The Most Expensive Mistake in the AI Era Isn't Tokens. It's the Wrong Question.

A team at a large financial services firm had spent months getting AI deeply embedded in how they worked. On the surface, things were moving faster, much faster.

Then a client brought them a real problem: an acquisition target in a market they didn't know well, with incomplete data and no clean comparable to work from. The team did what they'd been doing for months. They built the model, identified comparables, pressure-tested the assumptions, and landed on a recommended price. AI helped them move fast. The deck was sharp, the logic tight, the recommendation confident.

Nobody had asked the question underneath the spreadsheet: could the client even operate in this market at all? The regulatory environment, the customer behavior, the distribution model, none of it worked the way it did back home. The team had answered the question in front of them. No one had stepped back to ask whether it was the right one.

This wasn't a one-off lapse. It was the visible symptom of something that had been building quietly for a year: as AI absorbed more of the team's thinking, they'd gradually stopped questioning what it handed back. The story isn't really about one bad deal. It's a preview of what happens to an entire organization's judgment when nobody notices it eroding until the moment it's tested.

A Small Version of a Global Problem

This team's blind spot isn't an isolated incident. It's a small, contained version of a much bigger one.

Years ago, I gave a TEDx talk called "How to Ask Better Questions." At the time, I was researching the impact questions have on our careers, our growth, and our relationships. In preparing for that talk, I came across the Millennium Project, a global think tank of scientists, policymakers, and universities that publishes an annual State of the Future report tracking fifteen challenges facing humanity: clean water, peace and conflict, energy, population growth. And sitting right there on that list, next to problems that could reshape the planet, was something I didn't expect: "capacity to decide."

Decision making, named as a global challenge on par with the physical resources that keep civilization running. Not because we lack information. Because we're drowning in it, moving too fast, in a world too interconnected for anyone to fully grasp on their own. We tend to assume our problem is not having enough to go on. We're actually facing something closer to a systemic failure of judgment, and AI is about to make that failure faster and harder to spot, not easier.

Why This Is Your Career, Not Just Your Company's Problem

Here's the part I want to state plainly instead of leaving you to infer it. As AI gets faster and more capable, your value at work will be measured less by the answers you produce and more by the questions you ask. When anyone can generate a competent answer in seconds, the answer stops being the scarce thing. The question, the one that finds the real problem, catches the wrong assumption, or points the AI somewhere more useful, becomes the actual skill.

There's a name for what happened to that financial services team, and it isn't unique to them. Researchers Raja Parasuraman and Dietrich Manzey call it automation bias: a well-documented tendency to trust an automated system's output and quietly do less independent checking, not more, once that system has offered a plausible-looking answer. It shows up in experts as often as novices. It doesn't go away with training. It gets worse under time pressure, which is exactly the condition most of us are working under.

So let me ask you something. How many good questions did you ask yesterday? Not the polite kind that opens a meeting. By good, I mean the kind that could have changed your mind, or someone else's. When was the last time a question you asked caused a team, a client, or a friend to change direction mid-conversation?

If you went quiet just now, you're not alone, and it's not a knock on you. Almost nobody was ever taught to ask questions as a discipline. We're taught to have answers. Questioning gets treated as a thing you do when you don't know something yet, not a skill you keep sharpening for the rest of your career. Most of us ask questions that stay on the surface, stop at the first plausible answer, and never actually go looking for a different perspective. That's normal. It's also exactly the gap that's about to get expensive, because AI will happily meet you at whatever level your question was pitched at, and stop there with you.

Your Brain Is a Search Engine

This isn't only about corporate boardrooms. It's about how any system built on language, a person or a model trained on human text, responds to the frame it's handed. A restrictive question narrows what gets retrieved. An open one widens it. That's true whether you're prompting a model or talking to your teenager.

Your brain works like a searchable database. It stores information, and when you ask it a question, it goes looking for an answer. But the answer it hands back depends entirely on the quality of the question you asked. Ask yourself "Why do I always procrastinate?" and your brain will happily search its archives and hand you an excuse: you're lazy, you have too much going on, you never know where to start. Ask instead, "How am I going to get this done today?" and your brain switches modes. It stops looking backward for a reason and starts looking forward for a plan.

