Pgrust uncovered memory safety bugs in Postgres SP-GiST and dictionary loading during its test runs. On that basis alone, I think this entire rewrite project earns its place.
People love to complain that using AI on large engineering projects only creates vibeslop garbage. The entire difference comes down to how humans design and run the testing. Setting models loose on complex low-level database software works out fine, provided the test suite is thorough enough. Even if this project never replaces Postgres, that does not matter. Catching these obscure memory safety bugs in the original codebase makes the compute math add up.