Yes! Also, quality matters, cost matters, the low-hanging fruit of white-collar automation has already been plucked, and donโt forget all the ways that generative AI can be used to destroy economic value.
Thanks for this, Matt. I wonder as well, what the longer-term cost is of AI-induced cognitive decline, doesn't it also mean -- as LLMs and machine-learning apparati become more trained and sophisticated, so does the cognitive offloading mean that we as a whole get less equipped in our ability to think critically and analyze? How do we measure the cost of that, at the individual level but then also at the national level (i.e., India's experience of "brain drain" when its engineers get regularly recruited out of country)?
Great point, S.N. Yes, those are additional ways in which generative AI can (indirectly) destroy economic value in the long-run. It would be nearly impossible to measure. Heck, even the impact of having to divert more resources to cybersecurity and other forms of AI-necessitated quality control and anti-fraud measures is difficult to quantify. But in the long-run, there is a real risk that AI will destroy economic value simply by dulling people's ability to output and recognize high-quality (and economically valuable) work.
This is such a useful distinction. A lot of productivity advice treats โmore outputโ as the win, but the real leverage is often lowering the friction around what already matters. If the system adds another layer to maintain, it can quietly become part of the problem.
Thanks for this essay Matt, lots of interesting and relevant insights around how AI is currently delivering economic value. I wonder if you know how folks at these frontier AI labs think about the law of diminishing returns and how/if it applies to a general purpose tech like AI?
The narrative of โtoday is the worst AI will ever beโ is so pervasive โ especially in the Bay Area where Iโm based โ and Iโm wondering how strong the evidence for doubt is, or if we may be downplaying the potential value of this tech?
Hey, Danielle! Great questions. I wrote a long reply that was, in fact, long enough that I think I'll expand on it and turn it into a column :)
Short version:
(1) I don't think generative AI is a general-purpose technology (I allude to this in my piece where I question whether it can perform tasks as valuable as those that drove the Industrial Revolution).
(2) The tech bros seem to think that AI is different in terms of diminishing returns, but that's because they misunderstand the significance of Moore's Law (yes, we can achieve exponential growth by specific technical metrics, but those technical advances do not unlock exponential increases in economic value).
(3) I don't think we're downplaying the tech's importance. I think the Valley is overplaying it. Of course I could be wrong--I constantly question whether I am gaslighting myself about this. But I have been following the tech's progress (and lack thereof) for the past 11 years and am fairly confident that it's the industry that is gaslighting itself (and us) about its economic value.
A recent NYT story reports that โtwo business school professors, curious about A.I. books and whether anyone actually likes them, gathered data about 10 million books published on Amazon over the last five years. They found that the number of e-books published per month had tripled since the release of ChatGPT, to more than 300,000 at the end of last year, from around 100,000 in 2022.โ
So AI lets people publish an avalanche of low-quality slop writing to Amazon. Is this increasing Amazonโs productivity, or meaningfully lifting their bottom line? Doubtful.
I have been saying I have seen benefits at the micro local level, but nothing at the macro level. Office is still bloated, Windows is awful, most apps are still consuming massive amounts of ram but my co-worker can "one-shot" an excel based workflow app to web app in "a few hours" for 2 internal users in our company over and over again. "Coding is a solved problem" and " I can train someone to program (in our language of choice) in a few weeks with Claude, they don't need to know it" is what I heard this very week. I can quickly make a web app to the sole benefit of my family but for no one else.
This was an interesting read and dovetails with something that has been on my mind regarding the future of AI. How does AI, which devalues the work it does in that it removes scarcity, create value? If AI can write a million pop songs an hour, then pop music has no value. If AI can design a thousand bridges a day and is easily available to anyone who wants it, then civil engineering becomes a lot less profitable. And what really baffles me is what happens when an AI system is used to generate new AI systems? How does anyone make any money off of any of this? Of course AI companies, who don't give a shit about the Intellectual property of anyone they currently steal from (training material) will suddenly become very concerned about their own intellectual property when it's being stolen (used to train) by their competitors.