This isn't a trick of positive thinking. Research by neuroscientist Andrew Newberg, author of Words Can Change Your Brain, shows that negative or accusatory words activate the brain's fear circuitry, flooding it with stress hormones and partially shutting down the logic and reasoning centers in the frontal lobes. Tell your team "What should we do about this?" and you've unintentionally narrowed their thinking, because "should" implies there's one correct answer waiting to be found. Ask "What could we do?" instead, and the same team suddenly has room to generate real options. The word alone changes what gets retrieved, whether the system doing the retrieving is a brain or a language model.

Explorer or Confirmer

Before you can ask a good question, you need to know which mindset you're asking from. Psychologist Marilee Adams, in Change Your Questions, Change Your Life, calls this the Learner versus the Judger. I think of it as Explorer versus Confirmer, because that's what's actually happening in an era of AI.

Explorer mode is mapping territory you haven't seen yet. You're genuinely scanning for what you don't know, checking for patterns and hidden structure, and willing to change your mind when new information shows up. Confirmer mode is defending ground you've already staked out. You're subconsciously looking for data that fits a conclusion you've already reached, and your tone and body language broadcast urgency or judgment, whether you intend it or not.

Think back to the last time you sensed someone getting defensive with you mid-conversation, before they'd said a word that justified it. That feeling wasn't imagined. People feel your mindset before they hear your words. If you're in Confirmer mode, your tone sharpens, your tempo quickens, your body stiffens, your eye contact turns sharp. The other person senses it and pulls back, which gives you short, guarded answers, which confirms whatever suspicion put you in Confirmer mode to begin with.

Try this before your next hard conversation. Write down, quickly, how you see the situation, how the other person might see it, how someone with no stake in it might see it, and what unwritten rules or incentives might be shaping the whole thing. Say you're walking into a tense budget meeting expecting a fight. Running through those four lenses might surface something you hadn't considered, like the possibility that your colleague's resistance isn't stubbornness, it's fear of losing something her team depends on. Seeing it that way is Explorer mode in practice: you widened the list of possible explanations instead of defending the one you walked in with, and that alone can change your tone before you've said a word out loud.

And this dynamic isn't staying confined to human conversations. AI is getting fast at reading the same signals: tone, word choice, hesitation, even eye contact through a camera, and using them in real time to judge a person's disposition and intent. The Explorer/Confirmer difference won't just shape how people read each other anymore. It'll shape how the machines we work with read us too.

The Same Reframe, Everywhere

Once you're asking from the right mindset with the right words, the pattern is remarkably consistent across every part of your life:

Context	Confirmer question (narrow)	Explorer question (wide)
Your team	“What should we do about this?”	“What could we do about this?”
A friend	“Why does this always happen to me?”	“What can I learn from this?”
Your boss	“How can I help?”	“If I did this, would that be helpful?”
Your kid	“How was school today?”	“What made you smile today?”
Your partner	“What’s the matter?”	“How can we resolve this together?”
Someone you’re coaching	“What’s wrong here?”	“What do you want to be true six months from now?”
An AI tool	“What should this model tell me?”	“What could I learn by prompting it differently?”

Same topic, same intent, completely different door opened. I think this is the part that gets missed most: we treat "asking good questions" as a workplace skill, something for consultants and interviewers. It isn't. It's a parenting skill, a marriage skill, a friendship skill, and now, a prompting skill.

The Waterline Trap

Once you're pointed in the right direction, you need somewhere to send the questions. I use a systems-thinking tool called the Iceberg. Picture an actual iceberg: only about ten percent shows above the water. The rest, the part that determines whether the ship gets through, is hidden underneath. The Situation is the tip: what's happening, what you can observe directly. Below that sit the Patterns: what's been trending over time, the forces that produced this particular moment. Below that are the Structures: the processes, policies, and incentives quietly shaping those patterns. And at the very bottom are the Mental Models: the beliefs and assumptions that caused someone to build those structures in the first place.

