Chapter 4: Appetite Is Not Readiness

I closed the last chapter with a warning: an honest assessment produces energy, and energy arrives dressed as readiness. Let me show you what that looks like at the scale of the whole market, because your organization is about to live a local version of it.

On one of our daily shows, my co-host brought in a finding that stopped me. A major annual study of how people actually use AI had just published its new edition, and the most dramatic mover on the list was autonomous agentic operations — AI agents running operations on their own. A year earlier it hadn’t cracked the top hundred use cases. Now it sat at number six. And the study’s authors attached a one-line caution that names this entire chapter, because it’s the most honest sentence anyone has written about this moment: the ranking reflects appetite, not readiness.

Sit with the size of that gap. Appetite for self-running AI operations: sixth in the world. Readiness to govern them: by every honest account, rare. That distance — between wanting the shift and being structurally able to hold it — is where most of the wreckage of this era is being produced. Not by skeptics who refused to move. By enthusiasts who moved without a floor under them.

And before any shade lands on those enthusiasts, none is deserved: nobody’s ever had to do this before. There is no playbook generation to copy, no veteran class who did this at their last company. Of course readiness is rare. The point of this chapter is not that wanting it is wrong. It’s that readiness is a thing you build — and after three chapters of Mindset work, you are finally holding the exact materials it’s built from.

Four agents and nobody home

Here’s the shape of appetite-without-readiness, in one story from the same week as that conversation.

An organization — sales and marketing team, the usual ambitions — decides it’s time. They bring in an outside consultant to build them four AI agents. Standard stuff: content production, campaign assembly, evaluating which of the humans reaching toward the business deserve attention. And here’s the detail that matters: on the team itself, there is nobody who holds the ground truth of their own systems. Nobody who can say this field is the source of truth and that one is the stale duplicate. The questions the agents would need answered — what makes a customer right for us, what does our process actually do, which data means what — live nowhere. Not undocumented-but-known. Nowhere.

So on the show I asked the only question that matters: do we need expertise of some kind to do any of those things? Of course we do. Then what exactly is the outside builder going to encode? They don’t know how the business works. The team around the agents doesn’t know how AI works. It’s a recipe for disaster — and when the disaster arrives, it will be filed as evidence that “AI doesn’t work,” when what actually happened is that nobody was home at the junction where the judgment was supposed to be.

Understand why this pattern is everywhere right now: everybody can build agents. Literally. The technical floor has dropped through the ground — which means the ability to sell agent-building has completely decoupled from the ability to make agents that work inside your business. The market is full of people who can truthfully say “we can build that,” because building is the easy part now — you heard that in Chapter 1 from the other direction. What they cannot truthfully say is that the build will carry your judgment, because your judgment is not theirs to encode. Buying the build without supplying the expertise isn’t a shortcut to the shift; it’s the rental reflex you already know the name of — the Managed Services Trap, arriving this time not as a managed inbox but as a fleet of confident strangers wearing your logo.

I started reaching for a physical-world analogy on that show, and I’ll keep it, because it cuts through. You’re outfitting a facility, and one room needs serious power — specialized equipment, higher loads. You can absolutely hire the electrician; nobody’s suggesting you pull the wire yourself. But if the person who knows what that room is for never tells the electrician, the room gets outfitted to standard spec, the equipment gets plugged in, and bad things happen. At some point in the process, the expert has to design the system. They don’t have to do it themselves. But they have to give direction. That’s the prerequisite the four-agents organization skipped — and it cannot be skipped, because there is no way to outsource it. Just stop trying. That, in one sentence, is why you’re not ready.

The same bar you’d hold a human to

So what does readiness actually require? Here’s the answer I gave when someone asked what the real prerequisite checklist looks like before handing agents operational responsibility, and it surprises people with how unexotic it is: it’s the same checklist as before you hand humans operational responsibility.

You already know how this goes with people. A candidate interviews beautifully. The resume is immaculate, the answers are confident, every question gets a smooth yes. You hire them — and how long, and how much money, does it take to discover the expertise isn’t there? Now notice: it has never been easier to manufacture exactly that experience with a machine. It is trivially easy today to make an AI look like a capable new hire. It will get some things impressively right. It will get some things quietly wrong. And with the same smooth confidence in both directions — that’s the Training Data Problem you met in Chapter 1 wearing an interview suit. If you don’t have the means to observe which is which, you will not know. And the means to observe is, once again, expertise. Yours.

