OpenAI foresees millions of AI agents 'somewhere in the cloud' in just a few years — with human supervision (www.businessinsider.com)

OpenAI unveiled a vision and product push toward a future in which "millions" of autonomous AI agents run continuously in the cloud and inside companies’ data centers, but operate under human supervision. The company launched GPT-5 Codex, a model it says can run for hours on complex software projects (for example, large-scale refactorings) and integrate directly into developers’ cloud workflows. OpenAI executives described multi-agent systems that teams and organizations can steer to generate sustained economic value, and noted internal Codex-based tools (like automated code review) have already improved engineer productivity. The announcement is significant because it signals a practical shift in software engineering and infrastructure toward long-running, collaborative AI agents rather than one-off prompt interactions. Key technical implications include robust stateful execution (hours-long tasks), tighter cloud integration, and orchestration of large agent populations — all of which heighten needs for tooling around supervision, alignment and multi-agent steering. OpenAI emphasizes humans (or simpler AIs) will remain necessary to oversee and align agent behavior, reinforcing that these agents augment rather than immediately replace skilled developers while raising new questions about safety, governance and operational design.