Glad I didnt start writing my comment, before reading the others, as this is exactly what Ive been thinking about as well.
In addition, it has the ability to erase economic value, while at the same time dramatically increasing social value, breakthroughs in medicine, or our understandings there of how to keep people healthy (giving AI the benefit of the doubt here with future breakthroughs) could dramatically cut the " economic value" that gets "created" by illness while simultaneously providing so much value to human life..
Whats somewhat ironic about this piece, is in my opinion the real benefits of AI, making lives better, should decrease economic value overall. Our system is fixated on increasing value, finding edges, ways to use AI to create economic value...while the long term goals should be oriented, not towards how wall street implenents AI to create more accurate predictions and forecasting. Perhaps AI itself is inherently an Anti-Value.
But I suppose thats whats also being said here, that the people with the means are siezing the opportunity in the mean time to make as much money as possible. I think Elons point of money not really mattering at some point, has a greater than 50% chance of being true, because if society inevitably reaches the conclusion that an "abundance" of ideas and capability does deflate our economic values to virtual 0, social value will be all thats left to really matter.
The end result after many decades will be the automation of automation. The integration of physics simulations with AI with process control to build facilities that say build the machines that bore tunnels so the tunnels are continuously bored to expand the network of transport and service corridors where the supply chain of parts and materials is itself automated
Lowering the bar for creation, did not in fact, give people the requisite taste or experience for creating anything worthwhile. The quantity of things increased, the quantity of those capable of quality work did not.
Tl;dr much white collar work is about extracting rents as a middleman and doing it faster doesn't increase the total extractable rent.
Yes! Also, quality matters, cost matters, the low-hanging fruit of white-collar automation has already been plucked, and donโt forget all the ways that generative AI can be used to destroy economic value.
๐ฏ
Thanks for making this accessible.
I'm not a trained economist so I have to dumb things down to my own level.
I meant that thereโs no paywall haha.
Oh, well, in that case, youโre welcome ๐คฃ
Thanks for this, Matt. I wonder as well, what the longer-term cost is of AI-induced cognitive decline, doesn't it also mean -- as LLMs and machine-learning apparati become more trained and sophisticated, so does the cognitive offloading mean that we as a whole get less equipped in our ability to think critically and analyze? How do we measure the cost of that, at the individual level but then also at the national level (i.e., India's experience of "brain drain" when its engineers get regularly recruited out of country)?
Great point, S.N. Yes, those are additional ways in which generative AI can (indirectly) destroy economic value in the long-run. It would be nearly impossible to measure. Heck, even the impact of having to divert more resources to cybersecurity and other forms of AI-necessitated quality control and anti-fraud measures is difficult to quantify. But in the long-run, there is a real risk that AI will destroy economic value simply by dulling people's ability to output and recognize high-quality (and economically valuable) work.
This is such a useful distinction. A lot of productivity advice treats โmore outputโ as the win, but the real leverage is often lowering the friction around what already matters. If the system adds another layer to maintain, it can quietly become part of the problem.
The Mickey Fantasia analogy is so good. Not only does it require some serious cleanup, it can get away from you real fast.
Thanks for this essay Matt, lots of interesting and relevant insights around how AI is currently delivering economic value. I wonder if you know how folks at these frontier AI labs think about the law of diminishing returns and how/if it applies to a general purpose tech like AI?
The narrative of โtoday is the worst AI will ever beโ is so pervasive โ especially in the Bay Area where Iโm based โ and Iโm wondering how strong the evidence for doubt is, or if we may be downplaying the potential value of this tech?
Hey, Danielle! Great questions. I wrote a long reply that was, in fact, long enough that I think I'll expand on it and turn it into a column :)
Short version:
(1) I don't think generative AI is a general-purpose technology (I allude to this in my piece where I question whether it can perform tasks as valuable as those that drove the Industrial Revolution).
(2) The tech bros seem to think that AI is different in terms of diminishing returns, but that's because they misunderstand the significance of Moore's Law (yes, we can achieve exponential growth by specific technical metrics, but those technical advances do not unlock exponential increases in economic value).