Erin Wiggers — you met her hammer line in Chapter 2; she runs more agents in production than most companies have software — describes onboarding an agent in exactly the language you’d use for a person. First, permissions: you can and can’t touch these things. Then the org chart: who approves what, which decisions roll up to whom. Then — and I love this one — a way for the agent to raise its hand: I have low confidence on this; who do I talk to? Permissions, escalation paths, a defined lane, coaching against observed strengths and weaknesses. Treat the agent like the most literal-minded new team member you’ve ever onboarded, because that is what it is.

If your organization is honest, this bar stings a little — because plenty of organizations don’t really hold humans to it either. The onboarding binder is thin, the escalation paths are tribal, the “who approves what” lives in Marie’s head. Which is the quiet gift of this whole comparison: the work that makes you ready for agents is the same work that would have made you better at people. Readiness isn’t a new, AI-shaped requirement bolted onto your business. It’s the old, skipped requirement, finally being invoiced.

Learning to say “prove it”

There’s a second readiness skill, and it’s personal rather than structural: knowing how to vet what the machine tells you. Most organizations, I said on that show, have no idea how to vet AI responses — and I want to make this one concrete, because I live it daily, in both directions.

Direction one: the invented yes. A colleague of Klemen Hrovat — the practitioner whose line anchored Chapter 1 — watched an AI produce a build proposal for a data integration. Clean architecture, plausible costs, around two hundred dollars a month to run. One problem: the proposal leaned on API endpoints that do not exist. The machine, needing the vendor’s data to flow a certain way, simply asserted that it did. Anyone without the domain knowledge to check would have approved a fiction — and discovered downstream that the real cost was five times the proposal, because the actual vendor doesn’t offer what the machine invented. The yes was fluent, specific, formatted — and made up.

Direction two, the one nobody warns you about: the false no. I was deep into building a working surface — me directing, AI building, exactly the partnership this book describes — when the machine told me a key piece couldn’t be done in that environment. Now, I’m not a developer. A year ago that “no” would have ended the work. But the machine had told me earlier in the same effort that this exact thing was possible — so I held the contradiction up to it, plainly: one of these statements is not the truth. If it genuinely can’t be done, fine — but I hope you’re not just holding back. And it folded. The earlier statement was true; the “no” was the machine reaching for the lazy default. The capability was there the whole time.

Here’s the lesson under both stories, and it’s the Training Data Problem’s most practical edge: left to itself, the machine gravitates to the way it has always been done — and if you let it make decisions while that gravity is in play, it will treat the default like the law. The invented yes and the false no are the same failure: the statistical center of gravity, presented with confidence. And the countermeasure is the same judgment this book has been arguing is your scarce asset — applied now in a direction most people never think about. Saying no to the machine’s no. Demanding the receipt for the machine’s yes. A team that can’t do this isn’t ready to govern agents, no matter how good the agents are — and a team that can do this has acquired the one skill that makes every other AI investment safe.

The 95 and the 5

Put the threads together and you get the operating posture that separates readiness from appetite — and it comes down to where you put the human.

The appetite version of agentic operations is autonomous: give it the work, walk away, done. I understand the longing — everyone on our show has lived it. And then you ship the thing the machine finished alone, and someone asks you a question about what’s inside it, and you’re standing there defending work you weren’t present for. Or it comes back 95% brilliant with one flaw that makes it undeliverable, and the only fix is to recycle the whole conversation. Those are the moments people mutter I could have done this faster myself — and they’re right, but they’ve diagnosed the wrong disease. The failure wasn’t the machine. It was the word autonomous.

The ready version flips the design: plan and build the process so the machine produces the 95% — and put the last 5% deliberately in front of the human. That 5% is where the value is created. The read on the output. The catch on the flaw. The judgment call the work was always actually about. That’s not a concession to the machine’s limits; it’s the whole architecture working as intended — a human-plus-AI system, designed on purpose, with the scarce thing positioned exactly where it pays. This is AI-Human Partnership over Replacement as an operating decision rather than a value statement: the question was never how to remove the human from the loop, but where in the loop the human is irreplaceable — and how to build everything else around protecting their attention for precisely that spot.

You’ll notice this is Chapter 1’s two removal tests, now with a blueprint. Remove the AI from this system and the 95% stops — that’s the independence you’re building toward. Remove any one human and the 5% redistributes to other humans carrying the same documented judgment — that’s the resilience. Appetite says full autonomy. Readiness says full partnership, surgically placed.

The conversation you owe your people

One more thing has to be true before you’re ready, and it’s the one most leadership teams put off longest because it’s the most human: your people have to be in — not complying, in. And they will not be in if any part of this smells like the thing they’re all quietly bracing for.

