TL;DR: The strongest doubter argument is that AI is the next NFT, crypto, web3, or metaverse, all of which were hyped and faded. The pattern recognition is real and the doubter isn’t stupid for noticing it. The argument fails for one reason. Those technologies failed to deliver on the underlying capability claim. AI has delivered, and the delivery is visible in working products that millions of people use daily. Here’s how I use AI on a book.
Here’s the actual distinction between fad technologies and durable ones, where AI sits on that distinction, and why the pattern that retired the last few hype cycles won’t retire this one.
What is the strongest argument that AI is a fad?
The case against AI as a fad isn’t weak. The doubter remembers the promises. NFTs were the future of ownership. Crypto would replace banking. Web3 would replace the internet. The metaverse was where work and play would move. Each of those claims came with breathless coverage, massive funding rounds, professional credentialing, and confident predictions that holdouts would regret their skepticism. In each case the holdouts were right. The technologies either collapsed entirely or settled into niche uses far smaller than the predictions promised.
The pattern is real, and the doubter who sees AI as the same pattern is doing reasonable pattern matching.
The case against AI as a fad isn’t weak. It’s just incomplete.Share on X
The pattern matching has a specific shape that the doubter should name explicitly. New technology appears. Excitement builds. Money pours in. Companies promise transformations. Most companies fail. The remaining ones produce something narrower than the original promise. Five years later, the narrower thing exists but isn’t the world-changer it was sold as, and the doubters who waited get to feel justified. That shape repeats often enough that it’s reasonable to assume it’ll repeat again, and the doubter who pattern-matches isn’t being lazy.
Why does the fad pattern fail on AI?
The pattern fits cycles where the underlying capability never delivered. NFTs promised verified ownership of digital goods, which the technology did provide, but the goods weren’t actually valuable, so the verification didn’t solve a real problem. Crypto promised an alternative to banking, but the actual products were worse banking with more fraud and slower transactions, and the alternative-to-banking use case turned out to matter less than the marketing claimed.
Web3 promised decentralized internet, but the decentralization came at costs people didn’t want to pay, and the architectures that delivered on it were used mostly by speculators. The metaverse promised immersive virtual spaces, but the spaces were empty and the headsets were uncomfortable, and the demand never materialized.
In every case, the failure was that the working products didn’t deliver value users wanted. The technology functioned. The market didn’t show up. Doubters who said “this will fade” were right because nobody was getting genuine utility from the working products.
AI’s working products do deliver genuine utility, in ways that show up in working data instead of in marketing decks. Hundreds of millions of people use AI chat tools weekly for real work tasks. Code assistants are being used by working developers because they produce real productivity gains. Image generators are being used by working designers as part of actual production pipelines. The use isn’t concentrated among speculators waiting for prices to rise.
The use is concentrated among working professionals who have integrated the tools into how they earn their income, the strongest possible signal that the underlying capability delivers value.
How do you test fad versus durable technology?
Watch whether people keep using it after the novelty wears off, for ordinary work, without being told to.
From the podcast
Martin Ricketts has been reading technology trends for three decades and warns about the failure mode on the enthusiast side, on Leaders and Their Stories:
There’s tons of AI tools out there and people get shiny object syndrome around this stuff.Chasing every launch and dismissing the whole category are both ways of avoiding the work of getting good at one thing.
The cleanest test for whether a technology is a fad is whether it produces value people will pay for, repeatedly, with their own money, outside the hype cycle. NFT speculation produced trading volume but no recurring use case anyone would pay for if speculation stopped. Crypto produced trading volume but the same problem. Web3 produced a small core of true believers and almost nobody else. The metaverse produced billions in spending but no visible audience.
AI passes this test cleanly. Professionals pay monthly for AI tools the way they pay for other software, with their own money, for tasks they’ve decided are worth the price. Companies pay enterprise rates because the productivity gains are visible in their numbers. The payment is recurring, the use is sustained, and the payers aren’t waiting for someone to buy from them later at a higher price.
That signal is qualitatively different from the speculation that drove the previous hype cycles, and the difference is the reason the fad framing breaks down. A piece on what AI is actually good at covers the specific tasks where this value is showing up.
The “but the products are bad” objection
The doubter’s response is often that the working products are bad, hallucinate, produce flat output, and disappoint users repeatedly. The objection is partly true. Products do hallucinate, and the output is often flat, and the disappointment is real. None of that puts AI in the fad category. The products being imperfect doesn’t mean the capability isn’t there. The capability is the question, and the capability is what determines whether the technology endures or fades.
NFTs were perfectly functional. The blockchain did exactly what it was supposed to do. The technology worked. What didn’t work was demand for the use case. AI products are imperfect but the demand for the underlying capability is enormous, and that demand is producing rapid product improvement on a quarterly basis. The bad products of today are training data for the better products of next year, and the working professionals using the tools are providing feedback that improves them.
The trajectory is the opposite of fad technologies, where the underlying use case never materialized and improvement couldn’t save the products.
The “this is different than X past technology” trap
Every technology cycle includes someone arguing “this one is different,” and the argument is sometimes used to defend things that turn out to be fads. The doubter who notices this is doing legitimate pattern matching. But the trap is that the argument is also used to describe technologies that were different and didn’t fade. The internet was different from radio in the late nineties, even though both were communication technologies in their hype cycles.
Mobile was different from PCs in the late aughts, even though both were computing platforms. Email was different from fax in the eighties, even though both were business communication tools.
The way to tell which one applies in any specific case is to look at the working data instead of the rhetoric. Real working data on AI shows sustained paid use by professionals integrating the tools into income-producing work. Real working data on NFTs showed speculative trading by investors waiting for liquidity. Those data patterns are different categories of phenomenon. Both came with rhetoric. Only one came with the working-professional adoption that signals durable technology, and that one is the one to take seriously.
A piece on the realistic AI-and-jobs picture covers more of how the adoption is actually playing out in working environments.
What this means if you are still a doubter
The honest position for a doubter at this point is to acknowledge that the fad framing doesn’t fit the data and to engage on a different question. A doubter can reasonably ask whether AI is overhyped, which it is. They can ask whether AI is good for everything, which it’s not. They can ask whether AI is being adopted badly by many companies, which it is, with predictably bad results. All of those questions are legitimate and produce useful answers.
The fad question, though, has been answered by the data, and continuing to ask it after the answer is in produces the experiential gap covered in the anchor piece. Use the working professional energy for the real questions about how to adopt the technology well, where it helps, where it doesn’t, and what to do about the cases where it’s being adopted badly. Those questions are where the value of the doubter mindset compounds, instead of dissipating against a fad framing that the technology has already escaped.
