The reminder: what we had sensed six weeks earlier
In our article of 24 April on the release of GPT-5.5, we wrote this prescient sentence: "If OpenAI succeeds, the question will no longer be 'which AI is the best', but 'which AI is best integrated into my daily life'."
This user lock-in strategy can now be seen unfolding in black and white. On 19 March 2026, OpenAI made its plan official: to merge ChatGPT, Codex and the Atlas browser into a single desktop super-app. One application, one ecosystem, one gateway to all your digital work. And incidentally, one company that sees everything you do.
What exactly is the OpenAI super-app?
Three products that existed separately are to be merged:
- ChatGPT remains the conversational core. It is the interface that 900 million people use every week to ask questions, write content or analyse documents.
- Codex is the coding agent, launched on macOS in February 2026 and then on Windows in March. It already has 2 million weekly users, with growth of 70% per month. It codes, debugs and executes complex tasks autonomously.
- Atlas is the GPT-5-powered web browser, in beta since October 2025. It does not merely display pages: it understands them, summarises them, and can carry out actions on your behalf (booking, filling in forms, doing research).
The idea behind the super-app is to make them work together in a single window. You read an article in Atlas, you ask ChatGPT to summarise it, you ask Codex to turn it into a Python script that automates part of the process, all without ever switching applications. On paper, it is extremely practical.
The internal memo from Fidji Simo, CEO of Applications at OpenAI, is very clear on the reason: "We realised we were spreading our efforts across too many applications and platforms, and that we need to simplify our efforts. This fragmentation has slowed us down and complicated our ability to reach the level of quality we are aiming for." But between the lines, there is another reason: Anthropic is taking massive enterprise market share with Claude Code and Cowork. OpenAI can no longer afford to be scattered.
The real inspiring model: Apple
To understand what is really at stake, you have to look at Apple's history over the past 20 years. Steve Jobs spent his life building an ecosystem in which every Apple product works better with the other Apple products. iMessage works better between iPhones. AirDrop only works between Apple devices. iCloud synchronises everything, but only between Apple devices. The result: in 2026, switching from an iPhone to an Android is no longer a question of price or quality, it has become psychologically costly. You lose your synchronised photos, your blue messages with your friends, your Apple Watch, your configured Apple Pay.
This strategy has a name in economics: switching costs. The higher those costs, the less able the user is to leave. And the less able they are to leave, the more the company can raise its prices without losing customers.
OpenAI is applying exactly the same model, but to AI. Imagine two years from now: your ChatGPT conversation history is rich with thousands of contextualised exchanges. Your Atlas browser knows all your favourite sites, your passwords, your preferences. Codex has learned your coding style and your current projects. Your company data is indexed in the system. The day Claude or Gemini releases a model that is 30% better, will you leave? Probably not. The psychological cost will be too high.
The figures that show what is really at stake
OpenAI is not making this change by chance. The financial figures are dizzying:
| Metric | 2026 figure |
|---|---|
| Weekly ChatGPT users | 900 million (x2.25 in a year) |
| OpenAI valuation | $852 billion |
| Recent funding round (March 2026) | $122 billion |
| Share of revenue coming from businesses | 40% (target 50% by the end of 2026) |
| Businesses using OpenAI agents | More than 1 million |
The IPO is planned for the end of 2026 or the start of 2027. To pull it off, OpenAI has to prove to Wall Street that it can hold on to its users for the long term. And there is only one way to prove that: lock them in.
The problem: what OpenAI sees when you use the super-app
Here is the part nobody says out loud. When ChatGPT, Codex and Atlas were separate applications, each one saw a part of your digital life. Now, a single company will see the whole puzzle:
- Atlas sees what you search for on the web, at what times, on what subjects
- Codex sees what you work on, your code, your projects, your trade secrets
- ChatGPT sees your intimate questions, your doubts, your mental health, your personal life
Brought together in a single application, these three streams give OpenAI a near-complete picture of your daily intellectual activity. No other tech tool (not even Google, not even Facebook) has ever had access to this level of knowledge about its users. And OpenAI can now correlate in real time: what you search for (Atlas), what you create (Codex), and why you do it (ChatGPT).
If you search for a specific API error in Atlas, then fix it in Codex, OpenAI knows exactly which part of your code was broken and how you repaired it. For a startup or a developer working on sensitive intellectual property, this is a potential continuous leak of trade secrets into OpenAI's training database. Several American consultancy firms already advise defence companies to stop using the super-app for this reason.
Why Apple blocks the super-app on iPhone
Here is a fascinating detail that few media outlets pick up on. The OpenAI super-app will be available only on Windows and macOS. Apple blocks OpenAI on iOS, thanks to a specific clause in its rules: "guideline 4.7" of the App Store, which prohibits applications that run agentic code not reviewed by Apple.
Apple's message is very clear: Apple is itself a locked ecosystem, and has no intention of letting another locked ecosystem take root inside its own. It is a territorial war between two identical economic models. And the end user has no say in it.
The same strategy at the other giants
OpenAI is not alone in applying this logic. All the major AI players are converging on the same lock-in model, each in their own way:
- Microsoft Copilot is integrated into Word, Excel, Teams, Outlook and Edge. If your company uses Microsoft 365, leaving Copilot means leaving the Microsoft suite. A weighty decision.
- Google Gemini will be available natively in Gmail, Docs, Drive and Search from the Google I/O event on 19 May 2026. The same logic: lock in via the existing ecosystem.
- Anthropic Claude plays a different card with Claude Code and MCP connectors. It is more open technically, but attachment is created through the quality of reasoning, which brings developers back. It is lock-in through talent, not through friction.
- Apple Intelligence will never need to lock its users in: they already are, via the iPhone-Mac ecosystem.
What this changes concretely for you
If you are a private ChatGPT user, in the immediate term you will probably love the super-app. It is more convenient, smoother, more powerful. The problem will come in 18 to 24 months, when you may want to switch to a more capable competitor. You will then discover that your entire history, all your settings, all your habits are locked inside the ecosystem.
If you are a developer, the risk is more immediate. Building workflows in Codex means starting to depend on infrastructure you do not control. Migrating in two years could mean hundreds of hours of rework. The same question arose for developers who adopted Heroku en masse between 2010 and 2015, before Salesforce bought the company and raised its prices.
If you are in an enterprise, the calculation is strategic. Adopting the super-app means gaining immediate productivity, but it also means handing a single foreign (and private) company access to all your business data. Several cybersecurity experts now recommend maintaining a "decoupled stack": an independent browser, a local IDE, a local AI assistant where possible. More complex to set up, but far more resilient to lock-in.
The real problem is not OpenAI, it is the model
It would be easy to cast OpenAI as the villain of the story. But that would miss the real issue. Every tech company has a massive economic incentive to lock in its users. It is mathematical: a user who cannot leave pays more, for longer, without negotiating.
The real problem is the absence of countervailing powers. There is currently no regulation that would impose the portability of AI data between providers. No framework that would oblige OpenAI to let a user leave with their history, their settings and their custom agents for Claude or Gemini. Compare that with the banking sector, where account number portability has been mandatory in Europe since 2017, or with mobile telephony, where number portability has been guaranteed since 2003.
As long as these rules do not apply to AI, lock-in is the logical outcome of the market. And the sooner you are aware of it, the better you can limit its effects in your own technical choices.