When the chatbot stops being a chatbot
ChatGPT is about to look and work very differently. The tool that popularized conversational artificial intelligence is preparing to stop being a box where you ask questions and get answers, to become something more ambitious. OpenAI is rolling out its largest redesign since the chatbot launched, a transformation internally codenamed “Aria” that pushes it toward what the industry calls a “superapp”: a single interface that handles many different jobs instead of just one.
The content of that redesign reveals the direction. OpenAI is embedding AI agents — software that takes actions on the user’s behalf, not just answers questions — a coding tool called Codex, image generation and a growing list of third-party app integrations directly into ChatGPT’s main interface. The difference between answering and acting is the heart of the change: a chatbot replies; an agent books a flight, writes code or manages a multi-step task. The superapp aspires to the latter.
The scale it starts from explains the ambition. ChatGPT serves more than 900 million weekly active users and surpassed 50 million consumer subscribers, according to the company itself. Few platforms in history reached that mass of users so fast. Turning that audience into something more than a question-and-answer public — into users who delegate tasks and, above all, pay for it — is the bet that defines the company’s next phase.
The pivot that matters: from consumer to enterprise
Behind the interface change lies a deeper strategic mutation, and it is the one that really drives the decision. The company is reorienting its focus from the individual user toward the corporate client. The changes are part of a broader reorganization at OpenAI, which is shifting resources to target lucrative enterprise clients and intensify competition with rival Anthropic, according to the report citing more than a dozen current and former employees. The mass consumer public gave fame and scale; the enterprise business promises revenue.
The numbers explain why the pendulum swings toward businesses. Most of those who use Codex pay for it, and enterprise clients account for roughly 40 percent of OpenAI’s revenue, a share expected to rise to 50 percent by year’s end. The free user generates scale and data, but the business client generates cash. Reorienting the product toward tools companies deploy across their workforce is following the money, not just the trend.
That pivot brings OpenAI closer to its main competitor’s strategy. The move aligns OpenAI’s strategy more closely with that of Anthropic, which focuses on developing products for businesses; both companies are aiming for a listing and racing to reach public markets. It is a telling data point about the state of the sector: the company that defined consumer AI is moving toward the enterprise terrain its rival cultivated from the start. The convergence of strategies suggests where, today, the money in AI is.
The race to go public
The backdrop of the whole redesign is financial, and of unusual magnitude. OpenAI has reportedly filed confidentially to go public, with Goldman Sachs and Morgan Stanley advising, targeting a valuation of up to a trillion dollars by late 2026, which would make it potentially the largest US public listing on record. A trillion-dollar valuation for a company that was a research lab a few years ago gives the measure of the financial phenomenon surrounding AI.
The connection between the redesign and the listing is direct: the product is transformed to justify the valuation. The redesign seeks to generate more revenue through large businesses that would deploy the new ChatGPT across their workforce, which would help the company ahead of its listing. An enterprise superapp with recurring and growing revenue is a far more attractive story for public markets than a consumer chatbot, however massive. The product is redesigned, in part, with investors in mind as much as users.
The caution, however, lies in the timing. CEO Sam Altman has said the company is not focused on timing and will go public when it makes sense. The agency that reported the filing preparation could not immediately verify all the details of the plan. These figures — the trillion dollars, the dates — are best read as press reports about plans in progress, not facts confirmed by the company. The IPO race is real; its exact terms, still moving.
What “superapp” means and why the term matters
The superapp concept is not new, and understanding its origin helps size up the bet. In Asia, apps like WeChat proved that a single platform can concentrate messaging, payments, commerce and services, becoming indispensable in the daily life of hundreds of millions. The aspiration to turn ChatGPT into a superapp is to bring that model to the AI terrain: that the interface of artificial intelligence be the entry point to multiple services, not one tool among many.
The mechanism to achieve it is third-party integrations, and there is a data point that qualifies the scope. OpenAI’s new app integrations — including Booking.com and Canva — are not available for now in the United Kingdom, the European Economic Area and Switzerland, regions listed as “coming soon” with no committed date. The superapp rollout is uneven by geography: European users will receive the interface redesign, but the integrations that make it a true superapp could take months. The regulatory fragmentation of the digital world is reflected in which features reach each region and when.
That asymmetry is no accident: it responds to the weight of regulation. Europe, with its data-protection and competition framework, imposes conditions that delay the rollout of features integrating third-party data. The result is that the same platform offers different experiences depending on where the user is, and that the promise of the universal “superapp” collides with the reality of an unevenly regulated world. What in one region is an assistant that books and buys for the user is, in another, for now, just a prettier interface.
The battle of interfaces: against Copilot and against the desktop
The redesign does not happen in a competitive vacuum, but in a war to be the dominant interface of productivity. The redesign sets up a direct fight with Microsoft Copilot, the assistant the software giant integrated into its office applications. The underlying question is who will be the entry point to digital work: the assistant built into the word processor and spreadsheet, or the standalone superapp that aspires to replace both. It is a dispute over where people start their workday.
The incorporation of Codex at the product’s core is the most revealing piece of that strategy. The changes will give greater prominence and resources to Codex, OpenAI’s software engineering agent, capable of writing, fixing, explaining and executing code. Programming is the use case where AI agents have shown the most tangible value and where clients pay with the least resistance. Betting on Codex is betting on the segment that already proved willing to pay, and building the superapp around its most profitable capability.
