if properly reviewed and it works right, you can't argue with results.

The key word is "if".

This is rather the crux of the issue: most AI-generated code is not reviewed, let alone reviewed by humans, let alone reviewed by human experts within their expertise. Nor does AI-generated code have a good history of being well-tested to any particular formal standard of validation (eg ISO), against any defined criteria that isn't itself AI-generated and unreviewed. There are outliers, no doubt, which strive to lift themselves above this low-bar, though the effort to do so often exceeds the effort to just have the experts hand-write the code instead and then formally validate it, at least as of early 2026.

Some AI could plausibly be tolerable within an already-functioning software engineering team. But "all code in Trail Mate is 100% generated by AI under human guidance" is an abdication too far.