When Your Expertise Stops Being Scarce

By Michael Ulrich

Michael is the founder of Thornwright, focused on engineering leadership, technical recovery, and operational execution in complex product environments.

A recent The Wall Street Journal piece (The Unexpected Risk of Letting ChatGPT Fact-Check You: Financial Adviser) examined an odd effect: clients using ChatGPT to fact-check their financial advisors sometimes get worse service. These advisors feel challenged, disengage, and thus their relationship degrades.

It’s easy to dismiss it as a behavioral flaw with fragile egos and poor bedside manner. But it’s a structural signal. While etiquette may seem broken, it’s the model that made certain expertise valuable. This issue extend beyond just financial advisors but other areas where specialized expertise is required

The quiet assumption most expert businesses rely on

Across various fields such as finance, medicine, law, consulting, and engineering services, four fundamental assumptions have held significant weight:

1. Information is scarce.
2. Interpretation is the exclusive domain of credentialed experts.
3. The process is opaque enough that clients won’t verify it.
4. Trust is derived from the first three assumptions.

However, Artificial Intelligence (AI) has the potential to disrupt the first and third assumptions at scale. This disruption destabilizes the second assumption, and subsequently, the fourth assumption doesn’t survive the transition unscathed. If your offering is built upon these assumptions, you’re not merely dealing with a tooling change. You’re confronting a shift in how value is assigned.

What actually changed

Clients didn’t become experts overnight. The gained immediate access to plausible answers with the simple entering of a prompt into any of a number of AI interfaces. This produces as cheap way for cross-checking reasoning and gives a different starting point for conversations. That’s enough to reframe the interaction from “Tell me what to do” to “Explain why your recommendation is better than what I already have.” That becomes a very different job for the experts.

Where the model breaks first

In my work, I’ve noticed the fracture lines in five places.

1. Information is no longer a product. If your deliverable is a report, a plan, or an explanation, a client can now produce a competing version in minutes using an AI platform. This output might not always be correct and is certainly not equivalent, but it is close enough to force a comparison. Once that comparison exists, your output is no longer self-justifying. It had to win. The real tension occurs when close enough is not good enough in high-stakes domains. But most work is not high-stakes, and it is this middle market where issues develop.
2. Authority becomes testable. Historically, authority was buffered by credentials and process opacity. Now it’s continuously audited quietly and often without the expert’s knowledge. Professionals don’t always appreciate being fact-checked. Their performance and engagement with their clients drops. It’s not always a personal issue but rather a result of a system dependent on unchallenged authority. As clients misinterpret AI output, noise is created. This means the experts’ role now includes filtering bad comparisons and not dismissing the existence of comparisons.
3. The moat was built in the wrong place. A lot of capital goes into proprietary frameworks, repeatable processes, and production efficiency. Those were defensible when information was scarce. They’re not defensible when the generation is cheap. But still, some things are hard to replicate, like judgement under uncertainty, cross-domain synthesis, and accountability for outcomes. If a business can’t center on those, the moat is thinner than it looks.
4. Role compress, but organizations don’t. AI doesn’t eliminate roles cleanly. It compresses them with junior work of first drafts and basic analysis partially replaced, and senior work shifting towards evaluation, synthesis, and decision framing. Most organizations haven’t updated the role definitions or incentives. They still hire, train, and promote for production. The result: more output with less understanding.
5. The relationship flips to audit mode. Clients increasingly behave like auditors. They have options to compare, make probing assumptions, and ask about tradeoffs. Continuing to operate as an unquestioned authority degrades the interaction. Understanding the shift happens allows you to use it.

A simple diagnostic founders can run

To understand where you stand, you don’t need a full transformation program from an expensive consulting firm. 

1. What are you really selling? If a client can replicate a meaningful portion of your output with AI, you’re selling something that’s become cheap. The real test is to ask if you removed your deliverable, would a client still pay for your involvement?
2. How does trust hold up under scrutiny? If you are being challenged and it degrades your performance, your authority is conditional. Ask if every client you have verifies you in real time. Do you still win?
3. What can be replicated? Decompose your offering. Consider that information is replaceable, processes are partially replicable, judgement is harder to replace, and accountability is certainly the hardest for AI to duplicate. If your tools and templates disappeared, what remains that a client cannot source elsewhere?
4. What is your team actually doing? If your people are primarily producing outputs, you’re optimizing the wrong layer. If AI were to hand all the first drafts or preliminary work, what then is each role for?
5. Is your client relationship that of an auditor or an advisor? If clients are already comparing and challenging you, you’re in an audit relationship whether you acknowledge it or not. Are you being asked for answers or for justification?

What to change (without pretending it’s easy)

I see three practical shifts a specialized advisor could make.

1. Make reasoning legible. Opaque processes and tools used to protect you. And now this secretive black box approach invites replacement. Show assumptions, expose the real tradeoffs, and compare the alternatives, including those obviously AI-generated. This is less about transparency than it is about value. It’s making your work competitive.
2. Re-anchor value in decisions, not deliverables. If the output is the product, you’re in a race to zero. Tie your work to decisions made and risks managed. Stay engaged through the consequence, not just the recommendation. That is where accountability becomes real and billable.
3. Redefine roles around judgement. Reduce the time spent on first drafts and routine analysis and increase the time spent on framing problems, integrating constraints, and deciding under uncertainty. This is a structural change, not a tooling upgrade.

Where this argument overreaches

Accuracy still matters. While AI’s “good enough” output is suitable for most normal uses, it fails in edge cases where competence and experience matter. Medical, legal, and safety-critical engineering will always need expertise and judgement, which is unlikely to disappear anytime soon.

Relationships still carry weight. Trust doesn’t vanish just because information has become cheap. It has to survive verification. Many clients will still prefer to engage with a human they can trust instead of a software system, even if they check everything they are told. 

So I pose an uncomfortable question: If a client can generate a plausible version of your work, challenge your reasoning, and compare alternatives instantly, what exactly are you being paid for? If the answer is “this document,” you have a problem. If, however, the answer is “we help you decide and we stand behind it,” then you have something to build on.