Management as AI superpower

Ethan Mollick ran an unusual experiment at Wharton: give executive MBA students, most of whom had never written a line of code, four days to build a startup using Claude Code, Google Antigravity, ChatGPT, and Gemini. The results weren't just decent student projects. Teams shipped working prototypes with real functional features, from a verified-ticket marketplace called Ticket Passport to a blood-sugar prediction app named Invive. Mollick, who has taught entrepreneurship for fifteen years, says what he saw in two days outpaced what a full semester used to produce before AI entered the picture.

The interesting part isn't the speed. It's why the speed happened. Mollick argues the real lever wasn't prompting cleverness but plain old management competence — the ability to say clearly what you want, judge whether the output is any good, and correct course when it isn't. He backs this with numbers from OpenAI's GDPval study, where experts took an average of seven hours to complete complex professional tasks. GPT-5.2's Thinking and Pro models tied or beat those experts 72 percent of the time, and even when you factor in the hour typically needed to check an AI's work, delegating still nets roughly three hours saved per task on average.

Mollick frames this as an equation: weigh how long a task takes you against how long it takes to prompt, wait for, and verify an AI's attempt, adjusted by how often that attempt actually clears your bar. Raise the success rate or shrink the review time, and delegation becomes an easy call. Lower them, and you're better off just doing it yourself. Crucially, he says, subject-matter expertise is what improves both variables — experts write sharper instructions and catch failures faster, which is exactly why his non-coder MBA students outperformed expectations. They weren't AI whisperers. They were people who already knew how to write a product brief, a shot list, or a five-paragraph military order, and it turns out those formats work remarkably well as AI instructions.

What Mollick is really describing is a shift in what scarce talent looks like. Management has always existed because good labor was limited and expensive; AI flips that, making execution abundant and cheap while leaving a new bottleneck — knowing precisely what to ask for and what good looks like when it comes back. He points to software engineers at frontier AI labs who already say their day job has shifted from writing code to managing coding agents, and predicts that shift is coming for plenty of other fields that assumed they were safe because their work wasn't 'technical.'