You’ve Been Managing People for 25 Years. AI Is Just Your Newest Direct Report.
The skill nobody is teaching, and you already have it.
Tuesday morning, May twelfth. Liz Myers’ name shows up as engaging my An Intentional Age announcement on LinkedIn. I open the AI draft I’d prepared for messages like this.
The draft is good. Generic-good. It says I launched something, here’s the core idea, here are some essays, no pressure if it’s not for you. Clean prose. Reasonable tone, in my voice.
It wouldn’t embarrass me to send it.
But it couldn’t be sent to Liz.
Here’s what the AI gave me:
Subject: Finally did it
Liz,
Launched something I’ve been building for a while. It’s called An Intentional Age.
The core idea: the judgment and pattern recognition you’ve spent decades earning is exactly what makes AI most powerful in your hands. That’s a different story than what most people are telling right now, and I think it matters.
Four essays are already live. Take a look. If it’s worth sharing with someone you know, I’d appreciate it. If it’s not your thing, no pressure.
anintentionalage.com
Bennie
That email is fine. It’d be fine going to a stranger. It’d be fine going to a casual acquaintance.
It’s wrong for Liz.
Liz is a designer I’ve worked with for a few years. She cleaned up the SVG files on my logo here. I wrote her a LinkedIn recommendation last week for her outstanding design collaboration through the past year plus. And right now, she and her husband are designing the next chapter of their lives together. Actively traveling the country. Working from anywhere. Building a life that fits the people they’ve become.
That last bit isn’t in any database AI can read. It lives in conversations we’ve had. It lives in how she described their plans the last time we talked. It lives in the relationship.
The AI draft didn’t have access to any of that. So the AI draft was generic. And generic was the wrong language for Liz.
Here’s what I sent instead:
Subject: Saw you came through
Liz,
Just saw your name come through as one of the first subscribers to An Intentional Age. Thank you.
It means something to see you here, given the design partnership we’ve had. You’ll see this in the LinkedIn recommendation I wrote you last week, but it’s worth saying again away from the public version: working with you made my work better in ways I didn’t always see in the moment. The SVG files you cleaned up on the logo are quietly doing their job too. I’m using that mark on every surface now.
And given the chapter you and your husband are designing together, traveling the country, working from anywhere, building a life that fits the people you’ve become, you’re closer to what I’m writing about than you might realize. The publication is for people doing exactly that kind of designing. That’s you.
If anything sparks, reply anytime. I read every one.
— Bennie
Same recipient. Same launch announcement. Different email.
The AI version positioned me as the author requesting attention. My version positioned Liz as someone already living the story the publication is about. That difference isn’t a stylistic choice. It’s structural. It’s the load-bearing move of the entire exchange.
And the move I made to produce it has a name.
If you're tracking with this, the publication is for you. New pieces every week.
A move I’ve been making my entire career
I’ve been making this move my entire career. So have you, if you’ve been a senior manager for any length of time. It’s the part of management that doesn’t show up in org charts or performance reviews. It’s also the part that decides whether what you send actually does what it’s supposed to do.
A junior writer produces a competent draft. The senior reads it. The senior keeps what’s good. The senior adds the thing only the senior could add. The line the junior couldn’t know, didn’t see, hadn’t built yet. The line that carries the relationship forward, holds the strategic context, names what the recipient hasn’t been told but needs to feel.
That’s the move. It’s delegation discipline. It’s what management is, at its highest level, when the work matters and the recipient matters.
The move runs in four steps. I call it the Operator Paradigm. The name is mine. The discipline is older than my career. It’ll outlast every version of the technology it now manages.
Move one is to specify. This includes the decision of whether to delegate at all. Senior managers triage. Some work you hand off. Some work you do yourself because the stakes are wrong for delegation. Some work shouldn’t be done at all. Specifying starts with that triage and then frames the outcome, the constraints, and the context the agent needs to do the work well.
With Liz, the specification was: outreach to Tier 1 contacts about the Substack launch. The triage was already done. I’d decided in advance that AI could produce the first draft, but the final pass had to be mine, because Tier 1 outreach is relational, not transactional. That decision happened before AI ever touched the task.
Move two is to hand off. Set the boundary of autonomy. Let the agent run. Don’t supervise the typing. The AI produced its draft in seconds.
Move three is to verify. This is where the senior muscle lives. The AI gave me an email. I read it the way I’d read a junior writer’s draft. Not just for grammar. Not just for accuracy. For what isn’t in it. The relationship history AI couldn’t have. The shared chapter that exists only in conversation. The line that has to be added.
In Liz’s email, the line I added was the paragraph about the chapter she and her husband are designing. That sentence couldn’t have come from training data. It came from knowing her.
