By Dr. Cole Webb – 11/17/26

For as long as anyone can remember knowing something that other people didn’t was enough to make you valuable. In capitalist markets, its often that the supply of expertise in a particular field is what determines the price of that product or service. But something in modern economies is changing to flip this idea on its head. What happens when everyone has access to that knowledge? Knowledge that required years of education, special training, or expensive professional consultation can now be accessed and interpreted almost instantaneously from the advent of Artificial Intelligence (AI). We are building technology that makes it easier than ever to perform work without the experience of learning it, but it may be at the very moment that that experience is most valuable.

A Saturation in the Supply of Explicit Knowledge

Humans have been good at solving problems for a long time. Sometimes it was by acquiring some specialized theoretical knowledge through education; other times it was by risky investments in time and tools to learn through the school of hard knocks, but in either way there was a detailed understanding of the theory behind the problems at the end of it. But generative AI reduces the cost of this investment… dramatically. Both through time investments and financial ones. The cost of retrieving, interpreting, and producing many knowledge-based solutions to problems is near zero to the entire population. For example, my company’s entire Learning Management System (LMS) was “vibe-coded” using Anthropic’s Claude. Now, I am a physics teacher and have a self-taught acumen to understanding computers and technology but honestly, I didn’t need very much “skill,” and this project alone would have cost me tens of thousands of dollars.

What do we do when just having knowledge stuck in our personal, often less-than-efficient, organic neural network has become less economically differentiating than a mechanical neural network? Just like horses of old, we may be witnessing a saturation in the supply of what once allowed individuals to create economic value. In a 2026 Stanford study examining the employment effects of generative AI, researchers found that employment among workers aged 22–25 in occupations highly exposed to AI had fallen 19% relative to the growth experienced by their less-exposed peers. While this does not establish AI as the sole cause, it suggests that the economic consequences of this technological shift may already be appearing in the labor market.

So how do we avoid the same fate?

The Economic Premium on Competence.

In 1966, philosopher Michael Polanyi published The Tacit Dimension, in which he explored an important distinction in how humans acquire and possess knowledge. Polanyi argued that much of what we know cannot be easily explained, written down, or transferred to another person. He called this tacit knowledge, famously observing that “we can know more than we can tell.” Consider something as simple as riding a bicycle. You could read every book ever written on the physics of balance, understand the mechanics of a bicycle, and memorize every instruction on how to ride one, but none of that necessarily means you could get on a bicycle and ride it. That ability comes through experience, through practice, failure, and adjustment. It is the difference between possessing explicit knowledge of how something works and developing the tacit knowledge necessary to actually do it freely from scratch.

A motivated novice may be able to complete a task he has zero understanding of by using the available AI resources that replace his own internalized explicit knowledge; however, the experienced practitioner would be able to complete the same task, likely with more speed, quality, independence, and judgment. Economist David Autor explored this distinction in his 2015 paper, Why Are There Still So Many Jobs?, discussing what he described as the Polanyi’s paradox. The paradox describes a longstanding challenge in automation: many of the things humans know how to do are difficult to translate into explicit instructions for a machine. While AI is beginning to overcome some of these limitations, access to knowledge and the ability to apply it competently are still not necessarily the same thing. And its entirly still within the realm of possibility that AI may never develop a tacit knowledge of how things are done.

This distinction is important because the economic value of competence is not simply in what an experienced worker knows, but in the efficiency with which that worker can apply it. An AI-assisted novice may be capable of producing a similar outcome as an experienced practitioner, but differences in speed, quality, and judgment can still create an economic advantage. It is in this efficiency differential that an incentive for employers to develop competent workers appears.

AI and the Changing Economics of Workforce Development

In a 2025 study published in the Quarterly Journal of Economics, researchers Brynjolfsson, Li, and Raymond examined the introduction of an AI assistant among 5,172 customer-support workers. They found that AI increased productivity by an average of 15%, with the greatest benefits going to less experienced and lower-skilled employees, who resolved approximately 30% more customer issues per hour. Interestingly, workers with only two months of experience who used AI performed comparably to workers with more than six months of experience who did not. AI was effectively allowing inexperienced employees to perform work at a level that previously required substantially more experience.

