OpenAI foresees millions of AI agents 'somewhere in the cloud' in just a few years — with human supervision

OpenAI is placing its bets on a transformative future where AI agents play a pivotal role in software development. During a recent episode of the "OpenAI Podcast," co-founder and president Greg Brockman, along with Codex engineering lead Thibault Sottiaux, painted a picture of a future teeming with autonomous AI agents. These agents, while supervised by humans, will operate continuously in the cloud, acting as essential collaborators in various projects. Sottiaux expressed strong confidence in this vision, stating, "We have a strong conviction that the trajectory is leading us toward large populations of agents in the cloud, which humanity—individuals, teams, and organizations—will supervise and guide to create significant economic value." He projected that in just a few years, organizations could see millions of these agents functioning within their data centers, delivering valuable contributions. On the same day, OpenAI unveiled GPT-5 Codex, an advancement that allows the AI to tackle complex software tasks for extended periods. This new iteration can handle intricate code refactorings while seamlessly integrating into developers' cloud-based workflows. Kevin Weil, OpenAI's Chief Product Officer, highlighted the increased efficiency achieved through Codex-powered internal tools like code review systems during a discussion on tech entrepreneur Azeem Azhar's podcast, "Exponential View." Despite these advancements, the role of human coders remains vital. Experts have noted that while AI can assist in coding, it doesn't replace the need for skilled programmers. Brockman emphasized the importance of human oversight as AI agents take on more complex responsibilities. Since 2017, OpenAI has been strategizing how to ensure that humans or simpler AIs can effectively supervise more advanced AI systems, maintaining essential oversight and control. Sottiaux acknowledged the challenges ahead, stating, "Understanding this entire system and enabling it to be multi-agent and steerable by various stakeholders is where we are headed. While it may seem a bit nebulous, it is also incredibly exciting."