Why can't anyone trust AI coding as much as human developer?

There is something I have been thinking lately:

Even though AI can write codes & accelerate coding workflows, what becomes the real bottleneck? Why do startup founders & businesses struggle with apps built by the AI-coding platforms and still search for experienced developers to review the code, fix issues and deploy the codebase to server?

I have been a full-stack software developer working in IT job marketplace platform and I have seen a majority of vibe-coders post jobs to review, fix issues and deploy to server, the codebase they built using AI-powered vibe-coding tools and I personally picked up many of such jobs and found it frustrating to build around that codebase.

May be, the bottleneck in software development is slowly moving from execution to judgment.

Why are we trying to make each individual step of software development smarter when the entire process could potentially be automated with precision?

Requirements → planning → development → testing → debugging → review → deployment.

In my opinion, the harness is key to this bottleneck.

• How much of the software development lifecycle can AI actually operate?

• Can AI actually take responsibility for the entire software development lifecycle?

That seems to be a much bigger problem to solve. What do the community think on it?