August 7, 2026

AI for the Over 40 – Week 31: The Paradox of Expertise: What If You're Hallucinating Too?

For most of my 23-year career as a Microsoft Business Central partner, the right answer really was Microsoft.

When a client asked whether to build a custom feature inside Business Central or integrate with an outside platform, the answer was almost always build it in. That was not spin. On-premise integrations were limited, expensive, and fragile. They rarely delivered the functionality you needed. They meant operating across multiple environments, confusing for users, painful for IT, expensive to maintain. If you could solve the problem inside Business Central, you solved it inside Business Central. That was sound advice, and I gave it confidently because it was genuinely true.

The problem is that the world changed, and my advice did not change at the same speed.

Over the last several years, our clients have shifted aggressively to SaaS. That shift did not just change where the software runs. It changed the fundamental economics of integration. Open APIs turned what used to be six-figure custom development projects into configuration exercises. Platforms like Zapier connect over 8,000 applications with no custom code. Power Automate, Microsoft’s own tool, bridges to over 1,000 systems. MCP servers now let me communicate with external systems in natural language, which was unthinkable even a year ago.

I can now use the functionality of other platforms without ever leaving Business Central. I can build robust, inexpensive integrations that previously would have been complicated and cost-prohibitive. The world that made “keep it all in Microsoft” the right answer does not exist anymore, at least not in the absolute way it used to.

But I was still giving the old answer. Not because I was trying to mislead anyone. Because for so many years the right answer was Microsoft that it did not require me to look hard at alternatives. The habit of confidence outlasted the conditions that justified it.

And I am not the only one. Combine that momentum with Solutions Partner designations that reward Microsoft platform consumption, co-sell incentives that reward growing client platform consumption, and the simple reality that exploring alternatives takes time and money with no corresponding revenue, and it is no wonder the answer continues to be “Microsoft only” for those of us who make our livelihoods selling Microsoft.

The world is changing. Our thinking and our advice do not always keep pace.

A word for what I was doing

In 1986, Princeton philosopher Harry Frankfurt wrote a short book called On Bullshit that drew a distinction I had never considered: bullshitting is not the same as lying. A liar knows the truth and deliberately hides it. A bullshitter produces confident-sounding output without having done the work to know whether it is accurate.

That distinction matters because it describes something more common and more forgivable than dishonesty. It is not that you do not care about the truth. It is that you are operating inside a frame of reference that feels like the whole picture, and nothing in your environment pushes you to check whether it actually is.

That was me. I was not indifferent to whether my advice was good. I genuinely wanted to help my clients make the right technology decisions. But my frame of reference was the Microsoft ecosystem. My expertise, my revenue, my professional identity, my partner incentives, they all pointed in the same direction. And when every signal in your environment confirms the same answer, you stop asking whether the question deserves a wider lens.

I was not lying. I was not careless. But I was confidently advising clients based on a frame of reference I had not pressure-tested against alternatives, because nothing in my professional world rewarded that pressure test, and quite a lot penalized it.

I had a word for when AI does this, produces confident output without adequately grounding it in the full picture. I called it hallucination. I did not have a word for when I did the same thing.

The advice you actually get

Here is the uncomfortable truth that nobody in business wants to say out loud: most of the advice that drives significant business decisions does not come from people who have evaluated the full landscape. It comes from people, good, well-intentioned people, who know their corner of the landscape well and extrapolate from there.

Your ERP vendor recommends their product. Your cloud migration consultant recommends the platform they specialize in. Your golf buddy heard from his golf buddy that a particular approach worked great. The person at the conference who seemed confident told you exactly what you needed to do. The board member’s nephew “works in tech.”

None of these people are lying to you. Most of them genuinely believe they are helping. But every one of them is operating inside a frame, defined by their expertise, their incentives, their experience, and presenting that frame’s answer as if it were the whole picture.

Psychologists David Dunning and Justin Kruger showed that people with limited knowledge in an area tend to overestimate their competence. But the version that applies here is subtler than classic overconfidence. It is not that my peers and I lacked expertise. It is that our expertise was rooted in a world that had shifted beneath us, and nothing in our professional environment flagged the drift. We were not incompetent. We were operating in a narrow frame. And confidence does not come with an expiration date.

The other bullshitter in the room

In 2024, philosophers Michael Townsen Hicks, James Humphries, and Joe Slater published a paper in Ethics and Information Technology arguing that when AI produces false information, calling it a “hallucination” is the wrong metaphor. Hallucination implies the system is trying to perceive reality and failing. Large language models are not trying to perceive reality at all. They are generating text that looks like accurate text without a mechanism for verifying whether it is.

Their title was gloriously direct: “ChatGPT is Bullshit.”

Here is the recognition that changed my thinking: the confident business advisor and ChatGPT have more in common than either would be comfortable admitting. Neither is lying. Both are producing confident output from within a frame, the advisor’s frame shaped by expertise and incentives, the AI’s frame shaped by training data and architecture. Both may be right. Both may be wrong. And in both cases, the confidence of the delivery tells you nothing about the accuracy of the content.

We have a panic-inducing word for when AI does this. We call it hallucination. We write articles about it. We cite it as the reason we cannot trust AI for serious work.

When we do the same thing, give confident advice from inside a frame we have not pressure-tested, we just call it expertise.

The double standard

What would change if we recognized that human bullshit and AI hallucination have more in common than we thought?

For starters, we might notice that we hold them to completely different standards.