This finding could establish itself as a primary factor that incentivizes employers to invest in the development of new workers. It is no longer the same tedious and time-consuming process of building the minimum required theoretical knowledge that precipitates practice. Well-designed AI-supported training could shorten the transition from novice to independently competent worker.

However, there is an important caveat between making an inexperienced worker more productive and actually making that worker more competent. In a 2025 study published in the Proceedings of the National Academy of Sciences, researchers found that students using unrestricted AI assistance performed substantially better on mathematics practice problems, yet performed 17% worse on a subsequent exam when the AI was removed. Interestingly, students using an AI tutor designed to guide their learning rather than simply provide answers did not experience the same decline. The implication of this study is that it matters where AI-supported training introduces friction into the system. Unproductive friction consists of obstacles that consume time without necessarily contributing much to competence: searching for information or struggling to identify clear objectives. However, productive friction allows for experimentation, includes opportunity for individual judgment and troubleshooting, simply allowing for the process of making and correcting mistakes.

Experience becomes much cheaper when the learner has a baseline of competency. When workers can contribute productive labor earlier, the employer may recover more of the investment while the worker is still learning. The opportunity arises when AI doesn’t eliminate the learning process but rather reduces the cost of getting workers to the point where meaningful value and the development of tacit knowledge become possible.

Ushering in an Age of Doing Things Poorly

I received my first teaching position as a middle school intern in 2017, and it went horribly. I was in way over my head. Classroom management was terrible, writing lessons was exhausting, and I definitely said some things that, to be generous, were counterproductive. Then I joined a credential program and had an experience that, in my conversations with other educators, appears to be nearly universal.

I learned almost nothing about how to do my actual job.

Don’t get me wrong, I am not saying that learning the theory was useless. My study of B.F. Skinner’s operant conditioning and Albert Bandura’s theory of self-efficacy fundamentally shaped my understanding of learning and curriculum design. But knowing those theories did not make me capable of leading a classroom.

I became capable by leading one.

I learned by trying lessons, watching them fail, adjusting them, and trying again. I learned by observing what drew students into an experience and what made them reach for their phones. The truth is that competence develops through imperfect attempts, feedback, repetition, and progressively greater responsibility. But accepting temporary imperfect performance is not the same as accepting permanently poor expectations.

A cultural shift is needed: experience needs to be the new education. But in order for that to happen, employers, institutions, and individuals must make room for productive friction in the workplace. We cannot continue down the pathway of an entry-level job requiring 10 years of related experience. Experience needs to be as freely offered as high-interest student loans have been the last two decades. Employers can view novice inefficiency as an investment in the tacit knowledge of future supervisors, not a current expense.

Investing in our capacity to develop competence

The death of the expert does not have to mean the death of expertise. If anything, it presents an opportunity to reconsider how we develop it, like learning to ride a bike. AI has the potential to reduce the time and cost required to prepare inexperienced workers for meaningful work, but it cannot replace the value of experience itself. Competence still requires practice, mistakes, feedback, and the opportunity to do things poorly before doing them well. The faster we can develop that competence, the more economically rational it becomes to give inexperienced people the opportunity to develop it. Perhaps the future of work does not require fewer beginners, but rather a willingness to let more of them improve.

Sources

Brynjolfsson, E., Chandar, B., & Chen, R. (2026). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence [Working paper]. Stanford Digital Economy Lab.

Autor, D. H. (2015). Why are there still so many jobs? The history and future of workplace automation. Journal of Economic Perspectives, 29(3), 3–30. https://doi.org/10.1257/jep.29.3.3

Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. https://doi.org/10.1093/qje/qjae044

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), Article e2422633122.

Polanyi, M. (1966). The tacit dimension. Doubleday.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *