The Delegation Crisis.
Exploring how AI is breaking the delegation frameworks managers spent twenty years building — and what it takes to rebuild them.
For twenty years, Sarah built her career on being a great delegator.
She knew how to break down projects. How to match tasks to people’s strengths. How to give just enough guidance without micromanaging. Her teams loved her because she trusted them. Her bosses loved her because she got results.
Then her company gave everyone AI agents.
Now Sarah spends three hours a day trying to figure out what to give the AI.
She’s not alone.
The skill that broke
Management training spent decades teaching us to delegate to humans. Set clear outcomes. Trust the process. Empower people to figure out the “how.”
That framework assumed the person you’re delegating to:
Understands context without you spelling it out
Can ask clarifying questions when confused
Knows when to escalate and when to problem-solve
Brings judgement to ambiguous situations
AI agents can do none of these things reliably.
Which means everything we know about delegation is suddenly obsolete.
The “well-defined” problem
Here’s what’s breaking managers right now: they don’t know which tasks are “well-defined enough” to delegate to AI.
Ethan Mollick recently ran an experiment. He had MBA students build startups in four days using AI agents. The ones who succeeded had one thing in common: domain expertise.
They knew what “good” looked like. They could define deliverables precisely. They could evaluate AI output and give useful feedback.
The ones who struggled? They tried to delegate things they didn’t fully understand themselves.
Turns out “I’ll know it when I see it” doesn’t work with AI.
With human reports, you could say “make this presentation compelling” and trust them to figure out what that means for the audience.
With AI, “compelling” is meaningless. You need to specify: compelling to whom? What outcome? What tone? What length? What format?
The more precisely you can define the task, the better AI performs.
Which surfaces an uncomfortable truth: you can only delegate to AI what you already understand deeply.
The expertise paradox
This creates a paradox.
The tasks you understand well enough to delegate to AI are often the tasks you’re best at. The ones where your judgement is sharpest.
The tasks you’d most want to delegate — the ambiguous, exploratory, “figure this out for me” work — are exactly the ones AI handles worst.
So you end up delegating your strengths and keeping your weaknesses.
Which is backwards.
Traditional delegation worked because you gave junior people the well-defined tasks (they learned by doing them 1,000 times) and you kept the ambiguous strategy work (which required judgement).
AI delegation inverts this. You give AI the well-defined work. You keep...everything else.
Including the stuff you’re not actually good at.
The control paradox
Here’s the second problem: managers are terrified of both extremes.
Delegate too little to AI? You are wasting the tool. Your boss sees other teams moving faster and wonders why you are not.
Delegate too much? You lose control. The AI makes decisions you would have made differently. Mistakes slip through because you are not reviewing carefully enough.
The sweet spot is narrow. And it’s different for every task, every manager, every context.
Sarah told me she now spends more time thinking about delegation than she ever did with human reports.
“With people, I knew the framework. Set outcomes, trust the process. With AI, I am reverse-engineering every task to figure out if it’s ‘ready’ to hand off.”
She’s not managing anymore. She’s task-engineering.
The judgement gap
The real crisis is this: we trained managers to delegate outcomes. AI needs process.
Humans are outcome-oriented delegators. You say “increase conversion rate” and trust your marketer to figure out whether that means A/B testing, new copy, funnel redesign, or better targeting.
AI is process-oriented. It needs you to specify the exact steps: “Run an A/B test on homepage headline. Test 5 variations. Minimum 10,000 visitors per variant. Report confidence intervals. Recommend winner.”
The managers who are thriving right now? They’re the ones who were always a bit micromanage-y. The ones who naturally broke tasks into discrete steps.
The “empowering” managers — the ones who gave autonomy and trusted judgement — are struggling.
Their instincts are wrong for this moment.
What this means
If you are a manager feeling lost right now, you are not broken. The skill you spent twenty years building is suddenly mismatched to the tool.
Delegation used to mean: trust people to figure it out.
Now it means: be precise enough that a machine can execute.
Those are opposite skills.
The good news? This is learnable. But it requires unlearning a lot of what made you successful.
In Part 2, we will talk about what delegation looks like when you have three layers: you, AI, and the humans who report to you.
Because that is where it gets really weird.
The new org chart has arrived. And nobody knows how to draw it yet.