The anxiety is real. A Pew Research Center study published in August found that 71 percent of Americans believe AI will lead to fewer jobs over the next two decades, a figure that has risen 7 points in just two years. That number captures something genuine: a growing sense that the technology is moving faster than most people can track, and that the consequences will be unevenly distributed. What it does not capture is what to actually do about it. That is the gap Ethan Mollick is trying to close.
Mollick is an associate professor at the Wharton School and faculty director of its Generative AI Labs. He writes a newsletter called One Useful Thing, with nearly half a million subscribers, and his latest book is titled “Co-Existence: The Next Phase of AI.” His central argument is straightforward, even if it is uncomfortable: AI is not going away, and the most productive response is not resistance or denial but deliberate engagement.
What Most People Still Do Not Understand About Agents
The conversation around AI has shifted, but public understanding has not fully caught up. Mollick points to AI agents as the clearest example of this gap. Earlier versions of agentic systems required constant human guidance, made frequent errors, and hallucinated regularly. That description no longer fits the current generation of tools.
Agents can now handle extended, multi-step tasks with far less supervision. Mollick describes coders who report being more than a hundred times more productive than they were before. He also offers a more mundane illustration: he asked a personal agent to cancel a recurring subscription to airline wifi, a task that required navigating customer service channels. The agent handled it without further input. The subscription was gone.
This is what most coverage misses. The shift is not just about chatbots getting smarter at answering questions. It is about systems that take action in the world, on behalf of people, across real workflows. Walmart has enabled agents to shop on behalf of customers. Apple has announced a home display designed to work with AI to control household functions. The technology is moving from conversational to operational.
The Skill Trade-Off Nobody Wants to Think About
The harder question is not whether AI can do useful things. It is what happens to human capability when people stop doing those things themselves.
Mollick is direct about this tension. Students are the clearest case: homework is tedious, the incentives to complete it are often weak, and AI can do it quickly. But the act of doing homework is also how learning happens. Offloading it removes the friction that builds competence. The same logic applies to professional work. When AI handles the drudgery, people gain time. They may also lose the practice that keeps their underlying skills sharp.
There is a subtler problem too. When people use AI in assistant mode rather than as a genuine learning tool, the experience can feel like engagement even when it is not. The sense of being in the loop persists even as actual understanding erodes. Mollick describes this as one of the central personal challenges of the current moment: knowing when to let AI do the work, and when doing the work yourself is the point.
His own concern, stated plainly, is about de-skilling. If a person offloads enough cognitive tasks, they may eventually lose the ability to evaluate whether the AI’s output is any good. That is not a hypothetical risk. It is a structural feature of how these systems interact with human attention and habit.
Why the Coexistence Frame Matters
The word “coexistence” in Mollick’s title is doing real work. It is not optimism. It is not a promise that everything will be fine. It is a framework for thinking about agency in a situation where the technology itself is not subject to individual veto.
Most people are not in a position to shape AI policy, regulate its development, or slow its adoption across industries. What they can do is make deliberate choices about how the technology fits into their own work and life. That requires actually using it, not to endorse it uncritically, but to understand what it can and cannot do. Mollick’s argument is that ignorance of the technology is not a neutral position. It is a disadvantage.
This matters for how organizations think about AI too. The question is not whether to introduce these tools, but how to introduce them in ways that augment what people do rather than quietly erode the expertise those people have spent years building. The risk of getting that balance wrong is not abstract. It shows up in the quality of decisions, the depth of institutional knowledge, and the ability to catch errors that an AI system would never flag on its own.
In Short
71 percent of Americans fear AI will reduce jobs over the next two decades, and that fear is not irrational. But fear alone does not produce useful responses. Ethan Mollick’s argument is that the more important question is not whether AI will change work, but how individuals and organizations can engage with it deliberately enough to stay in control of what they know, what they do, and what they choose to hand off.
Based on reporting from Vox.