No Country for Beginners.
I Thought Learning a Trade Would Mean a Lasting Career
Learning technical skills carries a certain faith in the future: if you can do something others cannot easily do, there will still be a place for you, even as organizations change and the economy weakens. Anyone who has built a career in technology has probably held that belief at some point.
AI makes that belief harder to live with. Watching it perform, from a brief instruction, a task that took years of study to master can leave you unsure how to value what you already know, before you even begin to consider what to learn next. You cannot afford to stop learning, but having learned is no longer enough to feel secure.
Companies make decisions by criteria that differ from those by which individuals measure their achievements. A task can remain important while fewer people are needed to perform it. The growing importance of security does not mean hiring will increase across every security role. Existing staff may take on a broader remit, or external services may assume part of the work.
Thinking about employment in the age of AI therefore requires asking two questions together. Is the range of work I can do expanding? And how much additional demand is there for people to do that work? The answers do not always point in the same direction.
The Same AI Comes at a Different Price for Newcomers and Veterans
For experienced professionals, AI creates room to tackle work they have put off. It lowers the cost of writing analysis code, organizing unfamiliar data formats, and testing competing hypotheses. For someone who knows what to investigate and can recognize errors in the results, this is a substantial opportunity.
For someone seeking a first job, the same change can look very different. The tasks they expected to learn on the job may be automated before they ever get the chance to perform them. An organization weighing whether to hire and train a junior employee or give its existing staff better tools may decide to postpone hiring.
An August 2026 report from the Bank of Korea examined declining youth employment in industries with high exposure to AI. The researchers did not, however, attribute the trend solely to AI’s direct effects. They suggested that post-pandemic hiring adjustments, a preference for experienced hires, and a weakening of in-house training may also have contributed. There are still too many factors to disentangle to explain the difficulty of finding a job entirely in terms of AI.1
Even so, the underlying concern is clear. Employers may demand more experience while offering fewer places to acquire it. Those already working can develop alongside AI; those who have yet to start remain at the door, being asked to prove their practical experience.
Nor will the advantage enjoyed by experienced workers necessarily last. Once people can accomplish more in the same amount of time, organizations may adopt that productivity as the new baseline. What begins as breathing room can soon become the standard workload. Whether better performance with AI translates into job security or higher pay depends on the choices organizations make.
Who Is a Game Without New Players For?
Online games offer another way to think about this problem.
Imagine a long-running game in which only seasoned players remain. Strategies are refined, equipment is advanced, and established players make short work of difficult content. For a newcomer, however, everything is taken for granted. They cannot join a group because they lack experience, and they cannot gain experience because no group will take them.
The average skill level of the remaining players may look impressive. But that may simply reflect the disappearance of beginners. Looking only at the average can create the illusion that the game is healthier than before. When existing players begin to leave, there are too few people ready to take their places.
A labor market is not a game. Still, the analogy holds in one important respect: experienced people have to come from somewhere. Today’s skilled professionals exist because, at some earlier point, an organization gave an inexperienced person time and work to do. Hiring experienced staff can solve one company’s shortage. It does not follow that every company making the same choice will increase the industry’s supply of skilled workers.
An organization without junior staff may look more efficient for a while. Less time is spent explaining and reviewing, and people accustomed to working together communicate quickly. But an assessment of that performance should also ask who will take over the work a few years from now. Savings on training today may return as higher recruitment costs and unfilled roles tomorrow.
What Remains of the Skills We Have Learned?
I do not believe that past study and experience have become meaningless. If anything, this is a time to examine more carefully what our experience has actually given us.
In Applying AI to Rebuild Middle Class Jobs, David Autor explores the possibility that AI could enable more workers to exercise expert judgment. People with relevant knowledge could use AI to undertake work that was previously beyond their reach. It is a prospect for extending expertise, not a guarantee that this will happen on its own.2
The distinction becomes clearer in digital forensics. Knowing how to extract artifacts with a tool is a different kind of competence from understanding the conditions that produced them and the limits of what they can establish. AI may assemble a timeline quickly, but someone still has to interpret gaps in the record and decide what further investigation is needed when evidence conflicts.
Technical knowledge endures when it informs those judgments. The operation of a particular tool may change, but an understanding of file systems and operating systems remains useful when assessing new results. Possessing that knowledge alone is not enough, however. It must allow us to solve a broader range of problems with the tools now available.
I also expect it to become harder to judge someone’s ability from finished work alone. As polished reports and working code become easier to produce, explaining why a method was chosen and where it breaks down will matter more.
This change can create opportunities for newcomers as well. AI can help them complete small projects from beginning to end and gain experience across a wider range of tasks. But judgment is unlikely to develop if the experience ends with submitting a finished product. They need to compare results against the source material, correct errors, and have their work reviewed by others. The learning opportunities organizations provide should move in that direction too.