Move four is to own. I sent the message. My name on it. My reputation. My relationship with Liz on the line. If she’d received the AI’s version, she would have read it as a mass message. She probably wouldn’t have replied. Because I verified and rewrote it, she replied within hours, as a peer. The catch earned the response.
That four-move loop runs everywhere senior people work with AI. A board memo where the agent drafts the analysis and you add the strategic context only you have. A coaching note where AI suggests the structure and you add the one observation that lives in decades of watching that person. A hiring decision brief where AI summarizes the candidates and you flag the cultural fit issue AI couldn’t see. Same four moves. Different cargo. Same muscle doing the work.
Two witnesses from adjacent rooms
Two people are saying versions of this in public right now. Neither of them is a manager. Both of them describe the same pattern from adjacent vantages.
The first is Andrej Karpathy. Former director of AI at Tesla. Founding team at OpenAI. On the No Priors podcast in March 2026, he described what had changed in his work: “In December is when it really just... something flipped where I kind of went from 80-20 of writing code myself versus just delegating to agents to like 20-80.”
What flipped was not the AI. What flipped was Karpathy. He learned to be the manager. He said it directly at Sequoia’s AI Ascent in April 2026: the programmer is increasingly not just a code writer, but an orchestrator of agents. Designing specs. Supervising plans. Inspecting diffs. Writing tests. Creating evaluation loops. The work is no longer typing. The work is delegation discipline.
What makes the discipline necessary is what Karpathy calls AI’s jaggedness. In his words on No Priors: “I simultaneously feel like I’m talking to an extremely brilliant PhD student who’s been a systems programmer their entire life, and a 10-year-old.” The same model that produces work beyond what most humans could match also makes errors a careful human wouldn’t make.
Harvard Business School and Boston Consulting Group studied this directly. Seven hundred fifty-eight consultants. On tasks the AI handled well, the consultants using it produced more work, faster, with higher quality than peers without it. On tasks just outside what the AI could reliably do, the same consultants were nineteen percentage points more likely to produce incorrect work.
Researchers called it the jagged technological frontier. The senior manager has another name for it: trust, but verify.
That move is impossible without judgment. And judgment is the thing decades of management work builds. Karpathy is seeing this pattern from the engineering side. I’m seeing it from the management side. We landed on the same pattern from opposite directions, which is usually how you know a pattern is real.
The second is Simon Willison. Software engineer, writer, and one of the most consistently useful voices on what AI agents can and cannot do.
In February 2025, Willison wrote a short essay surfacing a slide from a 1979 IBM internal training. The slide reads: “A computer can never be held accountable. Therefore a computer must never make a management decision.”
Forty-six years later, the slide is more relevant than it was the day it was made.
Willison returned to the principle on the Heavybit High Leverage podcast in May 2026, this time from his own vantage as an engineering manager who’s spent years trusting and verifying other people’s code. He said: “Claude Code does not have a professional reputation. It can’t take accountability for what it’s done.”
The person managing the AI has to.
Your name on the work
This is where the Operator Paradigm earns its weight.
A computer cannot be held accountable. That principle is older than the consumer internet. In January 2026, California codified it. Assembly Bill 316 took effect. The law says, in plain terms, that anyone who developed, modified, or used an AI system can’t escape liability by pointing at the system. The accountability comes back to a person. There has to be a name on the work.
That name should be the operator’s name. Not because operators are the ones to be punished. Because operators are the ones who can be trusted with delegation. The person who specifies, hands off, verifies, and owns is the person whose reputation rides with the result. That’s what makes them durable in an era when junior workers are handing AI output to senior reviewers without checking it first. That’s what makes them the person their organization comes back to when something goes wrong.
The AI discourse keeps describing this as a young person’s game. It isn’t. It’s a senior person’s game played with a new kind of direct report. The discipline is older than the technology. The discipline is older than most of the people writing about the technology.
Karpathy said something at Ascent that I keep coming back to. “You can outsource your thinking, but you can’t outsource your understanding.”
That’s the operator’s principle in a single sentence. The agent can produce the artifact. Only you can verify whether the artifact carries what it needs to carry.
Liz replied to my email the same morning. “I really do appreciate the push to explore AI you have provided, when probably half of my colleagues of a similar age are giving up or sticking their heads in the sand.”
That reply was the proof. The AI couldn’t have predicted it. My edit earned it.
This week, take one email or message that AI drafted for you. Read it as a stranger would read it. Then add the line only you could add. The line that lives in conversation, not in any training set. Notice what your judgment caught. Notice what you owned.
You’ve been managing direct reports for decades. You know how to do this. The new direct report just happens to be made of code.
— Bennie