Why do so many AI products fail after launch? Is building becoming easier than finding opportunities?

I've noticed something interesting in the AI wave:

Building an AI product is becoming easier every month.

With AI coding tools, APIs, and open-source models, a solo founder can now build an MVP in days or weeks.

But the harder problem seems to be:

How do you know if you are building something people actually need?

I’ve seen many AI products that are technically impressive:

great UI
powerful models
interesting features

But after launch, they struggle to get users.

Maybe the problem is not execution anymore.

Maybe the bottleneck has shifted from:

"Can we build it?"

to:

"Should we build it?"

I'm exploring how founders can validate AI ideas before spending months building.

Some signals I think might matter:

Are people repeatedly complaining about this problem?
Are they already paying for alternatives?
Are there existing workarounds?
Is the problem urgent enough to solve now?
Are people actively searching for solutions?

Curious about other founders' experiences:

How do you validate an AI startup idea before building?

Do you rely on:

Talking to potential users?
Studying competitors?
Looking at communities like Reddit/HN?
Building quickly and letting the market decide?

Would love to hear how others approach this.