Artificial intelligenceTechnology
What People Get Paid For When Thinking Gets Cheap
Every major technological leap has changed what is valued in work. AI is no exception.
Translated from the Spanish original. Read the original
There’s a curious way to think about artificial intelligence that has nothing to do with technology. It simply involves asking what it changes in terms of value.
The history of work is, in part, the answer to that question. Whenever a technology makes something that used to be scarce cheaper, value doesn’t disappear; it shifts. I think understanding where it shifts this time is more useful than any prediction about which jobs are going to disappear.
When physical strength stopped being scarce
For centuries, a huge part of work depended on physical strength. In farming, construction, and transportation, whoever could carry the most or endure the most was more productive and therefore better paid.
Mechanization broke that connection. One person with a machine could do the work of many, and strength stopped being what set one worker apart from another. The change is clear in American agriculture:
Source: Our World in Data, based on historical U.S. data series
See the data
| Year | Agriculture |
|---|---|
| 1840 | 64% |
| 1880 | 43% |
| 1920 | 27% |
| 1960 | 10% |
| 2000 | 1.5% |
In 1840, almost two out of every three American workers were in agriculture. By 2000, fewer than two out of every hundred were.
It’s not that people stopped producing food: much more was produced by far fewer people.
What’s interesting is what happened to those people, or to their children. Value shifted toward those who knew how to design, build, maintain, and organize machines. Knowledge became the currency of well-paid work.
Knowledge is getting cheaper
What AI is doing looks a lot like that, but with knowledge. Writing, summarizing, programming, analyzing data, and translating are tasks that until recently required training and hours of work, and that a tool can now do in seconds for very little money.
And for less and less. According to various studies, the cost of achieving the same level of performance with these models has fallen by a factor of between 9 and 900 per year, depending on the task measured. Few things in the economy are getting cheaper that fast.
| Technology / sector | Annual reduction factor |
|---|---|
| Artificial intelligence (fixed performance) | ~13× per year |
| DNA sequencing (2001–2025) | ~1.84× per year |
| Classical computing / Moore’s Law (1940–2001) | ~1.51× per year |
| Lithium-ion batteries (1991–2024) | ~1.16× per year |
| Electricity (1892–1973) | ~1.05× per year |
I don’t think human knowledge is going to lose all its value overnight. But I do think the process is underway, that it will be gradual, and that, just as with physical strength, it will change what people get paid for.
Judgment as the new advantage
My answer is judgment. I define it simply: the ability to decide what’s worth doing, what to discard, and when something is well done—the what, the how, and the when; that sixth sense the average person so often lacks, and that the vast majority of freelancers, business owners, and entrepreneurs have more than enough experience with.
The difference between judgment and intelligence or knowledge matters:
| Knowledge | Judgment | |
|---|---|---|
| What it is | Knowing how to do something | Knowing what to do and whether it’s good |
| How you acquire it | Study and practice | Experience, exposure, and your own mistakes |
| What AI does to it | Makes it cheaper and multiplies it | Needs it to know what to produce |
Anyone can see this for themselves with a mobile device that has access to any AI tool.
For example, until now I’d always avoided producing video content, since neither the technical side of cameras nor video editing appealed to me. Over time, thanks to AI, the technical barrier to knowing about cameras, images, and sound came down, but not enough to get me to take the plunge just yet. Then, recently, something launched that changed that forever.
And no, it’s nothing disruptive, and it’s not a new company; it’s simply Claude, specifically Opus, Sonnet, and Haiku 5.5. Overnight, with the help of Motion Design, various libraries, and resources (which they install and use on their own), I can drop a completely raw video recording into a folder, and they edit the important parts, add subtitles, adjust the lighting and sound, cut silences and noises, insert graphics and animations, and much more.
Overnight, I felt the urge to finally start creating content, until I ran into the wall of creativity. Yes, ironically, the wall was creativity. Then I had the wonderful idea of telling a friend who loves this world about this new feature. In 20 minutes of conversation, he flooded me with ideas, video formats, animations, things I could build on, and much more.
That was exactly when I realized for the first time, beyond my own experience, just how much leverage and power AI offers when the right person knows how to use it.
Tools have lowered the barrier to entry. The bar for quality, by contrast, keeps rising.
When anyone can produce something, what people get paid for is producing something good, quickly, and at a low cost, all at once. And that depends more on judgment and experience than on the tool itself.
