Inc.
AI can make almost anyone look capable, but producing something polished isn’t the same as producing something good.
AI and the Death of Expertise: Why Experts Still Matter
Author: Louise K. Allen
If the vast, always-on, and easily accessible knowledge base of the modern internet ushered in the Death of Expertise, as Thomas M. Nichols lamented nearly a decade ago, then AI put the final nail in the coffin.
The technology, the line goes, makes it possible for anyone to do nearly anything, quickly and with little traditional training. In the age of the generalist, everyone’s opinion — whether they asked ChatGPT or have a PhD — is equally informed.
On some level, it is true. Formal training is less important than in years past. AI opens doors for individuals to explore new fields, dive into passion projects, and expand their skills. Then, it removes “knowing the details” from the list of barriers to execution.
Suddenly, salespeople can write code. Marketers can try their hands at analytics. Accountants can dabble in content creation on the side. The AI model fills in the gaps that once stood between interest and implementation.
It’s a whole new world. But there’s a distinction that’s missed in the “death of expertise” narrative that’s become so common. Walking away with a finished product isn’t the same as walking away with something valuable.
Knowledge, distance, and diminishing returns
With a good generative AI tool, anyone can produce nearly anything and have it look polished, finished, and functional. However, looks can be deceiving, especially to those who don’t know the right questions to ask or the right time to ask them.
A recent Stanford-Harvard study asked a group of workers from a single company to conceptualize and write two articles for the organization’s website, with varied access to AI tools throughout. These individuals were divided into three groups:
- Insiders: Web analysts whose regular responsibilities include website content creation
- Adjacent outsiders: Marketing specialists who work alongside analysts but with different responsibilities
- Distant outsiders: Technology specialists, including data engineers, software developers, and other similar roles
Researchers then scored performance on a five-point scale. What they found neatly illustrates the fallacy in the death-of-expertise story. Though it proved there are benefits to using AI, it also showed that there are limits to its potential to enhance workers’ skills.
Each of the three groups outperformed its own non-AI scores when using the AI tool, and access to the technology closed the conceptualization gap between all three groups. What it did not do was erase the intra-group execution gap. Though adjacent outsider execution was on par with that of the insiders, the distant group lagged far behind. Perhaps most surprisingly, distant outsiders only gained 0.04 points when using AI.
Insiders with AI gained 0.67, and adjacent outsiders added an impressive 1.11.
The study, though relatively limited in scope, identifies existing domain knowledge as the determining factor in their results. The tech specialists saw less improvement because, even aided by the tailored AI tool, did not have the context to make informed decisions about the output.
The conclusion they reach is that, yes, AI can help anyone brainstorm. It can help them plan. To help with the rest, the user needs to bring their own expertise to the table.
The gap AI can’t fill
The conclusion was that you can’t make something good if you don’t know what good looks like. AI tools operate in a black box, doing what they think is correct and providing only the finished product. The user must decide whether it’s right.
For someone with experience, that’s a true qualitative exercise, akin to reviewing a peer’s work. For dabblers, it’s different.
They either defer to the technology outright or overcorrect toward their own specialties. This is why perceptions of model quality diverge so profoundly along experiential lines.
While vibe coders tout their flawless, quick-made tools, developers’ trust in AI code is plummeting. So is favorability, with positive sentiments among this group falling from 72 percent to 60 percent between 2024 and 2025.
Another finding adds some color: 80 percent of respondents now use AI in their workflows. Nearly half of developers cited “almost right but not quite” outputs as a top blocker, and two-thirds said fixing this “almost right” code is increasingly time-consuming. Given time to work with AI tools, true experts end up frustrated rather than empowered.
This is why AI does not represent the dramatic workforce realignment that many vendors have touted to prospects nor the death of expertise. As credible-looking outputs become easier to produce, someone with experience will need to separate what’s possible from what’s useful.
Hiring a gaggle of generalists and arming them doesn’t eliminate the need for experts. It just moves demand for expertise elsewhere. And it begs the question: Why not keep it where it already is?
Credits: TCA, LLC.