Most meetings, most arguments, most performance reviews, stay at the Situation level. We describe what happened and stop there. But the leverage is almost always further down. A team keeps missing deadlines (Situation). Workloads have crept up for two years (Pattern). There's no formal capacity planning process (Structure). Somebody believes that saying yes to everything is what makes a good team player (Mental Model). You can reorganize the workload all day. Until that belief gets named out loud, the pattern comes back.

This is also where careers get made or capped. People who only ever operate at the Situation level get described as responsive, someone who handles what's in front of them. People who consistently work the Structure and Mental Model levels get described as strategic, and they're the ones handed the bigger, more ambiguous problems, because they've shown they can find what's actually driving the outcome instead of just reacting to it.

The Iceberg Meets AI

Go back to the financial services team. Their failure maps cleanly onto the Iceberg, and it's exactly why I built a set of five thinking practices for working alongside AI.

Question the Frame

The cognitive shift: moving from answering the question in front of you to interrogating why you're asking it in the first place.

• At work: the team never asked whether valuation was even the right question.
• At home: a parent stops asking "how do I get my kid to do their homework" and starts asking why the homework feels pointless to the kid in the first place.

Read the Pattern

The cognitive shift: moving from reacting to an isolated event to tracking the system that keeps producing it, and separating system speed from system health.

• With AI: AI made the team faster for twelve months straight. Faster was never proof the team was thinking better, and nobody checked.
• With relationships: noticing that every check-in with a friend turns into a crisis call, and asking what's producing a pattern of only connecting during emergencies.

Find the Leverage

The cognitive shift: moving from using AI as a quick workaround to using it to change the structural problem itself.

• At work: running the valuation faster wasn't the leverage point. Asking whether the client belonged in that market at all was.
• At home: nagging a partner about chores won't fix anything until you address the unspoken structure of who's assumed to notice what needs doing.

Surface the Assumption

The cognitive shift: treating AI's confident output as a mirror of your own assumptions and historical bias, not a window onto objective reality.

• With AI: an output that "looks objective" usually just reflects the data and incentives it was trained or fed on.
• With people: when someone says they're "just not a numbers person," the useful question isn't about their math skills, it's about what experience taught them to believe that.

Define the Machine's Role

The cognitive shift: deciding, on purpose, where AI's judgment ends and yours has to begin, before you're in a room with a client and it's too late to ask.

None of this is new thinking dressed up for the AI moment. It's the same Iceberg, applied to a tool that's gotten extraordinarily good at answering the question it's given and not particularly good at telling you it was the wrong one.

It's worth being specific about what's on the other side of this. People and teams who keep practicing these five moves catch flawed market assumptions before they commit resources to them. They test an AI's recommendation against the incentives actually built into their organization before acting on it. They question a data source before it gets codified into policy. None of that shows up as extra effort from the outside. It shows up as fewer expensive surprises and better judgment under the exact conditions where everyone else is moving on the same output from the same model. In a world where the answer is available to anyone, that judgment is the actual competitive edge, for a company and for a career.

Try This Today

You don't need a framework to start practicing this. Pick one of the two. The interpersonal version: have one conversation in the next 24 hours where you catch yourself about to ask a "should" question and swap in a "could." Try it with your team, your kid, or yourself, and notice what comes back differently.

The AI version: next time you're working with an AI tool, don't ask it for a solution. Paste in your current plan and ask it, "What major assumption am I making in this strategy that this data ignores?" Then go look below your own waterline before you look at what it says.

I'd genuinely like to know what you find. Drop a comment with what happened when you tried it, or who you tried it on. I read every one.

Warren Berger once wrote that questioning is like breathing, so basic we forget it's a skill at all. That was true before AI. It matters more now, because in a world where AI can answer almost anything you ask it, the human advantage isn't having better answers. It's knowing which questions are worth asking in the first place, and being willing to admit that skill needs work. That won't just make you a better leader. It'll make you a better friend, a better parent, and a better citizen of a world that keeps getting more complicated.

This is the first in a series. In upcoming pieces, I'll go deeper on each of the five practices above (Question the Frame, Read the Pattern, Find the Leverage, Surface the Assumption, and Define the Machine's Role) one at a time. For now, start with one question swap and see what it changes.

If you want to hear where all of this started for me, here's that TEDx talk I mentioned earlier: How to Ask Better Questions, TEDxMileHigh.