Research by Dietvorst and colleagues found that people abandon AI after a single mistake, even when the AI’s overall accuracy remains higher than the human alternative. One wrong answer from AI and trust evaporates. Meanwhile, our trusted advisor can be wrong repeatedly and we chalk it up to complexity. Philip Tetlock’s 20-year study of expert predictions found that the most confident, most frequently quoted experts were wrong more often than the tentative ones, and nobody stopped listening to them.

The pattern is simple: we scrutinize AI with a magnifying glass and accept human advice with a handshake. Not because human advice is more reliable. Because human advice comes wrapped in a relationship, a title, and a track record that may have quietly stopped being current.

How AI Broke My Microsoft-Only Pattern

Here is where the irony gets personal.

In Week 26 of this series, I needed to understand when organizations should use Microsoft Copilot versus Power Automate versus Zapier. Clients were asking. And I realized, uncomfortably, that I had been answering based on my position inside the Microsoft ecosystem rather than on actual comparative analysis.

So I did something I had never done before. I designed a research question carefully, then ran it through Claude, ChatGPT, and Gemini simultaneously, all in their deep research modes. Same prompt. Three different AI platforms. I compared the outputs and synthesized the results.

What came back was a strategic framework that revealed something I should have seen years ago: the answer is not always Microsoft. It is not always not Microsoft either. It is a hybrid architecture, what I now call Intake-Engine-Bridge, where each tool handles what it does best. Power Automate as the deterministic engine for business rules and financial commits. Copilot as the intelligent intake layer for unstructured input. Zapier as the bridge to the thousands of applications that Microsoft’s connector ecosystem does not natively reach.

No single tool was sufficient. The architecture required all three.

I could not have produced that analysis through traditional channels. I would not have hired a consultant to compare my own ecosystem against competitors, since the cost could not be justified and the political awkwardness would have been real. I was not going to run a months-long evaluation project when I had billable work to deliver. And honestly, I was not sure the question needed asking. I had been giving the Microsoft answer for so long that questioning it felt disloyal rather than diligent.

AI broke through all of that. Not because AI is smarter than human consultants, but because AI does not share my frame. It does not care about my partner incentives. It does not worry about co-sell revenue. It does not have a Solutions Partner designation to protect. It just does the comparative analysis without the structural constraints that had been quietly shaping my recommendations for years.

The tool I had been warned about because it “hallucinates” helped me see past my own limited frame.

From avoiding hallucinations to catching my own bullshit

This is the shift I did not expect.

When I started using AI seriously, I was focused on the hallucination problem. How do I verify what AI tells me? How do I catch its errors? How do I make sure I am not trusting something that sounds confident but is not grounded?

Those are the right questions. AI does hallucinate. Verification matters.

But somewhere along the way, I realized the more valuable question was the mirror image: how do I catch my own confident output that is not adequately grounded?

Because that is what AI turns out to be extraordinarily good at. Not replacing your judgment, but pressure-testing it. When I bring a recommendation to AI and say “challenge this,” it does not care about my ego or our relationship. It does not worry about telling me something I do not want to hear. It does not share my incentive structure. It just examines the logic and shows me where my frame has gaps.

That is not a replacement for human expertise. It is a corrective for the blind spots that expertise creates, including the ones I could not see because I was standing inside them.

I spent years worrying about AI hallucinations. Turns out the more expensive problem was my own frame going unexamined. AI did not make me more honest. It made it possible to do analysis I was structurally discouraged from performing. And the results changed how I advise clients.

Your Week 31 challenge: pressure-test the next piece of advice you receive

This week, the next time you get a recommendation from an expert, a peer, a vendor, or a friend, try this before you act on it.

Take that advice to AI and say: “Someone I trust recommended X. What are the strongest arguments against this approach? What alternatives should I be considering? What questions should I be asking that I am probably not?”

You are not asking AI to override the advice. You are asking it to do what the advisor’s frame may have prevented: examine the recommendation from outside the frame it was created in.

Pay attention to what comes back, not because AI is always right, since it is not, but because the gaps it surfaces will tell you something about the frame the original advice came from. You might discover the advice is even better than you thought. You might discover assumptions you did not know you were inheriting. Either way, you will be making decisions with a wider lens than any single advisor, including AI, can provide alone.

Then ask yourself why you did not do this before. Not with AI specifically, but with any counterpoint. How many significant decisions have you made based on advice from people inside the same frame, without ever hearing from someone outside it?

That is the habit worth building. Not skepticism toward the people you trust, but verification as a standard practice, applied equally to human advice and AI output alike.

The bottom line

I started this article with a confession: for years, I gave advice shaped more by my position inside an ecosystem than by genuine comparative analysis. Not because I was dishonest, but because the world changed and my frame did not change with it, because the system I operated in rewarded platform expertise and did not reward, actually penalized, the kind of cross-platform analysis that would have made my advice more complete.

Both AI and human advisors produce confident output that requires verification. The difference is that AI gives you tools to check it. You can compare across models, ask for sources, audit the reasoning. The advice you get from peers, vendors, and industry networks has no audit trail and no error rate, because nobody thinks to check.

The question is not whether to trust AI or trust humans. It is whether you are applying the same verification standard to both, or whether you have been scrutinizing the new tool while giving the old patterns a pass.

If the answer makes you uncomfortable, you are exactly where I was six months ago. That discomfort is where the useful work begins.

This post is part of my “AI Over 40” series. It first appeared on LinkedIn: AI for the Over 40 [Week 31]: The Paradox of Expertise: What If You’re Hallucinating Too?

Read more AI and Copilot blogs.

Trending Posts