OpenAI wants to be everything

GPT-5.5 just dropped, and OpenAI is not being subtle about what comes next. The company is merging ChatGPT, Codex, and its Atlas browser into a single desktop "super app," with president Greg Brockman calling it "setting the foundation for how we're going to do computer work going forward." Fidji Simo, OpenAI's CEO of Applications, told employees they'd been "spreading efforts across too many apps and stacks" and needed to simplify. The pitch is seductive: one app to rule them all. Chat, code, browse, deploy agents, all in one place. But the super app playbook has a terrible track record outside of Asia, and there are good reasons to think OpenAI's version won't be the exception.

The super app graveyard

The term "super app" gets thrown around like it's an inevitability, but the history is mostly failures. WeChat succeeded in China under conditions that don't exist anywhere else: a mobile-first population leapfrogging desktop computing, a regulatory environment that favored consolidation, limited competition from fragmented alternatives, and cultural norms around centralized digital ecosystems. Every Western attempt to replicate this has flopped. Elon Musk's grand vision for X as an "everything app" has gone nowhere. Google+ tried to be the connective tissue across all Google products and was shut down. Facebook's pivot to being a platform for everything from marketplace to dating to payments has resulted in a bloated app that people tolerate rather than love. Microsoft spent years trying to make Cortana the center of your computing life before quietly shelving it. Forrester published a report in 2023 titled "The Super App Window Has Closed," arguing that super apps like WeChat exploited first-mover advantages in a unique market and delivered the right utility at the right time to an audience that needed it. Those conditions are absent in the West. Cultural factors, from privacy expectations to regulatory frameworks to simple user behavior, work against consolidation. So why does OpenAI think it's different?

The argument for this time being different

The strongest case for OpenAI is that AI fundamentally changes the equation. Previous super apps tried to bundle existing services, essentially stapling a payments layer onto a messaging app. OpenAI is arguing that intelligence itself is the unifying layer. If the AI is smart enough, it doesn't matter whether you're coding, researching, or managing spreadsheets, the interface is just a conversation. GPT-5.5 lends some credibility to this. OpenAI describes it as "a new class of intelligence for real work," and the benchmarks back up real improvements in autonomous task completion. You can hand it a messy, multi-part task and it will plan, use tools, check its work, and keep going. Brockman highlighted a math professor who built an algebraic geometry app from a single prompt in 11 minutes. The enterprise angle matters too. OpenAI's blog post on the next phase of enterprise AI explicitly describes "a unified AI superapp: one place where employees can work with AI agents throughout the day to complete tasks and take action across the tools they already use." For enterprise buyers who want to simplify vendor management and reduce integration overhead, one app is genuinely appealing.

Best-in-class model vs. best-in-class product

Here's where the argument starts to crack. Building great foundation models and building great consumer products are fundamentally different disciplines. OpenAI is world-class at the first and mediocre at the second. ChatGPT's UX is still rough around the edges. The conversation interface works well for simple queries but becomes unwieldy for complex workflows. The app has accumulated features, from image generation to web browsing to code execution, without a coherent design language tying them together. Adding Codex and a browser into the mix doesn't simplify things, it compounds the problem. Contrast this with Anthropic's approach. Rather than trying to be everything, Anthropic has focused on doing a few things exceptionally well: a strong API for developers, Claude Code for engineering workflows, and a clean desktop app for knowledge work. TechCrunch reported that Claude Code has become "the tool of choice for many businesses," and Anthropic's revenue growth has been outpacing OpenAI's on a relative basis. This isn't an argument about which model is better. It's about strategic focus. Anthropic is making a bet on depth. OpenAI is making a bet on breadth. History suggests depth wins in competitive markets.

The Jevons paradox problem

There's a deeper issue with the super app thesis that doesn't get enough attention. The Jevons paradox, originally about 19th-century coal consumption, observes that when technology makes a resource cheaper and more efficient, total usage increases rather than decreases. Applied to AI: as models get better and cheaper, we don't converge on fewer tools. We get more of them. Satya Nadella invoked Jevons paradox after DeepSeek's efficiency breakthrough, arguing that cheaper AI would lead to more usage, not less. He was right, but the implication cuts against the super app model. If the cost of building AI-powered tools drops toward zero, the natural result is an explosion of specialized tools, each optimized for a specific workflow, not a single monolithic app trying to do everything. We're already seeing this play out. The AI tool ecosystem has fragmented into hundreds of specialized applications: Cursor and Windsurf for code editing, Perplexity for research, Midjourney for image generation, Granola for meeting notes, v0 for UI prototyping. Each of these does one thing better than any general-purpose tool could. The super app thesis requires users to accept "good enough" across every category in exchange for the convenience of a single interface. That's a hard sell when best-in-class alternatives are a tab away.

Platform risk just got louder

For the startup ecosystem building on OpenAI's APIs, the super app announcement is a flashing warning sign. Every feature OpenAI bundles into ChatGPT is a feature that becomes harder to build a business around. This isn't new, platform risk has been a concern since OpenAI started shipping consumer features that overlapped with third-party products. But the super app framing makes it explicit. OpenAI isn't just building a model company anymore. It's building a product company that competes directly with its own customers. Jason Calacanis has publicly warned developers about building on OpenAI's APIs for exactly this reason. The counter-argument is that OpenAI's platform grows the overall market, and there's truth to that. But there's a difference between a platform that empowers its ecosystem and a platform that absorbs it. The super app strategy leans heavily toward absorption. Startups building wrappers around OpenAI's API were already on thin ice. Now, even companies building genuinely differentiated products need to worry about OpenAI deciding their feature is the next thing to bundle into the super app.

What this is really about

Strip away the product strategy and the super app framing is really about distribution. OpenAI has 400 million weekly active users on ChatGPT. That's an enormous surface area. The bet is that consolidating everything into one app creates network effects and switching costs that lock users in, even if individual features aren't best-in-class. It's the same bet Microsoft made with Office, and to be fair, that one worked. But Microsoft had something OpenAI doesn't: decades of entrenched enterprise workflows and file format lock-in. OpenAI's moat is model quality, and that moat is eroding as competitors close the gap. Anthropic, Google, and open-source models are all improving rapidly. If the model advantage narrows, the super app becomes a collection of second-best tools held together by inertia. The more interesting question isn't whether OpenAI can build a super app. It's whether that's the right allocation of their resources. Every engineer working on browser integration or desktop app design is an engineer not working on the next model breakthrough. OpenAI raised $122 billion in committed capital. The question is whether that money is better spent making GPT-6 dramatically better, or making ChatGPT slightly more convenient.

The strategic tension

OpenAI is caught between two identities. It can be the company that builds the world's best AI models and lets an ecosystem of specialized tools flourish on top, or it can be the company that builds the world's best AI app and competes with that ecosystem directly. The super app bet is a bet on the second identity. It might work, OpenAI has the user base, the brand, and the capital to force it. But it requires being great at product development, design, and user experience in a way that OpenAI hasn't demonstrated yet. And it requires the super app model to work in a market where it has never worked before. GPT-5.5 is genuinely impressive. The model capabilities are real. But the strategic wrapper around those capabilities, the everything-app vision, feels like a category error. The best AI company in the world doesn't need to be the best app company in the world. Trying to be both might mean being neither.