Agent demos move fast. Enterprise readiness moves slower. Hiring spikes and product claims (agents that “do the work”) often outrun reliability, permissions, and audit trails that real organizations need.

OpenAI’s agents narrative and big-tech enterprise reorgs are part of the same cycle: pressure to ship autonomy while buyers still ask who is liable when the agent is wrong.

Quiet test: can the agent operate inside existing identity, logging, and approval systems — or only in a sandbox slide deck? Until that gap closes, treat agent hype as a roadmap signal, not an ops plan.