AI can't read your thoughts

There's a lot of hype around AI coding tools right now. A mantra I keep hearing is: Just tell it to make an app, and it works.

And what's amazing is that it does work, at least in many cases. Tools like Bolt, Lovable, and others have really lowered the barrier to getting an application up and running with very little prompting.

But the more you want that something to be yours — to actually fit how you work, how you think, how you want it to fit in, the problem you're trying to solve — the more you realize that "make me an app" is just the beginning of a long and potentially endless journey of thinking.

AI tools can do a lot

They can handle implementation. They can plan effectively. They can research, write code, and iterate. The one thing they can't do consistently, and will probably never be able to do, is read your thoughts.

Human thought is erratic

We jump around. We have half-formed ideas. We know what we want, but we can't always articulate it clearly. We often assume context exists when we haven't stated it.

I really don't think there's a replacement for actually sitting down and taking a moment. You don't need the full solution mapped out, but I think you should outline the problem. That can be as simple as speaking whatever comes to mind into a machine, as AI agents are really good at interpreting that erratic thought.

If you're going to use a tool to craft a real solution, make sure it is conducting the right research and implementing the solution in a way you are comfortable supporting, as you will ultimately be responsible for it.

The plan that an agent takes to begin implementing is just that — the beginning of the implementation process. At that point, you've likely done the work to understand the core of the problem, and this is the beginning of a solution.

"Make me an app" works, but ...

AI will work on a broad prompt. It will give you something. But it might not be anything like what you had in your head.

I wrote recently about my daughter building an application. Her prompt was, "I want you to build a candy website." And Bolt built a full-fledged e-commerce website to sell candy. What she actually wanted was to report on the candy she had received while trick-or-treating for Halloween. Context mattered.

Still, for many people, prompting with just enough broad information is fine, especially early on. But the reality is that, in those scenarios, you're relying on the context and programming the providers who created the agent have provided.

So you might get something that hasn't been created before, but it's very likely to look similar to other tools and sites created before. It's probably going to reuse placeholder content and follow familiar patterns. It's different from picking up a WordPress template and getting an identical copy of a site, but it's not that different. And soon, it will be viewed as not different to the general public, if it isn't already. (If you want to see this in action, check out why your AI keeps building the same purple gradient website.)

You've become a product manager

This doesn't even have to involve a technical background. I find myself doing this too — I don't really know what I'm going to build, I just know the core of the problem. And while I might have a longer prompt than someone without a technical background, the solution might end up just as generic.

One way to approach this is to just pick one thing at a time and keep changing it until you really like what you've got. That way, you're doing all the planning through building and iteration. You are, and have become, the product manager.

Product thinking still matters

That is not to belittle the art and science of product management. People who have built products before are a step ahead today. They can do more of that deeper thinking upfront — the kind that enables an agent to be more successful from the start in shaping a unique MVP focused on solving a well-thought-out problem.

Today, everyone is a developer, and everyone is a product manager. And the folks who've made that concept their reality are the ones who are navigating this new world most successfully.

Pre-implementation thinking isn't going away

I think there's a misconception that AI removes the need for planning. That you can just start prompting and iterate your way to something great.

And you can do that today.

But you'll get there faster and with better results if you think first. Outline the problem (even if it's totally erratic). Understand what you're actually trying to build. Work with AI to build a rock-solid plan. Be clear about the constraints and trade-offs. Then hand it off.

The pre-implementation phase doesn't have to take much time, but it still offers significant value. And if done well, it leads to a prompt from which an agent can craft an MVP of a solution that is 90% or more complete.

Think, plan quickly, and happy building!