The cost of adapting
It’s worth not idealizing the historical example. The fact that value shifted from physical strength to knowledge doesn’t mean those who made a living through their strength quickly found somewhere else to go. The chart shows decades of transition, and behind every data point are generations who had to retrain, move to the cities, or simply accept worse jobs.
If this time the change happens faster—and everything suggests it will—the question isn’t just where value is going, but how much time we’ll have to adapt.
In Spain, the change was also profound: between 1950 and 1975, agriculture’s share of employment fell rapidly as industry, construction, and services grew. But that aggregate shift didn’t mean everyone found another job at the same pace. For young people, moving to the cities could open a path into those sectors; for those who had spent a large part of their working lives in the countryside, learning another trade and starting over could be much harder. In many families, full adaptation came with the next generation.
Source: Historical series from INE and the Bank of Spain; exact reference pending confirmation
See the data
| Year | Agriculture | Industry and Construction | Services |
|---|---|---|---|
| 1950 | 48.8% | 25.6% | 25.6% |
| 1955 | 44.5% | 28.1% | 27.4% |
| 1960 | 39.7% | 31.8% | 28.5% |
| 1965 | 32.5% | 35.4% | 32.1% |
| 1970 | 25.1% | 37.2% | 37.7% |
| 1975 | 19.3% | 37.4% | 43.3% |
| 1980 | 16% | 35% | 49% |
Which tasks are most automatable
Not all knowledge gets cheaper at the same time. The first tasks to go are the ones with three characteristics:
- They can be described with clear instructions.
- Their results can be checked easily.
- There are lots of examples of how to do them well.
Writing an email, summarizing a document, or writing standard code meets all three. Deciding what product to launch, how to handle a difficult customer, or what risk is worth taking meets almost none.
That difference is exactly what gives judgment its value, and yes, it’s uncomfortable to understand, because suddenly we see that those difficult things that take time, effort, and thought—those things seen as boring and intimidating in today’s society of quick, easy dopamine—are going to become, overnight and by a wide margin, the tasks that receive the greatest rewards.
Exposure doesn’t mean that this percentage of jobs will disappear: it indicates how much generative AI could affect the tasks performed by each group. The pattern is clear: administrative support stands out for its automation potential, while complementarity matters more in professional and technical occupations. Estimated exposure is much lower in physical jobs, though not zero.
The important takeaway isn’t that AI is simply going to replace clerical workers, technicians, or professionals. It’s that it can automate a larger share of administrative tasks, while in many skilled occupations it tends to act as a complement: it makes execution cheaper but leaves more of the work to deciding what to do and evaluating the result.
Lessons from building with AI
At Jurenza, I’ve built a product, a brand, and a website with the help of AI tools. And unlike 1 or 2 years ago, it’s nothing to be ashamed of.
After hundreds of hours on this project, and thousands of hours on many others that saw little, or barely saw the light of day, I can say with great pride:
The experience confirmed the idea. AI dramatically sped up execution, but none of the tools made the decisions that mattered: what problem to solve and for whom, what to leave out, and when something was ready.
All at once, every failure, every mistake, every problem I’d faced across the many projects I’d put so much effort and passion into had been worth it.
Every afternoon spent in front of the screen trying to figure out why things were going wrong, wondering whether I was steering the ideas, target audiences, and production in the right direction. Absolutely everything made sense.
So, how do I gain experience?
If judgment comes from experience, and AI does many of the tasks people used to learn from, how will someone just starting out develop it?
It’s the most serious objection, and there’s no definitive answer. My own view is moderately optimistic: AI also makes it possible to do many more iterations. Someone who used to complete only a few projects a year can now run dozens of tests in a week, compare them, and learn from them. Used this way, it accelerates learning. Used to avoid thinking, it destroys it.
How to invest in judgment
If value is shifting toward judgment, it makes sense to invest in it:
- Expose yourself to the best in your field. Judgment develops through comparison.
- Do lots of real things. Every attempt with a visible result teaches you more than ten readings.
- Specialize. Deep judgment about something specific is worth more than a superficial opinion about everything.
- Use AI to execute, not to decide.
When thinking gets cheap, what matters isn’t producing more answers, but deciding which one deserves to become an action. AI can multiply the options and speed up execution; someone still has to choose and take responsibility for the result.
And if someone is going to be responsible for that, it’s you.