Ethan Mollick

I think AI’s near-term future will be shaped by a tension between rapidly improving capabilities and much slower human institutions. There is already a substantial “overhang”: models can do more than most organizations, schools, and professionals have learned to use. But technical possibility does not translate automatically into social change. Professional rules, existing workflows, incentives, and simple organizational inertia will slow adoption and make it uneven. The capabilities themselves are jagged. In experiments, AI can perform impressively on one task and then fail on a nearby task that seems equally easy. That means the productive future is not simply “automate everything.” People need to learn where the frontier lies, when to rely on AI, and when expertise and judgment remain essential. At its best, AI can expand creative and intellectual exploration—helping people engage with literature, learn unfamiliar subjects, or receive forms of tutoring and feedback that were previously scarce. But fabrication and bias make uncritical substitution dangerous, especially in education. The central organizational risk is that we optimize for visible output. AI makes it easy to produce more documents, analyses, and ideas, but volume is not the same as value. If organizations use it mainly to industrialize knowledge work, they may erode craft, meaning, and the apprenticeship through which novices become experts. The better future preserves genuine human participation, taste, agency, and learning while using AI to extend what people can do. Which future we get depends less on capability alone than on how institutions redesign work around it.

Overall, I expect a mixed and highly uneven impact rather than a clean positive or negative verdict. AI can make expertise, tutoring, feedback, and creative exploration much more widely available. There is a large gap between what current systems can already do and what most people or institutions actually use them for, so substantial benefits may come simply from learning to deploy existing capabilities well. But those gains will arrive alongside real losses. Organizations may use AI to maximize the volume of knowledge work, standardizing jobs that once involved judgment and craft. If AI supplies the first draft, analysis, or answer by default, novices may lose the difficult practice through which expertise develops. Model providers may also capture more of the application layer, concentrating economic value. The jaggedness of the technology complicates any overall assessment: impressive performance on one task does not guarantee reliability on an adjacent one. Institutions will adapt slowly, and some will use AI thoughtfully while others substitute it where human oversight still matters. So I expect meaningful gains in access and capability, but also disruption to work, learning, and professional development. The eventual balance will depend heavily on whether institutions reward learning, judgment, and valuable outcomes—or merely faster production.

The biggest change would come from evidence that the jagged frontier had largely disappeared: AI systems becoming consistently reliable across adjacent real-world tasks, rather than brilliant in one setting and unexpectedly wrong in another. That would weaken the case for many current forms of human checking and accelerate pressure to redesign—or automate—whole workflows rather than individual tasks. I would also update substantially if institutions adapted much faster than expected. If schools, professional bodies, and organizations quickly developed effective ways to preserve apprenticeship, assessment, and meaningful human participation while capturing AI’s benefits, I would become more optimistic. Conversely, clear evidence that AI-assisted work systematically hollowed out expertise and craft—even where measured output improved—would push me toward a more pessimistic view. The crucial evidence is not another impressive demonstration; it is what happens to reliability, learning, and judgment when AI becomes embedded in institutions.