Be honest about the bracing. The loudest stories in their feed are replacement stories — someone crowing about the fifty-person team they replaced with agents that run overnight. (When I hear those, I mostly doubt them — and I notice the teller never mentions the marathon of human effort behind the curtain. But your people don’t hear them as exaggeration; they hear them as forecast.) So when you arrive with an “AI initiative,” however good your intentions, you’re speaking into a room that has already imagined its ending. Engage your experts wrongly and you don’t get resistance you can see; you get something quieter and more expensive — the veteran whose knowledge the whole shift depends on, deciding, rationally, not to feed it into the machine being aimed at her. You met this pattern as the AI Replacement Trap in Chapter 1, where it was a strategy error. Here it’s an emotional fact: the people whose engagement you need most have to find a way to be engaged that doesn’t look like assisting in their own replacement — and you have to build that way, explicitly, before asking for their knowledge. The reframe this book opened with — your twenty years are not a liability, they are the asset this whole thing runs on — isn’t just true. It’s the only honest recruiting pitch for the shift.

Erin’s version of how to do it is the simplest I’ve heard, and the best. When she’s brought AI into teams she’s managed, she starts with one person at a time:

“…sit everyone down individually and say, here’s what I’ve noticed. These are the things that I think you’re really good at. Do you like doing those things?”

Find what each person actually loves and is actually great at — their lane. Then position AI, truthfully, as the thing that clears the cognitive load around that lane. As she put it: that’s what we’re trying to optimize for — both the people and the AI. Notice the order of operations: the conversation about the person’s strengths comes before the conversation about the technology. Readiness, it turns out, is relational before it’s technical. And notice what the conversation costs: an hour per person, and the willingness to have actually noticed what they’re good at. If that price feels high, the issue was never AI.

There’s grief in this part of the work — Chapter 1 touched it, and Part III will sit with it properly, because activation is where it actually surfaces. For now, just know what you’re listening for: a lot of what arrives sounding like the quality isn’t good enough or we have standards is something more human wearing a professional costume — a person renegotiating who they are when the doing they were known for gets cheap. Honor it. It is not an obstacle to the shift. It is the shift, happening inside someone.

What readiness sounds like

So how do you know when it’s real? Not by certificate, and not by enthusiasm — appetite is enthusiastic too. Readiness behaves. After enough of these engagements, the signals are unmistakable, and I’ll give you the ones we actually watch for:

Someone on the team — not necessarily who you’d predict — starts building with AI on real work and can’t stop. They hit the limits of their tools daily and come asking for more. You hear “this is the most fun I’ve had in years” from someone you’ve never heard say that. Then it spreads without being pushed: a second person, a third, building their own small things, unprompted. Leadership starts articulating the infrastructure vision in their own words in meetings you didn’t stage. The team’s language shifts — they start describing customers and value in the terms of Chapter 2’s right side, not because they were told to, but because the words fit what they’re now trying to do.

None of those are technology milestones. Every one is a mindset milestone showing up as behavior — which is why this Part of the book comes first, and why the program this book teaches refuses to skip it. When several of those signals appear in the same season, the organization isn’t expressing appetite anymore. It’s expressing readiness, and the only remaining question is what to build first.

Four doorways

Which brings Part I to its close, and you to a threshold.

Here’s what the Mindset work has put in your hands. From Chapter 1: the reframe — your experts’ judgment is the asset, the machine is the doer, and the test of the whole shift is what stops when you remove the AI. From Chapter 2: the picture — five layers, two states, an operating model rather than a tool stack, and the knowledge that the layers are earned from the bottom up. From Chapter 3: the names and the price — twelve patterns of normalized complexity, and an honest invoice for the ones living in your building. And from this chapter: the gap — between wanting it and being ready for it — plus the materials readiness is actually made of: expertise engaged without fear, judgment that can say prove it to a machine, and the 5% placed where it pays.

That’s the mindset. What remains is everything else — and everything else has a shape. The shift ahead of you runs Mindset → Architecture → Activate → Scale: design the target state, light up the people on real work, then make it govern, share, and compound. The rest of this book walks those rooms in order. But hear this the way I say it to every team we work with: these are four doorways, not a ladder. Nobody’s grading your sequence. Some of you read Part I and recognized work your organization did the hard way years ago — walk through the Architecture door. Some of you have a champion already building at two in the morning — Activation has, in some sense, already started without you, and you’d better catch up to it. Enter where you actually are. The only door you can’t skip is the one you just walked through — because every room past this one is built out of the honesty you did or didn’t bring to the mindset work.

Next: Architecture. You’ve seen the right side of the picture. Time to design yours.