(3) I don't think we're downplaying the tech's importance. I think the Valley is overplaying it. Of course I could be wrong--I constantly question whether I am gaslighting myself about this. But I have been following the tech's progress (and lack thereof) for the past 11 years and am fairly confident that it's the industry that is gaslighting itself (and us) about its economic value.
A recent NYT story reports that โtwo business school professors, curious about A.I. books and whether anyone actually likes them, gathered data about 10 million books published on Amazon over the last five years. They found that the number of e-books published per month had tripled since the release of ChatGPT, to more than 300,000 at the end of last year, from around 100,000 in 2022.โ
So AI lets people publish an avalanche of low-quality slop writing to Amazon. Is this increasing Amazonโs productivity, or meaningfully lifting their bottom line? Doubtful.
https://www.nytimes.com/2026/07/16/technology/ai-slop-books-biography-amazon.html
I have been saying I have seen benefits at the micro local level, but nothing at the macro level. Office is still bloated, Windows is awful, most apps are still consuming massive amounts of ram but my co-worker can "one-shot" an excel based workflow app to web app in "a few hours" for 2 internal users in our company over and over again. "Coding is a solved problem" and " I can train someone to program (in our language of choice) in a few weeks with Claude, they don't need to know it" is what I heard this very week. I can quickly make a web app to the sole benefit of my family but for no one else.
"itโs just as likely that todayโs generative AI is about as good as it will ever get"
These "deep learning will hit a wall" predictions have consistently proven false over the past 10 years or so.
Well, the only reason they have not hit the wall is the exponential increase in resource usage.
This was an interesting read and dovetails with something that has been on my mind regarding the future of AI. How does AI, which devalues the work it does in that it removes scarcity, create value? If AI can write a million pop songs an hour, then pop music has no value. If AI can design a thousand bridges a day and is easily available to anyone who wants it, then civil engineering becomes a lot less profitable. And what really baffles me is what happens when an AI system is used to generate new AI systems? How does anyone make any money off of any of this? Of course AI companies, who don't give a shit about the Intellectual property of anyone they currently steal from (training material) will suddenly become very concerned about their own intellectual property when it's being stolen (used to train) by their competitors.
Glad I didnt start writing my comment, before reading the others, as this is exactly what Ive been thinking about as well.
In addition, it has the ability to erase economic value, while at the same time dramatically increasing social value, breakthroughs in medicine, or our understandings there of how to keep people healthy (giving AI the benefit of the doubt here with future breakthroughs) could dramatically cut the " economic value" that gets "created" by illness while simultaneously providing so much value to human life..
Whats somewhat ironic about this piece, is in my opinion the real benefits of AI, making lives better, should decrease economic value overall. Our system is fixated on increasing value, finding edges, ways to use AI to create economic value...while the long term goals should be oriented, not towards how wall street implenents AI to create more accurate predictions and forecasting. Perhaps AI itself is inherently an Anti-Value.
But I suppose thats whats also being said here, that the people with the means are siezing the opportunity in the mean time to make as much money as possible. I think Elons point of money not really mattering at some point, has a greater than 50% chance of being true, because if society inevitably reaches the conclusion that an "abundance" of ideas and capability does deflate our economic values to virtual 0, social value will be all thats left to really matter.
This names something I see in real work.
Faster drafts can feel like progress and still leave you with more to fix, more to check, and nothing that actually moved.
Speed isnโt the same as value. A tool that multiplies output can also multiply noise.
The question I keep coming back to: did this change anything that matters, or did it just fill the hour?
The end result after many decades will be the automation of automation. The integration of physics simulations with AI with process control to build facilities that say build the machines that bore tunnels so the tunnels are continuously bored to expand the network of transport and service corridors where the supply chain of parts and materials is itself automated
Listening to the AI podcast host read this article
Lowering the bar for creation, did not in fact, give people the requisite taste or experience for creating anything worthwhile. The quantity of things increased, the quantity of those capable of quality work did not.
The rich get richer.
Empty metrics can generate endless corporate growthfroth.
Here is Harry Ferrell
https://substack.com/home/post/p-210192056