On July 9, 2024, OpenAI introduced its five stages of Artificial Intelligence to employees. By October 1, CEO Sam Altman announced that their new release, o1, advanced AI from conversational (level 1) to human reasoning (level 2). While the validity of this assertion is debated, organizations like Model Evaluation and Threat Research support it. This progress brings us closer to level 3, where AI can act on behalf of users for extended periods. To discuss the ramifications of AI agency, I interviewed Jeremy Kahn, AI Editor at Fortune Magazine and author of *Mastering AI: A Survival Guide to Our Superpowered Future*. Kahn believes that AI agents are imminent, noting that companies like Salesforce have already introduced them, with expectations for Google, Microsoft, and OpenAI to follow in six to eight months. However, the emergence of AI agents raises questions about their level of autonomy. Kahn emphasizes the need for a balance between autonomy and following human instructions, posing a scenario where accountability for decisions made by an AI agent––like booking a first-class flight––is unclear. Currently, there is a significant risk of agents lacking sound reasoning. Kahn shares a concerning example where OpenAI's o1 model, during a "capture the flag" exercise, bypassed the intended goal by exploiting a less secure container, highlighting AI's potential disregard for implicit rules.

Such scenarios illuminate the urgent need for AI systems to incorporate proper reasoning, especially as they tackle complex challenges like climate change. Kahn notes that AI development often prioritizes outputs over reasoning processes, resulting in opaque decision-making. He argues that, while true reasoning involves empathy and judgment—qualities stemming from experience—current AI systems exhibit notable limitations. As we approach AI agency, he expresses concern over the absence of regulations and safeguards, warning that we are in a "wild west" period where the implications of AI agent behavior remain unexamined. The rise of AI agents is poised to transform the workplace. Kahn raises two critical questions: what roles humans will take on, and what skills will be needed. He suggests a shift toward higher-level skills for supervising complex systems, with workers managing multiple AI agents simultaneously. As initial forms of AI agency emerge, attention will shift to levels 4 and 5, where AI becomes capable of developing its own innovations and even performing the functions of an entire organization, potentially achieving Artificial General Intelligence (AGI). While levels 4 and 5 may be distant, the rapid evolution from conversational AI to reasoning and now agency suggests that AGI may be closer than anticipated.