The Most Expensive Sentence in Business

A few weeks ago, I visited the Fort Knox of plutonium.

That's not my phrase; it's how the people who work there describe it. No, I do not have proof. It's the kind of facility where, if you take out your phone, you will get shot. (So I left mine in the car.)

I was leading a series of workshops with a nuclear waste management company, surrounded by some of the smartest people I've ever met. At one point, an engineer took me on a two-hour driving tour of the site, and over those two hours I understood maybe one minute of what he said… and that minute came in one-second increments. (Radiological this, vitrification that. I nodded a lot.) These folks operate at a level of technical depth I can barely describe.

Which is why “the sentence” stopped me cold.

After an energizing (dare I say “eye-opening”) session, one of the most advanced engineers in the room pulled me aside. He was gracious. Thoughtful. Sincere. And then he gave me “the sentence.” You might have heard it, or even said it, yourself:

"But in some cases, AI is just not there yet."

He said it the way you'd comment on the weather. A little shrug. A little "come on, we can all admit this, right?" with a wink and twinkle in his eye, like we both knew that was the truth.

And I get it. I've said versions of that sentence myself. Sometimes it's even true. There are real limits around accuracy, security, long-horizon autonomy, and making up nonsense with astonishing confidence. (Technical term.)

But this time I pushed back. Hard. (Harder than a polite guest probably should. But they'd already shown me the plutonium, so I figured we were close.)

I told him: "You need to stop saying 'AI is not there yet.' You need to start saying, 'I'm not there yet.'"

Notice why the first sentence is so much easier to say. "AI is not there yet" costs nothing. It's a status report on someone else's product. "I'm not there yet" is an admission. The first protects you. The other has the power to change you, if you let it.

The Sentence That Lets Us Off the Hook

"AI is not there yet" sounds discerning. It sounds technically mature. But watch what it actually does: it functions as a permission slip.

If the models aren't there, waiting is prudence. If the models aren't there, my current workflow is fine, my skepticism is rigor, and the future is conveniently somebody else's job. I'll just wait for Claude 6. (Or 7. There's always another version to wait for.)

That's why I've started calling it the most expensive sentence in business. The price never shows up anywhere you track. A failed project at least generates a lesson; a wasted license at least shows up on an invoice. This sentence bills differently — in projects never attempted, questions never asked, and weeks your competitors spend practicing while you wait for a version number to save you. Comfort now. Compounding, invisible bill later.

Here's the logic that unravels the permission slip. The output of an AI model is substantially a function of the inputs the human provides — the context, the examples, the expectations, the feedback. If you accept that premise (and everything in my research says you should), then a disappointing output isn't a verdict on the model. It's a reflection of the collaboration.

In my book, I make the case that the most useful way to think about AI is as a capable colleague who thrives under good management. Now picture how most people actually work with it: a vague assignment in one sentence, no background, no audience, no examples of what good looks like, no review, no redirection. One glance at the first draft, and the verdict comes down: "Not there yet."

Imagine doing that to a human teammate: walking up to a brilliant new colleague on day one, briefing them in nine words, disappearing while they work, then announcing to the office that they're not ready for real work.

That would be negligence, masquerading as management.

It's not a technology problem. It's a management problem. (To the ~3% of my audience who pick up on AI-writing-ticks: I fully realize the “it’s-not-this-it’s-that” construction is supposedly a tell-tale AI sign. I happen to have the text message to myself to prove I authored those words. I like them enough to have left them despite sounding AI-ish.) And the cycle is self-sealing: low expectations produce lazy collaboration, lazy collaboration produces mediocre output, mediocre output confirms the low expectations. Then everyone goes home feeling very reasonable.

I'm Not There Yet, Either

Before this sounds like a lecture, let me confess: I don't think I'm getting 10% of the value from AI that I could be. And I'm trying. I teach this for a living. I wrote a book about it. "I'm not there yet as a collaborator" isn't an insult I hurled at a nuclear engineer — it's the sentence I say about myself, weekly.

If you think I'm exaggerating the size of the gap, here's Ethan Mollick on Simon Sinek's podcast:

"The other thing I worry about is people don't know how good these systems are. They are better than you think... I have a doctorate. I've been a professor for a while. I publish in journals. The AI writes a pretty damn good academic paper now — not just a parody of an academic paper — if you give it a data set to work from. It is proving math at a level where you really need to be one of the best math professors in the world to know whether the system is right or wrong. It's often right at this point. It is doing really good images and marketing work that beat professionals."

Proving math that only the world's best mathematicians can even check. And the thing that was "not there yet" for the engineer? A workflow he'd tried a handful of times with a few sentences of instruction.

Which raises a question I don't know how to ask politely, so I'll just ask it: what have you actually tried? Have you given it your real context? Not a summary, but the actual documents? Have you shown it an example of what great looks like? Have you let it interview you before it attempts the work? Have you asked it to critique its own first draft? Have you done that? And that? And that?

You have to exhaust the list before "AI can't do this" is a finding. Until then, it's a forecast. And it's not a forecast about the AI.

The Wildest Part: We're Under-Dreaming

Where this gets fun with teams is, I frame this flip as a thought experiment:

Assume, just for a moment, that it's not the AI. Imagine you were completely confident in its abilities. What would you ask it to do?

You know what I get? Blank stares. Most people don't have a dream project.

That's the deeper discovery hiding underneath "not there yet" — a crisis of ambition. Our imaginations were trained under old constraints. We learned what was "reasonable" when intelligence was scarce, analysis was expensive, and every new project required headcount, budget, meetings, and pain. Then AI showed up and changed the cost structure — and our ambitions didn't update.

So we use a 10x amplifier to pursue 10% improvements. We ask for a faster memo when we could ask for a new operating rhythm. We ask for a cleaner deck when we could ask which customer problem nobody has modeled yet.

This is where the sentence sends its real bill. Not the workflow that disappointed you, but in the work you never attempted.

The engineer didn't have an AI capability problem. He had a portfolio of small questions.

Try This Before You Say "Not There Yet"

Three experiments, all runnable this week:

1. Rewrite the sentence — for one week. Every time you catch yourself (or a colleague) saying "it can't" or "it's not there yet," swap it: "I haven't figured out how to ask yet." You don't even have to believe it. Just notice how differently you behave for the next ten minutes. And if the first response is generic, don't conclude the system is generic. Conclude the conversation is young.

2. Earn the right. Take one task you've already decided AI can't do, and paste this: "I tried to get you to [task] and was disappointed by the result. Interview me, one question at a time, about what I actually wanted — the audience, the stakes, what great looks like, what I'm secretly worried about. Then tell me exactly what inputs you'd need to deliver this at a world-class level, and let's run it again."

3. Commission your dream project. Paste this: "Assume you are fully capable of anything a world-class team could accomplish. Interview me to uncover a dream project I've never pursued because it always felt out of reach. Then propose how we'd start this week."

The More Honest Sentence

I don't want us to become AI maximalists who pretend every limitation is user error. Some things are genuinely not there yet.

But a lot more is there than we're currently reaching. And "AI is not there yet" is too often where curiosity goes to die. It ends the experiment right at the moment the human would have to improve.

The more honest sentence stings, and that's exactly why it works. It isn't shame. It's agency. If AI is not there yet, all you can do is wait. If you are not there yet as a collaborator, you can practice. You can brief better, steer better, review better, expect more — and finally name the dream project you haven't let yourself say out loud.

"AI is not there yet" may be the most expensive sentence in business.

Unless you let it become the most useful one.

Related: Do You Feel The Pain Of Idle Agents?
Related: The Answer Is Yes. What's Your Question?
Related: The Bug Is the Brief

Join over 52,147 creators & leaders who read Methods of the Masters each week.