When using #GSD wth #Codex, I've found it doesn't split between the available models as well as Claude does.
I end up burning through usage with GPT 5.5 instead of simpler tasks being passed off to 5.4 / Terra / Luna.
Any solutions to this problem?
It really grinds my gears that a MySQL UPDATE statement with a join is handled differently than a SELECT.
Just update ALL the join-matched rows instead of choosing a single one, seemingly arbitrarily.
I always forget this nuance because it only comes up once a year. End rant.