"Cadillac Tasks" have now gotten way cheaper in software projects.

These are expensive nice-to-haves, like great docs, or a nice landing page, or internationalization.

Those used to be hard-to-fake signals of the quality of a software project.

Now, LLMs can do those tasks orders of magnitude cheaper, so it's less of a strong signal of project quality than ever before.

Arguably, an inverse signal, one of codeslop, where human judgment was only superficially applied.

The same way that typos in writing are now a signal of humanness, the lack of Cadillac features in software might show that it's full of human judgment.