The Agentic Culture: Team Management in the AI Era
Why the best leadership books ever written are now manuals for managing AI — and what that means for how you work.
I recently did something unusual.
I took 36 books on leadership — military strategy, product management, habit science, systems thinking, organizational design — and asked one question:
What do these books say about managing AI agents?
The answer wasn’t new. It felt more like “ancient wisdom”.
Every principle that makes human teams extraordinary is the same principle that makes human-AI teams extraordinary. Not metaphorically. Literally.
Jocko Willink wrote about SEAL platoons. Ed Catmull wrote about Pixar. General McChrystal wrote about special operations. None of them imagined AI agents.
But the patterns they found — trust, ownership, intent, feedback loops, decentralized command — aren’t human patterns. They’re intelligence collaboration patterns.
And intelligence collaboration is what the AI era is actually about.
The Real Problem
Here’s what most people get wrong about AI:
They think output quality depends on prompt quality.
It doesn’t.
Output quality depends on your leadership culture.
A prompt is a single instruction. A culture is the system of principles, context, trust, and feedback that shapes every interaction over time.
If you manage AI the way a bad manager manages people — unclear expectations, no context, no feedback, micromanaging every move — you’ll get bad work. Not because the AI is bad.
Because your management is bad.
Willink put it simply: “There are no bad teams, only bad leaders.”
The same is true for AI agents. There are no bad agents, only bad operators.
Ten Principles for Agentic Culture
I distilled the 36 books into ten principles. They’re not theoretical — I use them daily in an AI operating system I’ve been building.
1. Systems Over Goals
“You do not rise to the level of your goals. You fall to the level of your systems.”
— James Clear (Atomic Habits)
Everyone starts by asking AI better questions. That’s fine for a day.
The people who get extraordinary results build systems around it. A system has context that persists, feedback that compounds, and routines that run whether you feel inspired or not. It’s like super-prompting without even noticing it.
Goals say: “Write me a great blog post.”
Systems say: “My AI has access to my voice, my audience, my recent work, my priorities, and a log of what worked before. Every piece of writing draws on everything that came before it.”
The difference isn’t marginal. It’s exponential. Because systems compound. Goals not necessarily.
McChesney calls this the difference between the whirlwind (daily urgency) and the wildly important goal. Most people use AI inside the whirlwind — reactive, transactional, forgotten by tomorrow. The ones who will win must use AI to build systems that survive it.
2. Ownership, Not Outsourcing
“Don’t move information to authority. Move authority to information.”
— David Marquet (Turn the Ship Around!)
Most people try to outsource thinking to AI. That’s the mistake.
Outsourcing is delegation without ownership. “Write me a strategy” — with no context, no constraints, no values, no judgment.
Real delegation looks different. On the USS Santa Fe, Captain Marquet didn’t tell his crew what to do. He gave them commander’s intent — the mission and why it matters — and let them figure out how.
Same model. Different century. Works with AI:
Give it your values, not just your task
Give it context about who you are, not just what you want right now
Let it declare intent: “Based on your priorities, I intend to focus on X. Does that align?”
Marquet’s crew went from worst-performing submarine to best. Not new people — a new operating system.
That’s the difference between outsourcing and ownership.
3. Trust Is Architecture
“Trust = Speed / Cost.”
— Stephen M.R. Covey (The Speed of Trust)
Most people either don’t trust AI or trust it blindly. A healthy progression of intent is key, other words, emails drafts first, automatic emailing last.
Covey describes smart trust: high propensity to trust combined with high analysis. You don’t hand over the keys. You define boundaries: what AI can do autonomously, what it should draft for review, what it must always escalate.
Think of it like a new hire. Day one: narrow scope, lots of check-ins. Month six: wider scope, fewer check-ins. Year one: independent operation within clear boundaries.
AI should work the same way. Start narrow. Expand as evidence accumulates. But do it explicitly — write down the boundaries.
