Hacker News Reader

> new versioning primitives/architectures which can handle machine-level code generation - 1 commit per second per repo.

This is not a particularly novel level of scale. Facebook's mercurial backend had to handle >5,000 developers committing to the singular monorepo long before LLMs were a thing

Yes, on a single repo. Now multiply that per bazillion companies on github, some of which are trying that.

Why the heck would you want to do this. Using git as your undo chain sounds like a pretty awful thing to do.

I think it'd be pretty neat to be able to rebase my undo history on to a remote branch someone else is working on.

This is smart but also hilarious.

This seems odd to me. Why would you need to commit every second?

And push to remote as well? Seems not thought out

> GitHub is not agent scale.

Is the scaling issue with git or github?