Thinking

The End of the Front End

Before the internet, there was no front end. You called a travel agent, told them where you wanted to go, and they handled it. You walked into a bank and a person processed your transaction. The interface was a human. Then the web happened, and suddenly the front end was the product. Companies spent two decades perfecting pixels, optimizing funnels, A/B testing button colors. The screen became everything. Now we're watching that era close. AI agents are starting to handle tasks end-to-end — booking flights, filing expenses, managing workflows — with back-end logic and APIs doing the real work. The "product" stops being a screen you design and starts being a system you orchestrate. It's a return to form. We're going back to the travel agent model, except the agent is software. But here's what makes it harder this time: trust. When you handed your trip to a human travel agent, trust was built through conversation, reputation, eye contact. You could read their confidence, push back in real time, catch a mistake before it was made. With an AI agent, all of that disappears. The user can't see what's happening. They don't know what the agent considered, what it ruled out, or why it chose what it chose. So the product problem shifts from "how do we make this easy to use" to "how do we make this easy to trust." That means solving for transparency — showing your work without overwhelming the user. It means giving people the right amount of control: enough to feel safe, not so much that you've just rebuilt a front end with extra steps. And it means earning trust incrementally, the same way a good human agent would — by being right often enough that the user stops checking. For PMs, this is the new design challenge. The best products of the next decade might be ones you never actually look at — just like the best products of the 1980s were. The difference is that back then, you trusted a person. Now you have to build that same trust into software.