Feedback loops when building is abundant

As building software gets easier, I keep thinking about how long it still takes to find out whether what I’ve built is useful.

The loop used to look something like this:

Step	What took time
Have an idea	Usually the easy part
Build it	A large part of the work
Get feedback	Wait for someone to use it and understand where they get stuck

AI can shorten the building step enough that feedback takes a much larger share of the loop. Taste matters more when there are many versions I could make quickly. I still have to decide which one is worth putting in front of someone.

Sometimes I can spend tokens to get useful feedback before anyone tries the product. A model might notice that three clicks could be two, or recognize a confusing interaction from patterns it has already learned. Those are changes I can make and test cheaply.

Other problems take more empathy. Someone might reach a screen and have no idea what to do next, even though every button works. A model can suggest why that might happen. Watching a person use the product gives me a way to check the suggestion.

That distinction affects what I’d build next. If the missing knowledge can come from another prompt, I can keep iterating immediately. If it depends on how people use the product, I need to get it into their hands and leave enough time to learn from them.

It also changes the build-versus-buy question. An existing product may have years of feedback behind an interface that looks simple enough to recreate in an afternoon.