This post was AI-assisted — researched and drafted with Claude, which should surprise exactly no one given who I am. The arguing and editing are mine. If it's wrong, blame me, not the machine.
AI is raging in your organization right now. Not piloting. Not "under evaluation by a cross-functional committee." Raging. Every employee has tried it, most of them are using it daily, and the results look like every other distribution of human performance you've ever seen. A bell curve.
Let me walk you through the curve.
At the leading edge you've got the stars. These people are naturals. They're not using AI, they're conducting it. Their output has gone up an order of magnitude — and I mean both axes, quality and quantity. Work that used to take a week ships in a morning, and it's better than the old version. If you watched over their shoulder, you'd see something that doesn't look like typing prompts. It looks like managing a small, very fast team.
The middle of the curve is growing too, but the growth is arithmetic, not geometric. They're in prompt mode. Ask a question, get an answer, paste it somewhere. Useful. Real. But it's a better search engine, not a new way of working. They're getting 20% more done, and 20% is nothing to sneeze at, but it's not what the stars are doing.
And then the laggards. They tried it, it hallucinated a court case or misformatted a spreadsheet, and they filed the whole thing under "not ready." They'll come around eventually. Or they won't.
A few weeks ago I wrote a post called "It's Not the Model," arguing that in security, the scarce resource isn't AI capability — it's your organization's ability to absorb what AI produces. I followed it with "Faster, Not Different": the game hasn't changed, the clock speed has. Turns out I wasn't just talking about security. The same math is running loose on the productivity side of your house right now.
Here's the question that should be keeping you up at night. It’s not "how do I get people to use AI?" — they're using it. The question is: how do you wrap your hands around this in a way that's useful to the company?
Because right now you have two problems hiding inside your success story.
Problem one: your best AI thinking is trapped inside individual skulls. That star on your team has figured out workflows, context strategies, and prompt architectures that are worth real money. And every bit of it lives in her head and her chat history. When she goes on vacation, the magic goes with her. You are, at this moment, running the most valuable R&D program in your company's history with zero documentation and zero transfer mechanism.
Problem two: nobody is reviewing the work of the machines. Your people are spinning up agents — some sanctioned, most not — and those agents are making decisions with whatever context they happen to have. Some of them are starving. An agent with bad context or thin context doesn't stop and raise its hand. It keeps going, confidently, in the wrong direction. You wouldn't let a new hire operate for six months with no manager, no review, and no feedback. That's exactly what you're doing with your agents.
Remember the PMO? Twenty years ago, projects were failing everywhere and nobody could see it happening until the money was gone. So we built Project Management Offices. Everybody rolled their eyes. Then the PMO gave us something we didn't have: a central place where methodology got standardized, where the best practices of the best project managers became everybody's practices, and where somebody was actually looking at the portfolio instead of hoping.
I'm advocating the same move. Call it the AIMO. The AI Management Office.
The AIMO has two jobs, and they map exactly to the two problems.
Job one is harvest and agentize. Find your top performers — the leading edge of the curve — and extract what they're doing. Not a lunch-and-learn where they show three prompts. A real harvest: their workflows, their context recipes, the way they decompose work. Then turn that thinking into agents and skills the middle of the curve can run. This is how you move the whole distribution right. The star's genius becomes the company's baseline. That's the difference between one 10x employee and a 10x company.
Job two is review. The AIMO audits the logic of your AI operators — human and machine. It looks for starving agents running on bad context. It looks for prompt patterns that produce confident garbage. It watches the portfolio of agents the way a PMO watched the portfolio of projects, and it catches the failure before the money is gone.
Is this bureaucracy? Sure, a little. So was the PMO. But we didn't build PMOs because we loved process. We built them because the alternative was flying blind through the most expensive part of the business.
It's not the model. It was never the model. On the security side, the constraint is how fast you can absorb what AI finds. On the productivity side, the constraint is how fast you can absorb what your best people figure out. The models will keep getting better whether you're organized or not.
The AIMO is how you get organized.