An AI Confession
I use AI every day and still have nothing to show for it. Between getting the job done and building a product there is a gap, and I think that gap is sensitivity to pain points.

Time for an AI confession: I'm admitting how ordinary I am, ha.
I really envy people who get real results with AI, whether they ship a beautiful product or actually make money. That's impressive, and I respect it, because I'm the person who uses AI every day and still hasn't gotten a result out of it.
That said, AI has made me more productive at work, and it has let me do plenty of things I couldn't do before. That feedback is real, otherwise I wouldn't have kept using it this long. But all of it sits inside "get the job done", and there is a huge gap between that and "create".
What is the gap? I'll hold off for a moment. Saying it directly sounds abstract, so I'll start with what I managed to do and what I didn't, and use a few examples to make the idea clear.
What I managed to do
First, what I did.
A very simple example: Excel.
Before AI, the most I could do in Excel was write a SUM and pull a table together. I had no idea how to use advanced functions, let alone VLOOKUP.
AI expanded that ability. I no longer worry about how to build the sheet; I only worry about the more important question of how the data should be defined.
Same with writing a tool, a script, or extending an existing project.
Pause here and notice that all of these are "get the job done": work needs something, so I use AI to do it. Sum a column, build a table, make a deck, write a tool.
Before I start, I have a clear goal and I know what "good enough" looks like. The process? I don't fking care.
Is that a problem? Not really. Work is results-driven. Get the job done.

What I didn't manage
Now, what I didn't do.
I once tried to rebuild Codex from scratch with an agent. Partly as practice, and then the plan was to keep going: open source it, keep adding features, version it, and gradually turn it into my own agent harness product.
I actually did it. After burning 3.5 billion tokens, I had all of Codex's basic features at the time.
Everything looked fine, and then I stopped the project.
Why? Because once that version was done, I lost my goal. I didn't know what to do next.
Waiting for the next Codex release and copying it again didn't seem worth much. The Codex CLI is open source anyway.

Putting the two examples together
Put the two examples side by side. If you are always getting the job done, the process doesn't matter: the moment you hit the goal, you move on to the next one. That works fine, and AI can help you do it very well.
The moment you want to build a product, that approach stops working. You need short-term goals and a clear long-term goal, which means you have to know what you want, or what someone else wants, or what a particular group of people wants. Otherwise, after you hit one or two goals and don't know where to go next, you end up lost, anxious, miserable, and eventually you quit. Worse, you start grasping at anything.
The gap is sensitivity
So back to the gap from the start. I think it's sensitivity.
At the very least, you need to be sensitive to pain points.
The reason is simple: if a pain point doesn't hurt you, you won't notice it, you won't be motivated to fix it, and you certainly won't turn it into a product that helps other people with the same problem.
That gap may be wider, deeper, and harder to close than any vibe coding trick, any harness engineering, any software engineering knowledge, or which agent and which model you use.

So instead of paying for courses on techniques, ask yourself what you actually want.
If you use AI to handle what's in front of you and free up energy for things that have nothing to do with AI, that's completely fine. That's exactly what AI is for.
If you want to create a product, first figure out which group of people you're solving what problem for, then start building. When you hit something you can't solve, go find outside knowledge then.
I used to think this was old, vague self-help talk. Now I think it's a hurdle you have to clear if you want to build production-grade products.