The chatbot window is the bottleneck, not the model

Ethan Mollick's piece makes a point that is easy to miss because the interface is invisible until someone names it. Chatbots answer in walls of text, and the work of reading, sorting, and pasting that output back into real tools is itself a tax. He cites studies of financial professionals whose productivity gains were partly eaten by the mental strain of managing the conversation rather than the task.

The alternative he describes is agents that meet you where the work already lives. Instead of a chat window, the model reaches into your files and applications, with a short message to your phone when it needs you. He gives two concrete examples. One is a morning briefing that pulls from calendar, email, and channels into a single report on your phone. The other is a request, sent from a phone, to check whether a chart on a slide was current: the agent found the PowerPoint, searched the machine for newer data, downloaded a referenced PDF, pulled a fresher graph, and updated the deck on its own.

The argument is narrow and practical, not a manifesto about the future of work.

Why it matters

If you are deciding how to deploy AI to a team, the lesson is to judge the interface as carefully as the model: the same capability buried in a chat box can cost more attention than it saves, so favor agents that act inside existing tools.