The idea behind AI in products makes sense. You reduce wait times. You remove the friction of talking to a human who might not understand you. You make things more flexible. People can interact on their own time, from anywhere. In theory, it should make life better.
And in some places, it does. But somewhere along the way, companies stopped asking where AI helps and started putting it everywhere. Every section of life now has a chatbot. Finance, healthcare, customer service. It doesn't matter. There's a chatbot.
The problem isn't the technology. The problem is what happens when you let it take over the identity of your product.
When AI Takes Over #
Take Duolingo. It used to be one of the best apps for learning a new language. It had a clear identity. You knew what you were getting. Now it's becoming an AI app that happens to teach languages. Open it, and AI is front and center. The experience has shifted.
Here's what I think they're missing: if I want an AI app, I'll go to ChatGPT. When I open Duolingo, I want Duolingo. The moment you try to combine both, the original idea gets lost. You're not improving the product anymore. You're drifting it into something that other companies already do better.
And not everyone wants this. Some people like writing things out by hand. Some people prefer traditional methods. Some people learn better without AI generating answers for them. When you force it in, you take that choice away.
When AI Works #
Then look at Apple. They've been integrating AI in a completely different way. The structure stays the same. The system stays the same. AI just makes it easier. It summarizes things for you. It works in the background. It doesn't prompt you to generate an email. It helps you with the one you're already writing.
That's the key difference. The product keeps its identity. AI is a tool, not the interface.
Another good example: AI as a study companion. You have your material on one side, and AI on the other. You can ask questions, follow up, simplify an idea, go deeper. It's interactive. It's useful. And it solves an actual problem: understanding something you're stuck on, without replacing the learning itself.
When AI Gets in the Way #
Now look at Microsoft. Copilot is everywhere. You open Windows and it's Copilot to generate an email, Copilot to summarize a document, Copilot in every corner of the operating system. They even put it in Notepad, the simplest text editor there is. At some point, you don't even feel like you're using Windows anymore. You feel like you're using Copilot that happens to run on Windows.
They went further than software. New Windows laptops now come with a dedicated Copilot key on the keyboard. A physical button built into the hardware. It gives you the impression that Copilot is everywhere, not just in the operating system but on the machine itself. Even the hardware is telling you to use AI.
The backlash was so strong that Microsoft's own Windows lead admitted the OS "went off track" with aggressive AI expansion. They've started rolling back Copilot from apps like Photos, Notepad, and Snipping Tool. Even Microsoft realized they overdid it.
What if you don't trust AI? What if you don't like it? What if it makes mistakes? There's no way around it. That's the problem with forcing it into every possible space.
I had a similar experience today. I needed to call a doctor, and the service had replaced the human agent with an AI assistant called Jakob. All it did was ask my name, ask what I needed, and pass that to the doctor's office, where a human agent would handle it anyway. That's it. A simple form could do the exact same thing. Instead, someone deployed an entire AI system that solves no actual problem. The process didn't get shorter. The experience didn't get better. It's just a more expensive way to collect a name and a reason for calling.
When AI is everywhere, it stops feeling helpful. It starts feeling like noise.
The Infrastructure Problem #
There's another side to this that people don't talk about enough. All of this AI depends on a handful of companies. AWS, Microsoft Azure, Google Cloud. They're the infrastructure behind almost everything. When you make your product rely on AI for core functionality, you're not just adding a feature. You're adding a dependency on systems you don't control.
If one of these providers has an outage, and they do, it doesn't just affect your AI feature. It can take down the entire experience. A doctor's office that replaced its phone system with AI can't take appointments. A customer service platform goes silent. The more you weave AI into everything, the more fragile the whole system becomes. And then there's the cost. Running AI at scale is expensive. The energy consumption is enormous. The environmental impact is real. Every chatbot answering a question that a simple FAQ page could handle, every AI layer processing what a basic form could collect — all of that burns compute, burns energy, and costs money that gets passed on somewhere.
When AI solves a real problem, that cost is justified. When it's there because someone decided every product needs it, you're paying a steep price for very little return.
What Would Actually Help #
Here's what I think companies are missing. AI is smart. It's efficient. So use it to solve real problems, not to be the interface for everything.
Take healthcare. If a doctor already has your profile, and the system already knows your history, AI could process all of that in the background and generate a meaningful report before you even walk in. That reduces the actual process: the calling, the waiting, the repeating your history every time. It doesn't replace the doctor. It makes the system around the doctor work better.
That's AI as a tool. Not AI as the product. Not AI as the first thing you see when you open an app.
The companies that understand this, that use AI to strengthen what they already do well instead of replacing it, those are the ones getting it right. The rest are just sticking AI everywhere and hoping it looks like progress.