I interpreted “random” in the parent comment to mean “ad hoc, but planned”. Meaning, not random, untrusted or unknown engineers, but engineers from an obvious trusted group (like my team), just accessed in an ad hoc manner.

The point being that short of choosing a convention for which person’s copy (or which machine, etc) counts as the primary source of record (which would then just be poor man’s centralization with the admin costs and maintenance of any centralization tools now a burden on the team rather than a third-party product), then interacting in an ad hoc way with even just my known team’s set of distributed clones of the project turns into a bookkeeping problem nobody wants.

I can also imagine a non-centralized model can lead to many more complicated workflow failure modes. I’m just thinking about how often novices get stuck with rebasing errors or squashing commits incorrectly, and bikeshedding arguments over whether it’s ok to ever revert master or if you should intentionally keep mistakes in the history and correct with new commits. I can imagine it being even worse when there are fewer conventions, since conventions are often the only way to avoid these bikeshedding debates about preferred git hygiene methods.