A New Year’s letter to a young person

Take the messy job

I am often approached by students and other young people for advice about their careers. In the past, my answers were often based on a piece of advice I myself got from Bengt Holmstrom: “when in doubt, choose the job where you will learn more.” In the last few years, there is a new variable to consider: the likelihood that artificial intelligence will automate all or large pieces of the job you do. Given that, what should a student choose today? The answers below are motivated by a book on artificial intelligence and the organization of work on which I am currently working with Jin Li and Yanhui Wu.

One way of thinking about this is that all knowledge work varies along one important spectrum: messiness. On one end, there is one defined task to execute, say helping clients fill their taxes. You get the expenses and payslips on email, you use some rules to put them on a form, you obtain a response. Over time, you become better at this task, and get a higher salary. On the other end of the spectrum, there is a wide bundle of complex tasks. Running a factory, or a family, involves many different tasks that are very hard to specify in advance.

The risk of the single-task job is that artificial intelligence excels at single tasks. Humans are still often in the loop, since the rate of errors in many fields is still too high to allow for unsupervised artificial intelligence. But the rate of errors is rapidly decreasing.

You may bet that fields vary in their tolerance for errors. Certain simple tasks, like content moderation, involve a high tolerance for risk: tech companies are comfortable with ‘shooting first, asking questions later’. But many others, from diagnostics to corporate communications, involve extreme risk-aversion on the part of the customer.

As long as a human is still needed to check the outputs, some value accrues to them. If AI drafts a contract that a lawyer reviews, signs, and takes responsibility for, the lawyer remains the supplier of legal services.

But the models are continuously getting much, much better at single tasks. If there aren’t any legal requirements to keep humans in the loop, even risk averse fields will eventually switch to unsupervised AI. When AI can produce finished code for straightforward tasks without human intervention, junior developers who once supplied that work compete with systems that produce it nearly free. The supply of programming services is no longer limited by human time, and the price of the service collapses to 0.

The result is that workers with simple tasks will become continuously more productive (and richer), until their work is worth nothing. A junior customer support agent gets more and more effective while the AI provides her the accumulated knowledge of senior customer support agents, as in the recent Brynjolfsson, Li; Ramond (2025) paper, until the AI is good enough that she can be replaced.

Notice that the autonomy threshold is not just given by technology. Firms and governments have a huge say in this, and they may choose to block adoption. The pressure to adopt better technology is strong, but do not overestimate it. Remember that many European countries still de facto ban Uber (don’t get my friend Nicolas Petit started about getting a cab in Florence!), despite the enormous quality of life improvement over traditional taxi monopoly, and that notaries are not a requirement of current technology, but a legal constraint that their strong lobby will always protect in continental legal systems. Even the most single-tasked civil servants will still have a job for a long time.

The end of work? Not so fast

The other option is to go for a messy job, where the output is the product of many different tasks, many of which affect each other.

The head of engineering at a manufacturing plant I know well must decide who to hire, which machines to buy, how to lay them down in the plant, negotiate with the workers and the higher ups the solutions proposed, and mobilise the resources to implement them. That task is extraordinarily hard to automate. Artificial intelligence commoditizes codified knowledge: textbooks, proofs, syntax. But it does not interface in a meaningful way with local knowledge, where a much larger share of the value of messy jobs is created. Even if artificial intelligence excelled at most of the single tasks that make up her job, it could not walk the factory floor to cajole a manager to redesign a production process.

A management consultant whose job consists entirely of producing slide decks is exposed. A consultant who spends half of her time reading the room, building client relationships, and navigating organizational politics has a bundle AI cannot replicate.

In 2016, star AI researcher Geoffrey Hinton leaped from automation of reading scans to the automation of the full radiologist job, and gave the advice to stop training radiologists.¹ But even fields that can look simple from the outside, like radiology, can be quite messy. A small study from 2013 (cited in this Works in Progress article) found that radiologists only spend 36 percent of their time looking at scans. The rest is spent talking to patients, training others, and talking with the nurses and doctors treating the patient.

A radiologist’s job is a bundle. You can automate reading scans and still need a radiologist. The question is not whether AI can do one part of your job. It is whether the remaining parts cohere in a manner that justifies a role.

To me, a key characteristic of these “messy jobs” is execution. Execution is hard because it faces the friction of the real world. Consider a general contractor on a building site. Artificial intelligence can sketch a blueprint and calculate load-bearing requirements in seconds. That is codified knowledge. But the contractor must handle the delivery of lumber that arrived late, the ground that is too muddy to pour concrete, or the bickering between the electrician and the plumber.

Or consider the manager in charge of post-merger integration at a corporation. Again, the algorithm will map financial synergies and redraw org charts, but it will not have the “tribal” knowledge required to merge two distinct cultures and have the tact to prevent an exodus.

Corporate law is increasingly vulnerable to automation because contracts are essentially code, but I would expect trial attorneys to subsist.

AI implementation itself could be the ultimate messy job. Improvements will require drastically changing existing workflows, a process that will be resisted by internal politics, fear, and legacy business models. For instance, law firms have always relied on “billable hours” to charge clients, a concept that will be useless in an AI world. But this organizational inertia is a gift: the transformation will be messier and more delayed than the charts suggest and it will require a lot of consultants, managers and workers, well versed in what AI can do, but with sufficient domain knowledge to know how to use it and how to redefine the process.

