AI keeps accelerating, but many of the products built on top of it don't seem to be keeping up. We see app after app climb the charts for a day or two and then never hear of it again. The models are improving faster than ever, so what separates the products that last from the ones that fade?
The easiest way to feel chat model progress is to build a chat app - the so-called GPT wrappers that exploded in early 2023. But most of those apps weren't sticky. They may have changed the output slightly, but the core experience stayed adjacent to what the frontier labs were already offering.
We are past the point of simple chat interfaces being able to break into the mainstream in a lasting way. Image generation applications that let you replicate yourself as an anime character quickly faded as people realized you can do the same thing in ChatGPT and Gemini. Products whose core purpose can be achieved with a prompt or two to an AI model are doomed to fail.
This can be discouraging to many builders, but there is so much opportunity left on the table. Humans have so many interests and aspects of their lives that AI is just not aware of. It doesn't know the decisions you made, doesn't learn much of your preferences unless directly told, and can only learn from you based on the limited interactions you have - which for most people are also limited by topics. Leveraging that missing context is exactly where the next wave of AI applications will come from.
Developers should build AI experiences around context that is gathered outside of core AI flows. One example I am currently experimenting with is an AI nutrition and strength coach. When I started lifting consistently a couple months ago, I found myself constantly going to ChatGPT to ask questions about specific exercises, macronutrients, or even changes that I noticed in my body. The answers were helpful, but never fully exact. I always ended up with it pointing me in several different directions depending on several different factors. I knew there had to be something better.
Problems like this are leading my current line of exploration. How can we leverage the unused context in our daily lives to provide more tailored and specific experiences across different domains? You could easily imagine an AI coach that can find correlations between your weight, exercise, and nutrition and is always there to answer questions on demand. Or even an AI co-worker that knows exactly the right context for a project you are working on in the long term and is not bound by a specific chat session.
The first challenge is building an app that is useful to users enough to have them share context with it on an ongoing basis. An extensive onboarding to get a feel for your user's taste will just make people ditch the app before they sign up. The product itself has to encourage meaningful human input. The second challenge is rewarding your users for that input with an ultra-useful AI. If I just keep inputting information into an app that doesn't leverage it, the app might as well be a note on my phone.
The products worth building are the ones that grow the more you live with them. Every workout logged, every link shared, every decision and action becomes context that makes the AI smarter about you. That becomes the moat for any app you build. Context creates switching costs. Not because users are trapped, but because the product is already tailored exactly for them. In an age where anyone can ship an app in a weekend, the competitive edge isn't being able to build anymore. It's knowing what's worth building.