OpenAI has spent the last eighteen months doing something that doesn’t quite fit the narrative of a chatbot company trying to monetize its way to AGI. It’s been quietly positioning itself underneath other software - not beside it.
The pattern is clearest when you look at the operator tier of the ChatGPT API ecosystem. Businesses aren’t just calling GPT-4o to answer customer emails. They’re building entire product surfaces on top of it: scheduling tools, legal document review systems, internal knowledge bases. OpenAI is less a product those companies compete with and more a layer they depend on. That’s a very different business.
The Consumer App Was Never the Point
ChatGPT crossed 300 million weekly active users earlier this year, and the number gets cited often enough that it’s started to feel like a destination stat rather than a meaningful one. But consumer engagement numbers are a distraction from where OpenAI’s actual leverage accumulates.
The real play is making model access so deeply embedded in developer and enterprise workflows that switching costs become structural. Anthropic understands this too - Claude’s API has been aggressively courting developers who want more predictable output formatting and longer context windows. Google’s Gemini is doing the same inside Workspace. Everyone is racing to become the runtime that nobody replaces.
Why This Matters More Than Model Benchmarks
The AI industry has a benchmark obsession problem. MMLU scores, GPQA, coding evaluations - these get treated as product reviews when they’re closer to controlled lab results. What actually determines which model wins enterprise adoption is stickiness: how many internal tools get rebuilt around it, how many engineers learn its quirks, how much institutional knowledge accumulates in fine-tuned versions.
OpenAI’s Custom GPT and fine-tuning infrastructure are explicitly designed to accelerate that stickiness. Once a company has fine-tuned a model on its own data and integrated it into its CRM, that’s not a subscription they’re casually cancelling.
The Risk Nobody Is Talking About
Platform strategies fail when the platform owner starts competing directly with the builders on top of it. Microsoft did this to enterprise software vendors for decades. If OpenAI’s operator-facing tools eventually shade into full vertical products - an AI CRM, an AI legal research product - the developers currently building on its API become competitors rather than customers.
That tension hasn’t broken into the open yet. But the incentive structure is already there, and platform histories suggest it usually does.