Genuinely sounds like the kind of challenge that could be solved with a swarm of Codex coding agents. I'm surprised they aren't treating this as an ideal use-case to show off their stack!
Getting the sharing in-place, yes, but maintaining it operationally would still be a headache. Things like schema migrations across shards, resharding, and even observability.
