According to Inc, new research from OpenAI, Duke, and Harvard shows almost 80 percent of ChatGPT usage is for basic guidance and simple writing help, with most users on the free plan. This pattern likely extends to other major platforms like Claude, Gemini, and Copilot, where few users touch advanced features. The article breaks the AI landscape into four layers: foundational model providers (like OpenAI and Anthropic), development solutions (like Replit and Praxie), retrofitted software (like Canva Magic Studio or Zoom AI), and AI-first applications that couldn’t exist without it. The author, who co-founded Praxie, argues the foundational layer is rapidly becoming commoditized infrastructure, akin to cloud providers, where features flatten and price drops.
The Infrastructure Is Commoditizing
Here’s the thing that jumped out at me. The big LLM providers are in an arms race on specs, but for 95% of users, they’re basically the same tool. You ask a question, you get an answer. One might be slightly better at coding this month, another at reasoning the next. But that’s a fleeting advantage. They’re becoming utilities. Think about it: do you care which power plant generated the electrons charging your phone? Not really. You just want the lights to turn on. That’s where foundational AI is headed. It’s the underlying research showing most use is simple and free that really drives this home. The value is shifting up the stack.
The Real Action Is in the Next Layers
So if the base models are a commodity, where’s the opportunity? The article points to two key areas. First, development platforms like Replit (powered by Claude Opus) and Praxie (powered by Google’s Gemini) that let you build apps on top of these models with minimal code. They’re the pickaxes and shovels for the AI gold rush. Second, and more interestingly, true AI-first applications. These are things that literally couldn’t exist before, like the author’s own examples: Bias Breaker for analyzing news article bias or Face Ager for simulating aging. These aren’t features tacked on; the AI *is* the product. This is where novel interfaces and new business models will emerge.
Don’t Believe the Overnight Hype
Now, the article makes a crucial point that’s easy to forget. This won’t all happen in 2026. It compares AI to the early internet. Amazon was founded in 1994 but didn’t turn a full-year profit until 2003. The infrastructure matures first, then the applications, then the world-changing business models. AI will move faster, but the pattern is probably the same. Widespread adoption in non-tech industries? That takes time. People need to learn, workflows need to be redesigned, and a ton of legal and ethical guardrails need to be built. The hype cycle suggests everything changes tomorrow. The reality is we’re laying the foundation for a change that unfolds over years.
What to Do About It Now
So what’s the practical takeaway if you’re not building the next GPT-5? Stop obsessing over which model is “best.” That’s a losing game. Instead, think in workflows. What annoying, repetitive process could be automated with a simple tool? Start small. Use the AI features already in your existing software. The biggest value isn’t in building another chatbot. It’s in solving a specific, painful problem with intelligence and elegance. The barriers to creating these solutions are collapsing. With platforms like Replit and Praxie, what took a dev team months can now be prototyped by a solo builder in days. That’s the real shift. AI is no longer the product you sell; it’s the ingredient you cook with. And the chefs who understand that are the ones who will define 2026 and beyond.
