Earlier this year, I attended Davos in the Swiss Alps alongside influential CEOs, political leaders, academics, and economists. After the conference, I joined a series of follow-on virtual sessions and noticed a strange trend: not all of the attendees were human. In fact, a surprising number of invitees sent AI agents in their place—bots that joined the conversation, took notes, and later emailed summaries to their human counterparts. In one instance, a group of 12 was expected, but we ended up with six humans and six AI agents. It was an unusual experience for me, but it appears to be one that’s threatening to become more prevalent: anywhere from 50% to upwards of 79% of executives say that their organizations are adopting AI-agents. While not all agents are acting as stand-ins for executives in meetings, clearly that’s a use case that busy executives are interested in. This got me thinking. In theory, AI agents like these are incredibly helpful. They can attend an hour-long meeting and send you detailed notes within minutes, freeing you to focus on other tasks. But in practice, are they truly a substitute for being there yourself? Do you obtain the same deep understanding of the issues discussed, or just a surface-level recap? And what about the quality of the conversation for everyone else in attendance? For me, the appearance of AI agents sucked the air out of the virtual room. In sessions they attended, conversations increasingly felt flat and empty, devoid of the rich discussion I was expecting. Consider something even more prevalent today—when we skip a meeting and read the Microsoft Teams or Zoom summary of the meeting. Is that sufficient? What are the pros and cons of doing this? These summaries have found their way into nearly every meeting now. Of course if the meeting was a simple conversation which included details of a straightforward routine project, maybe reading the summary is fine. But what if everyone starts doing this? Do we need guardrails on how many people actually turn up to conversations these days? That experience sparked for me a deeper reflection on AI and human metacognition—specifically, how humans reflect on their own thinking and learn most effectively, and how AI might help or hinder that process as it continues to advance and integrate into our lives. AI offers the tantalizing prospect of shouldering many of our daily mental burdens. But what non-obvious benefits of human cognition are we surrendering in the process? AI has undeniable upsides. In fact, we’re already seeing significant productivity gaps between those who use AI well and those who don’t use it at all. So the question is not whether to use AI, but how do we use it without losing our ability to think, process and reflect, and experience the benefits that result—the very qualities that make us human? In other words, how do we preserve the “stuff of thought,” and what do we risk losing if we don’t? The power of attention Sending an AI agent to a meeting may rob us of one of the most impactful experiences for the human brain: focusing on ideas with other people in real time. Being in the presence of others, even in virtual settings, increases the strength of our attention on those ideas and the neural circuits that support the ideas. Neuroscience has shown that when we focus on ideas in the company of others we encode the information in our brain more deeply and with greater understanding. This is because we evolved to pay close attention to social signals—who’s speaking, how others respond, and what that might mean for our own group standing. So just being in the presence of others will more strongly activate the brain. In fact, when we’re thinking and processing with others, we begin to build a shared understanding. In the brain, this leads to neural synchrony, an alignment of the brain’s electrical signals. Research shows that increased synchrony across multiple brains is linked to greater shared understanding, with the more brains involved, the stronger the synchronization. We don’t get this effect of attention, or the activation of robust neural circuits, by reading an AI-generated summary of a meeting. It’s similar to the way reading the cliff notes version of a novel doesn’t give us the same experience as reading the actual book. We might grasp the plot, but we miss the richness, depth, and characters that make it memorable. We’re not moved by the book summary, and we’re not moved by the summary of the meeting either. In the same way, sending a team member or direct report into a meeting for you, to summarize the findings, is not going to give you the quality of understanding or insight into the conversations at hand. The value of “spreading activation” Thinking and processing ideas with others in real time triggers something in the brain called spreading activation. When you think about an idea, this thought process activates a certain number and pattern of circuits across the brain—a neural chain reaction. When you then speak about that same idea to another person, you activate a different set of circuits—a lot more circuits, in relation to that idea—in other parts of the brain. The result? Spreading the neural activation in real time across the brain. This activation triggers thoughts or ideas about other concepts, enabling us to find implications and applications for ideas more easily, consider multiple angles, recognize patterns, and integrate new information with existing knowledge. The stronger the network, the easier it is to connect to other networks, in the form of implications, concerns, or next steps. When AI tools short-circuit this physiological process by providing instant summaries or answers without the space for discussion, there aren’t opportunities for spreading activation to occur, and our thoughts remain surface-level and one-dimensional. This is one of the secondary benefits of increased real-time attention—by aligning with others in real time, we trigger the activation of robust circuits. Our thoughts become richer and deeper, thereby enhancing our ability to understand, remember, and make decisions about our work. In fact, a recent, and frequently cited study highlighted the immediate impact of relying on AI to produce our work, and offload some of our thinking. Here, 83% of individuals who used gen AI to help them write an essay struggled to remember the content of their work, whereas of those who were allowed to use another type of search engine or no tool at all, only 11% struggled to remember. Relying on AI effectively degrades our ability to think about and remember our own work. The motivation that comes from having your own insights In addition to the power of attention and the enablement of spreading activation, working with AI arguably puts at risk the most important element of human thought: generating our own insights. Every company, invention, scientific breakthrough, and creative masterpiece begins with an insight—that “aha” moment when someone sees a problem or concept in an entirely different way. The dopamine rush that comes from an insight fuels our motivation to take action, whether that means solving a challenging work problem, inventing a labor-saving device, or experimenting with a new way of doing business. Some insights are so strong that their memories last a lifetime. The last thing we want is for AI to rob us or others of the opportunity for generating our own insights, or for making those deep connections that inspire us to change the world. However, this can happen when we rely on AI to represent us in the conversation. We may quickly have access to the solutions, but we lose the motivating power that results from reaching a solution ourselves, or we may even rob others of generating their own insights when we aren’t present to share our perspective in a meeting. Time to Pause In our rush to embrace these tools and the processing speed they promise us, we may need to pause to reflect on what we’re outsourcing—not just in terms of tasks, but in terms of thinking itself. The promise of cognitive offloading is undeniably appealing: Who wouldn’t want to think less about the mundane? Yet this convenience may come at a cost. When we consistently offload our attention, we miss opportunities for deeper thought, and sacrifice our moments of insight, we risk dulling the very qualities that make us human. . . . So, the real question isn’t whether to use AI, but how to use it wisely. How should we use AI to enhance, not replace, our best thinking? The answer may lie in what we choose to hand over, and what we choose to keep. It’s time we think a lot more about the stuff of thought.