In the extreme instances, the feared AI transformation may not take place. Jobs defined by empathy, care, and real-time judgment will become the economy’s ‘luxury goods.’ In these fields, artificial intelligence is not your competitor; it generates the wealth (and lowers the costs of goods and services) that will fund your higher wages.²

More leverage

Usually, better companies are not created by transforming old ones, but by starting new ones. Starting a firm is the messiest job in the world, and artificial intelligence gives you a lot more opportunity to compete head on with larger companies..

For a century, large organizations dominated because only they could afford the fixed costs of specialized functions. If you needed both engineering and marketing, you needed scale to justify the overhead. AI reduces both the fixed and variable costs of that specialization. A single professional can now relinquish supporting tasks to AI and operate as a generalist. In a world where single tasks in areas like HR and finance are automated, you can ride a good idea all the way to a large company.

In our new book, we write about Base44, an AI-powered app builder. Founded by Maor Shlomo, a 31-year-old Israeli programmer, the company was built as a side project. He invested about $15,000 of his own money. He hired no employees, and he used Claude to write 90 percent of his frontend code and push updates daily, without any of the overhead that bogs down traditional engineering.

Within six months, Base44 had attracted over 250,000 users, generated $189,000 in monthly profit, and signed partnerships with major companies. Wix acquired it for $80 million.

Shlomo had no sales team, no marketing department, no HR. The tasks that once demanded human specialists were absorbed by the technology.

Conclusion

Given all the above, several investments appear to matter if you want to engage in careers in knowledge-intensive work (as opposed to the many crafts and trades, such as hair-dressing, plumbing, playing piano or being a chef, that are likely to remain untouched for a long time):

First, build deep, substantive knowledge in your field. As models get better, fewer humans will be good enough to add any value at all. Artificial intelligence can predict that a scan shows a tumor with 94 percent probability. But should the patient undergo surgery, radiation, or wait and monitor it? That decision depends on tradeoffs, some of which the algorithm cannot weigh. To exercise this type of judgment in a helpful way, you must know a domain deeply.

Deep domain knowledge will also make you good at AI implementation. Being able to change a given company’s processes to take advantage of the new technology is likely to employ a large number of people for a very long time.

Second, openness to new experiences, and the ability to learn new things quickly, will become even more important than it already is. Choose the job where you will learn most, but also the one where you will learn to learn. You will have to reinvent yourself throughout your life. New jobs will appear that we have not even imagined. If knowledge is the largest constraint that we face, then cheap knowledge will change everything, from medicine or the law to every field of research. The specific knowledge you learn today will depreciate faster than ever. What matters is the slope of your learning curve and your ability to adapt. For instance, in a job such as an early-stage founder or first employee, you do not have a fixed role; you have a set of problems. You must learn enough law or sales by the afternoon to survive, distinguishing essential signals from noise. Or, as a management consultant, you enter a room where everyone knows more than you, so you must absorb a new industry’s logic in days to structure vague problems. In tech, the role of product manager sits between the engineers and the sellers as a translator.

Third, seek leverage. In the past, your output was limited by your time—a chef can only cook for so many people. AI breaks this constraint. It allows a single writer, coder, or entrepreneur to serve a global market without a massive support staff. The constraint is no longer your production capacity; it is your ability to direct the machine. There is never been a better time to undertake entrepreneurial projects.

Fourth, if you do want to do a task that can plausibly be done by a model, location is more important than ever. A small number of cities, starting with San Francisco, Paris, London, and New York, are where almost everyone working and thinking about artificial intelligence is based. Go to these cities, or the closest approximation of them available to you, not just to work on these problems but to understand what possibilities may come your way.

Fifth, install Twitter. Twitter is a huge time sink, but it is also where all progress is happening, out in the open. You will have a huge advantage over virtually everyone not on the platform in understanding what is happening.

Sixth, learn to supervise the machine. The essential new skill is meta-cognition: recognizing when the AI is hallucinating, directing it toward the right problems, and verifying its outputs.

Finally, if all of these changes in the nature of work do happen, we are going to have much more leisure. In a recent paper, Betsy Stevenson points to the Japanese concept of ikigai, “that which makes life worth living”. The ability to derive meaning from sources other than your work is itself a form of human capital. Try to cultivate the habit of reading fiction. Try to spend less time watching videos and other forms of TV. Pick up hobbies: the hobby I would recommend most is starting a blog!

Happy new year and thanks for reading Silicon Continent!

1. “If you work as a radiologist you're like the coyote that's already over the edge of the cliff but hasn't yet looked down so doesn't realize there's no ground underneath him. People should stop training Radiologists now it's just completely obvious that within 5 years deep learning is going to do better than Radiologists.” 2016 Machine Learning and Market for Intelligence Conference in Toronto.

2. Without it, the future is stagnation or (given the adverse demographic situation) worse. So “Oh, let’s turn the whole economy into a Baumol economy” is not a solution.