Microsoft poorly manages token use of most expensive models in a pilot. Then they use that failure to advertise/position their own Github Copilot agents to procurement teams, over the now widely validated Claude Code-based agents.

At least Codex is trying to win validation on merit.