Does Cheaper Code Make Starting a Business Easier?
Uncertainty about employment naturally draws attention to entrepreneurship. If AI allows a small team to build a product, perhaps we can create work for ourselves instead of waiting for someone else’s hiring decision.
The lower cost of trying is undeniably attractive. An idea once abandoned for lack of developers can now be tested directly. It becomes possible to explore businesses serving smaller markets or more narrowly defined tasks.
But the calculation goes wrong when the claim that coding costs are approaching zero is applied to the whole business. Beyond the prototype, validation, operations, security, and customer support still cost money. Above all, customers need a reason to abandon their existing approach and choose the product.
In AI, tools and transformation, Benedict Evans identifies problems that easier coding does not resolve: discovering that a problem exists, determining what to build, and getting an organization to use it. In a company where work spans several departments, the performance of a single tool is rarely enough to determine whether it will be adopted.3
The same applies to a security service intended for Korean businesses. A prospective user liking the features does not close the deal. The provider must explain where data will be stored, whether the service can integrate with existing systems, and who will respond when something goes wrong, and how quickly. What a founder can solve technically and what they can reliably promise customers over time require separate calculations.
The fact that competitors have access to the same tools deserves an equally sober assessment. A founder may retain the benefits of lower development costs, but a growing number of similar products may instead drive prices down. As building becomes easier, deciding what to build and reaching customers account for more of the work of running a business.
An AI Business’s First Customers Buy Something More Concrete Than Technology
Paul Graham’s Do Things That Don’t Scale emphasizes finding customers directly and giving them close attention in the early days of a business. Activities that cannot yet be automated or repeated at scale help founders understand their customers and learn how to improve the product.4
I believe this advice remains relevant even when AI makes building fast. Faster development also makes it easier to keep adding features before meeting customers. The feeling that development is progressing can be mistaken for evidence that the business is progressing too.
Consider a service designed to support forensic work. A useful starting point would be to identify where analysts repeatedly get stuck. The product they need will differ depending on whether they spend too long organizing data, validating results, or explaining findings to another department. The feature that looks impressive in a demonstration may not be the one that actually reduces costs.
If I were assessing a business opportunity, I would spend more time looking at what happens after the first payment. Do customers still use the service the following month? Do they renew after something goes wrong? Is the work still profitable once the time spent on support is accounted for? There needs to be repeated evidence that the customer’s work has improved, beyond the fact that AI is involved.
Leaving a company also means leaving behind the organization that handles these matters on your behalf. While an analyst concentrates on an investigation, someone else negotiates contracts, collects payments, and manages customer expectations. A salary comes with that division of labor and a degree of predictability in income.
Going independent means choosing to take on those responsibilities as well. Uncertain employment does not, by itself, make entrepreneurship safer. Yet by understanding customers and establishing what it will take to deliver on promises, an idea that began in anxiety can become a workable business plan.
I Want to Leave Room for the Next Person to Begin
The more I think about AI, the harder it becomes to separate individual survival from the long-term health of an industry. For those already working, learning the tools and expanding the range of work they can take responsibility for are pressing concerns. I, too, want to keep testing what problems my skills will allow me to solve, rather than relying on the fact that I have acquired them.
At the same time, I do not want to mistake changes that benefit experienced workers for progress across the entire industry. If one person’s productivity rises while another loses the opportunity to gain experience, the benefits and costs fall on different people. The fact that my own work has become easier is not enough to judge the change as a whole.
As the lead of D4C, I want to pay closer attention to whether the people I research with can explain and verify their conclusions themselves. There is still a period of learning between obtaining an automated result quickly and becoming capable of taking responsibility for it. I believe teams that allow time for that learning will endure.
Starting a business remains an option worth keeping open. But I do not want to attach significance to the act of leaving a company in itself. I think I will be ready for the weight of entrepreneurship when I can identify work that future customers need and deliver it consistently—and, if there is someone I want to work with, take their livelihood into account as well. Until then, I would like to give the decision the time and thought it deserves.
In the age of AI, it has become difficult to say that learning a technical skill will secure a place for life. There are still reasons to learn: to understand a changing environment and to develop a sound basis for judging our own choices.
But we should not pretend that learning happens through individual determination alone. Everyone once needed the chance to do work they had not yet learned to do well. I would like to devote some of the time AI saves me to giving the next person that opportunity to learn. This field will still need people to carry on the work after today’s experts are gone.
References
- Bank of Korea. “Is AI Behind the Decline in Youth Employment? Changing Career Ladders and Policy Challenges” (title translated from Korean), 2026.
- David Autor. Applying AI to Rebuild Middle Class Jobs, 2024.
- Benedict Evans. “AI, tools and transformation”, 2026.
- Paul Graham. “Do Things That Don’t Scale”, 2013.