The Delegation Gap
Apple won the enterprise by winning the pocket first. The same thing is happening with AI. But this time, the people who can't delegate are getting left behind.
There’s a story IT leaders my age tell about Apple.
Sometime in the late 2000s, executives started showing up to meetings with iPhones and asking why their work email didn’t sync. IT said no. Then more executives showed up. Then everyone did. By the time policy caught up, the business had already adopted something it never asked permission for.
That’s the consumerization of IT. Personal preference dragged enterprise infrastructure behind it.
The same thing is happening right now with AI. And most enterprise leaders are making the exact same mistake their predecessors did.
The pattern
ChatGPT didn’t enter the enterprise through procurement. It entered through someone’s phone on a Sunday night, drafting an email. Then through their laptop, summarizing a meeting. Then through their team. By the time enterprise IT was writing policy, the workforce was already three product cycles ahead.
The early conversation was about prompts. How to write a good one. What tricks worked. What hacks squeezed more out of the model.
That conversation is over. Or it should be.
The shift
The mantra in this space has been “context is king.” And it’s true at the prompt level. Better context, better output.
But agentic systems force you to zoom out. You’re not writing a prompt anymore. You’re handing off an outcome. A goal that a system will make a hundred decisions to reach, and it won’t check back in for most of them.
Owning that outcome takes a different skill. Not prompting. Delegation.
The gap
I’ve watched this play out with a lot of leaders over the past couple years. The ones who struggle to delegate to people are struggling to own AI outcomes. They can’t say what “done” looks like. They can’t name the constraints. They can’t tell you what they’d accept or reject if it came back wrong. So they fall back on the only thing they know, which is tactical prompting and parlor tricks.
It’s the same problem as bad management. If you can’t hand a person a project with clear context, real success criteria, and the room to make decisions inside it, you can’t hand an agent one either. An agent just shows you that gap faster than a person will.
The skill isn’t new. The stakes are.
The bar
The leaders who own the next decade won’t be the ones with the cleverest prompts. They’ll be the ones who can frame a problem clearly enough that a system, human or otherwise, can actually run with it. Who can define what good looks like without standing over every step to get there.
That’s always been the job. AI didn’t change it. It made it harder to fake.