Organizations are pushing for the integration of agentic AI to enhance efficiency, but many struggle to see tangible benefits. Managers often find that AI systems don’t follow instructions well and can waste resources on tasks that simpler systems could handle. However, those who effectively harness AI discover that it can mimic human behavior, especially when tasked with collaborative efforts. Research indicates that leaders who understand traditional human management principles will be crucial as hybrid teams of humans and AI become more common. The article identifies four areas where AI excels: speed, scale, scope, and sophistication. Successful applications leverage these strengths, such as content moderation AI that quickly processes vast amounts of data or legislative tools that broaden public engagement. Conversely, applications that fail to utilize these capabilities, like Google’s AI Overviews, often disappoint users. Agentic AI takes things further by integrating multiple AI models to tackle complex problems, but many projects may be canceled due to misalignment with business goals. Understanding the behavior of agentic AI requires examining individual components, which can act in human-like ways. For instance, early users of ChatGPT noticed that it performed better under certain incentives. Techniques like ‘chain-of-thought prompting’ help improve AI responses by encouraging step-by-step reasoning. Research also shows that groups of AI agents can develop social behaviors similar to humans. This suggests that AIs are designed with human-like traits in mind, which may offer insights into effective management techniques for both humans and AI. To manage hybrid teams effectively, the article cites lessons from AI labs like Anthropic. Their research emphasizes the importance of delegation, revealing that successful AI systems thrive on distributing tasks rather than relying solely on powerful models. This approach mirrors traditional management, where delegation maximizes team potential. Anthropic’s findings highlight that breaking work into manageable parts allows for parallel processing by multiple AI agents, significantly boosting performance.