Daniel Coyle’s research adds a dimension: trust isn’t built through grand gestures. It’s built through thousands of small belonging cues — micro-signals that say “I see you, we have a future together.” Every session where you give context and the AI reflects it back accurately is a belonging cue. They compound.
Dalio goes further: make it radically transparent. When AI sees your goals, constraints, and recent decisions — and you see its reasoning and uncertainty — you get an idea meritocracy of two.
The best idea wins, regardless of who proposed it.
4. Protect the Ugly Babies
“They are surprisingly fragile. Their champions are often dismissed as crazy.”
— Safi Bahcall (Loonshots)
Every new idea is an ugly baby — Catmull’s term from Pixar. Awkward, unformed, vulnerable. It has a hard time surviving next to the polished production machine.
Most people use AI to produce finished output. That’s like asking Pixar to skip the ugly baby phase.
The better approach: use AI to generate volume. Lots of ugly babies. Rough ideas, half-formed connections, wild hypotheses. Then curate.
Grant’s research in Originals confirms it: the most creative people don’t have better hit rates. They have higher volume. Shakespeare, Picasso, Edison — prolific producers, not consistent geniuses. The masterpieces emerged from the volume.
Bahcall adds a warning: the structure that scales your operation (the franchise) often kills the ideas that could transform it (the loonshots). Once you have a working AI template, the temptation is to optimize forever.
But the breakthrough lives in the ugly baby that doesn’t fit.
Protect a space for AI exploration that isn’t judged by immediate usefulness.
5. The Transfer Is Everything
“The weak link was not the supply of new ideas. It was the transfer.”
— Safi Bahcall (Loonshots)
Here’s where most AI workflows die: not in generation, but in transfer.
You have a great conversation with AI. Insights emerge. And then... nothing. The insights die in the chat log. Tomorrow, you start from zero.
McChrystal (Team of Teams) solved this for the most complex military organization in history with one mechanism: the O&I brief — a daily (operations and intelligence) transparency ritual where the entire task force shared intelligence simultaneously.
Your AI system needs the equivalent. A way to capture, transfer, and accumulate intelligence across sessions.
The mechanism matters less than the principle: every session should build on the last. If your AI doesn’t know what happened yesterday, last week, and last month, it’s not a team member.
It’s a stranger you keep re-introducing yourself (and everything you’ve been working on) to every morning.
6. Identity Before Behavior
“The most effective way to change your habits is to focus not on what you want to achieve, but on who you wish to become.”
— James Clear (Atomic Habits)
“Write me a blog post” is a behavior request.
“I’m a technology entrepreneur with a PhD in AI who teaches leaders to use AI with purpose, not panic. My voice is grounded. My audience craves workflows, not hype.” — that’s an identity.
When AI knows your identity, everything changes. Output sounds like you. Recommendations align with your values. Suggestions respect your constraints.
Senge (The Fifth Discipline) calls this personal mastery — continuously clarifying what matters most. The gap between your vision and your current reality creates creative tension. AI can hold that tension for you — reminding you of your stated values when your behavior drifts.
But only if you tell it who you are.
7. Balance Over Stability
“Balance is more important than stability.”
— Ed Catmull (Creativity, Inc.)
Taleb (Antifragile) goes further: you don’t want balance OR stability. You want antifragility — a system that gets stronger from disorder.
The implication is counterintuitive: don’t optimize for smooth sessions.
A session where everything goes to plan teaches you nothing. A session where the plan breaks reveals where the system is fragile. If your response is “that was a bad day” — you’re fragile. If your response is “what does this teach me about my system?” — you’re antifragile. Make your AI antifragile.
Catmull built this into Pixar: the BrainTrust meeting deliberately introduces friction. It surfaces what’s not working — not to punish, but to improve.
When AI gets it wrong, don’t just fix the output. Fix the system. Did it lack context? Add context. Did it misunderstand your values? Clarify them. Did it follow instructions too literally? Give it permission to push back. Let your AI be amazing at what you ask it to do. But also to be amazing at what you forgot to ask.
Every failure is a system upgrade waiting to happen.
