Management Is the AI Skill

Ethan Mollick ran an experiment at Wharton: executive MBA students with little to no coding experience had to build working startup prototypes in four days using Claude Code and other AI tools.

What they produced in a couple of days was, by Mollick's estimate, an order of magnitude further along than previous cohorts got in a full semester.

The interesting part wasn't the AI. It was what separated the students who built something real from those who flailed: management skills. Not prompt engineering. Not technical knowledge. The ability to break a problem down, hand off the pieces, and tell whether what came back was any good.

The Delegation Equation

Mollick identifies three variables that determine whether delegating to AI makes sense. Boiled down, the math looks something like:

Delegate when:

The variables:

• Human Baseline Time — how long it takes you to do the task
• Probability of Success — how likely the AI produces acceptable output
• AI Process Time — time to prompt, wait, and evaluate

Say a task takes you seven hours. AI has a decent shot at it — maybe three out of four attempts — and each attempt takes an hour to prompt and review. Even accounting for the failures, you come out ahead. The math works because the wins more than cover the losses.

You can shift that equation by getting better at delegation.

Your SOPs Are Already Prompts

Every field has invented paperwork for delegation. The military has Five Paragraph Orders. Film directors use shot lists. Software devs write PRDs. Consultants scope engagements. Architects create design intent docs.

Different formats, same function: getting what's in one person's head into instructions someone else can execute.

They work as AI prompts because they were always solving the delegation problem. AI just made the feedback loop faster.

If you already know how to write a clear spec, brief, or scope document, you already know how to prompt AI effectively. You don't need to learn a new skill. You need to apply one you already have.

The Evaluation Problem

Mollick's framework assumes you can evaluate AI output.

Many people can't.

AI produces work that looks competent. Proper formatting, confident tone, reasonable structure. But "looks competent" and "is correct" are different things. A doctor can spot when AI hallucinates a drug interaction. Can your marketing team spot when AI hallucinates a market trend?

The danger isn't that AI produces bad work. It's that bad work looks exactly like good work to someone who can't tell the difference.

This is where subject matter expertise earns its keep. The expert knows what to ask for and can see when something's off. Without that, you're just accepting whatever the AI generates and hoping.

What This Means for Organizations

The companies getting the most from AI aren't the ones with the best tools or the biggest budgets. They're the ones with:

1. Clear processes — documented well enough that delegation is possible. As Mitchell Hashimoto put it: the bottleneck is "improving my own workflows and tools so that I can have a constant stream of high quality work to delegate. Which, even without AI, is important!"
2. Subject matter experts — people who can evaluate AI output in their domain
3. Management culture — experience setting goals, giving feedback, and iterating

Technical AI skills matter less than you'd expect. The MBA students in Mollick's experiment had zero, and they still shipped working prototypes in four days. What they had was experience managing people and projects.

AI amplifies management skills. If you're good at delegation, you get leverage. If you're bad at it, AI just generates more garbage faster.

The Uncomfortable Implication

If management is the AI skill, then the professionals most threatened aren't the ones whose jobs AI can do. They're the ones who never learned to delegate because they always did everything themselves.

The solo expert who hoards knowledge. The senior IC who "just does it faster than explaining." The founder who can't hand off because nobody else "gets it."

AI doesn't replace these people. It routes around them. Their colleagues with clearer processes and better delegation skills will simply get more done.

---

Most AI training teaches people the tools. That's the relatively easier part. The hard part — delegating well, evaluating output, knowing when to override — is management. As Mollick puts it: "The skills that are so often dismissed as 'soft' turned out to be the hard ones."

If yours does and you're still stuck on where AI fits, let's talk.