AI Agents Could Break the Banking System — With Ethan Mollick

Ethan Mollick argues that AI has already moved far beyond chatbots: the models are more capable, the gaps between what they can and cannot do are still jagged, and the biggest constraint is now human organizations rather than the technology itself. He says the real overhang is not just model quality but adoption, because companies, schools, and regulators change far more slowly than the systems they are trying to absorb.

He frames the most durable human advantages as wide knowledge, deep expertise, taste, and agency. Broad knowledge helps people prompt and connect ideas across domains, deep knowledge lets them check and steer AI, taste helps select good outputs from lots of plausible ones, and agency determines whether someone actually uses these tools to create value.

A major theme is that AI will transform jobs more often than it simply destroys them, though some roles are more exposed than others. Mollick sees customer service and junior training pipelines as vulnerable, while coding, law, and medicine are more likely to be reorganized around AI than erased outright. He also thinks replacing workers too aggressively can backfire, because companies can end up looking and sounding generic if everyone uses the same model outputs.

Mollick is bullish on personal agents and agent-to-agent negotiation, from travel booking to financial frictions that could eventually pressure banks. He thinks universities may benefit rather than collapse because they remain central to assessment, verification, and teaching people how to learn with AI.