminates real daily friction.

But here are four reasons to temper the enthusiasm:

• Agent mode already failed once. The same autonomous AI capability promised in the superapp was already available — and fell from 4M to under 1M users. Bundling it into a larger app doesn't automatically solve the core confusion about what agentic AI is actually for.
• No launch date. "Over the coming months" is not a timeline. This could ship in Q3 2026 or Q1 2027 — or later.
• Desktop-only. If your primary ChatGPT use is on mobile, this change doesn't affect you — and the desktop-mobile split will remain.
• Fierce competition at the destination. Microsoft Copilot (embedded in Office 365 for 300 million users), Google Workspace AI, and Anthropic's Claude Cowork are all competing for the same enterprise customers OpenAI is targeting.

You can track OpenAI's official product updates at openai.com. If you're comparing Claude Code vs Codex for your workflow right now — before the superapp ships — our AI coding tools guide breaks down which one fits your actual use case in plain English.

Related Content — Get Started | Guides | More News