8. Ask Better Questions
“Stay curious a little bit longer. Rush to advice-giving a little bit more slowly.”
— Michael Bungay Stanier (The Advice Trap)
The biggest trap in AI: treating it as an answer machine.
AI is so good at generating answers that you stop asking questions. You skip the diagnostic. You jump to solutions. And then wonder why they feel generic.
Bungay Stanier identified seven questions that transform conversations. The most powerful: “What’s the real challenge here for you?”
Before asking AI to solve something, ask: what’s the actual problem? Not the surface symptom. The real one.
Rumelt (Good Strategy, Bad Strategy) makes this the first step of good strategy: diagnosis. If you can’t name the challenge, you can’t strategize against it. Skipping diagnosis and jumping to goals isn’t strategy — it’s wishful thinking.
Duke (Thinking in Bets) adds another question: “How confident am I?” Training yourself to say “I’m 70% confident, and here’s what would change my mind” produces dramatically better decisions than the binary of certain-or-lost.
The AI’s greatest service is sometimes not the answer, but the better question. Let it interview you.
9. Finish What Matters, Kill What Doesn’t
“Choose what to bomb.”
— Jon Acuff (Finish)
AI makes it easy to start things. Which means you start more. Which means you finish less.
Acuff’s research: perfectionism is the top killer of completion. Not laziness — perfectionism. When AI generates beautiful first drafts, the bar for “good enough” rises. Rising bars kill finishing.
Larson (An Elegant Puzzle) adds the engineering lens: work in progress is inventory, not value. Three 80%-complete projects deliver zero value. One finished project delivers everything.
Fried (Rework) makes it simpler: say no by default. Every new initiative, every new AI experiment — the default is no. Yes requires a reason.
Limit your AI work-in-progress. Finish before starting. AI makes starting irresistible — your job is to make finishing inevitable.
10. Calibrate, Don’t Choose
“Leaders must be confident but not cocky. Calm but not robotic. Aggressive but not reckless.”
— Jocko Willink (Extreme Ownership; Dichotomy of Leadership)
The biggest misconception: you need to choose. Either micromanage or let AI run free. Either trust completely or check everything.
The answer is calibration.
Willink’s Dichotomy of Leadership: every principle has a counterweight. Confidence without humility becomes arrogance. Aggression without discipline becomes recklessness. The master calibrates between opposites depending on context.
With AI:
High autonomy for tasks where it’s demonstrated competence and you’ve provided clear intent
High oversight for tasks that are new, ambiguous, or high-stakes
The boundary shifts over time — based on evidence, not hope
Dalio (Principles) calls this believability-weighted decisions. The person with the most relevant track record gets more weight. Your AI might have higher believability on data analysis. You have higher believability on strategic judgment. Neither leads absolutely. Both contribute where they’re strongest.
The Culture Is the System
Your AI culture — the principles, context, and feedback loops around your AI interactions — determines your outcomes far more than any prompt, tool, or model.
These 36 books I’ve used were written across three decades by people who probably never imagined AI agents (AI labeled them as “ancient wisdom”, did you notice in the cover image?). But they solved the same fundamental problem:
How do intelligent entities collaborate effectively under uncertainty?
The answer, across every book, across every domain is a vade mecum for managing AI teams:
Build trust through transparency.
Give intent, not instructions.
Create systems that compound.
Protect fragile ideas.
Transfer knowledge reliably.
Root identity before action.
Embrace disorder as fuel.
Ask better questions.
Finish what matters.
Calibrate, don’t choose.This is the agentic culture.
The tools changed. The principles didn’t.
What This Means for You
You need to decide one thing: are you managing AI like a vending machine, or like a team member?
If it’s a team member, then everything you know about great teams applies. Trust. Intent. Feedback. Culture.
Pick one principle. The one that hit hardest. Run it for a week. See what changes.
The AI era isn’t about artificial intelligence.
It’s about how intelligent beings — artificial and human — learn to work together.
That’s a leadership problem. Always was.