If AI can build anything, why do so many products still fail?
Short answer
Because building was never the constraint that killed them. Distribution was. Lowering the cost of building increases the number of products competing for the same fixed amount of attention, which makes distribution harder rather than easier. The bottleneck moved; it did not disappear.
The longer version
Because building was never the constraint that was killing them. The graveyard is not full of good products that could not be finished; it is full of finished products nobody ever heard of. Removing the build cost removes a bottleneck that was, for most failed products, not the binding one.
Worse, it makes the real bottleneck harder. Attention is roughly fixed — there are only so many hours, feeds and inboxes. If the number of products competing for that fixed attention goes up sharply, the average product's share goes down. Cheaper building means more competitors for the same scarce thing.
Which is the whole argument for treating distribution as a first-class discipline rather than the bit you do at the end. When building was expensive, being able to build was an edge. Now that it is cheap, the edge is being able to get something seen — and that skill has not become any more common.
The full argument, with the rest of the picture around it:
Vibe Coding Killed the Idea Backlog. Now Everyone Has the Same Problem