The enterprise rollout already shows how the agent will be monetized. ChatGPT Enterprise added workspace agents for shared workflows, with new admin controls and activity visibility, and extended the free period for those agents until 6 July 2026, when credit-based pricing will begin. The business model takes shape: agents companies deploy across teams, with centralized administration and pay-per-use. The transition from free tool to paid enterprise service is not an abstract promise, but a calendar with dates and rates.
What the redesign says about the product’s maturity
Beyond the new features, the product’s recent evolution reveals an industry moving from growth to consolidation. One symptom is the cleanup of the model catalog. OpenAI will retire the o3 model from ChatGPT on 26 August 2026 after a 90-day sunset period, and GPT-4.5 on 27 June 2026 after 30 days, to better serve its newer, most capable models. Retiring old models is a sign of a maturing offering: instead of accumulating versions, the company concentrates resources on a few frontier lines. It is the logic of a product that stops experimenting and starts optimizing.
The appearance of safety and control features indicates the same maturity. ChatGPT introduced a “lockdown mode” that restricts network-enabled capabilities — live web browsing, deep research, agent mode, downloads — and a “trusted contact” feature to notify someone in serious self-harm-related safety concerns. That the product incorporates brakes — safety limits, admin controls, safeguards for vulnerable users — shows it is no longer an experiment, but an infrastructure that must answer for its effects. Containment features arrive when a tool becomes too central not to have them.
There is also a change in the revenue model worth noting for its precedent. ChatGPT began rolling out ads for free and Go plan users in the United Kingdom, while paid plans remain ad-free. The introduction of advertising in the free plan is a step the internet industry knows well: monetizing the non-paying user by showing them ads. It is another sign that the phase of mass acquisition without monetization is ending, and that the pressure to generate revenue — the same that drives the enterprise pivot and the IPO race — extends also to the consumer user. Unlimited free access was a stage, not a destination.
What the agent changes from the chatbot
The conceptual difference between a chatbot and an agent is the key to the whole transformation, and it deserves precision. A chatbot operates within the conversation: it answers, suggests, drafts, but the user executes. An agent crosses that boundary: it acts in the world, clicks, fills forms, chains steps. OpenAI believes AI agents that can perform tasks on users’ behalf will become more valuable than traditional chatbots. That belief is the thesis underpinning the whole redesign.
The change has implications beyond convenience. An agent acting for the user needs permissions, access to accounts and data, and the ability to make decisions with real consequences. That multiplies both utility and risk: the same autonomy that lets an agent book a trip can, miscalibrated, execute an unwanted action. The leap from chatbot to agent is also a leap in the surface of responsibility, and it forces thinking about governance, oversight and limits that a simple answering assistant did not require.
For the user and the company in the region, that nuance matters when assessing adoption. An agent integrated into the workflow promises real time savings, but it demands trust that it will act within the intended limits. The practical question is not whether the superapp is impressive — it will be — but whether the organization adopting it has the governance to delegate actions, not just answers, to autonomous software. The lesson from this same year’s AI adoption data applies: the tool works; what decides the outcome is the framework around it.
It is also worth situating the scale of the phenomenon in its historical context to understand why the bet is so aggressive. Few companies have gone from research lab to aspirant to the largest public listing in history in so few years, and that speed has a flip side: the pressure to turn a gigantic audience into sustainable revenue before competition erodes the advantage. The introduction of advertising, the pivot to the enterprise client and the redesign toward the superapp are three faces of the same financial urgency. They are not independent moves, but coordinated pieces of a strategy that seeks to prove, before the listing, that the company can monetize its scale. The product the user will see is, in part, the argument investors will see.
That dynamic poses a tension the user would do well to recognize. A company that redesigns its product with a market valuation in mind optimizes for revenue growth, not necessarily for the best individual user experience. The two can coincide, but they do not always: advertising in the free plan, the push toward paid features and the prioritization of enterprise clients reflect where the economic incentive lies, which is not identical to the interest of the user who just wants a useful, free tool. Understanding that partial misalignment is part of using these platforms with judgment.
The balance of the mutation
ChatGPT’s transformation into a superapp is the clearest signal of where the AI industry is moving in 2026: from free mass consumption to paid enterprise business, from the chatbot that answers to the agent that acts, and from the productivity tool to the platform that aspires to concentrate everything. Behind the redesign lies a financial logic — the race for a listing of historic magnitude — and a strategic convergence among the main players toward the corporate client.
The verdict, read with caution, is one of transition rather than completed revolution. The valuation figures and listing timelines are press reports about plans in progress, not certainties; the superapp rollout is uneven by geography and regulation; and the leap from chatbot to agent, however promising, shifts to organizations the burden of governing an autonomy they did not previously have to manage. For Latin America, which adopts these tools mostly as a client and not as a creator, the mutation is both opportunity and warning: the superapp will bring more capability in a single interface, but also more dependence on platforms whose rules, prices and availability are decided outside the region. The AI of 2026 no longer wants only to converse; it wants to act, charge and list. It is wise to understand that pivot before delegating the work to it, because the tool that today seems free and neutral responds, in reality, to a financial strategy the user did not design but that shapes what they receive.