How to get good at something useful: Part 8 Which skills will be most valuable in the future?

Ten years ago we told a lot of people they should learn to code. This worked out well for those who did: software engineering jobs grew rapidly for years afterwards.¹ But today, the question of whether to learn to code is much more ambiguous.

As I write this, the best AI models can already complete short, real-world coding tasks better than most human experts. But by the time you read this, that sentence will almost certainly be out of date. And these changes don’t only concern coders. Around half of people are worried they’ll lose their job to AI,² and they’re right to be concerned.

AI can now generate photorealistic videos, drive a taxi more safely than humans, and perform accurate medical diagnoses.³ Although many economists still dismiss the possibility, experts in the technology itself believe it will be possible to eventually build AI and robotic systems that will be able to do every economically important task more efficiently than humans.

Many assume this will drive down wages. But the reality is more complicated. Even if wages do eventually fall, that’ll likely be preceded by a period in which automation generates a huge amount of wealth, driving wages up on average. During that period, as AI drives down the value of skills it has mastered, it’ll drive up the value of skills it hasn’t.

As a job-seeker, your best strategy is to learn skills that AI will make more valuable, staying one step ahead of automation. It’s hard to predict exactly what those skills will be, but I think an even bigger mistake would be to ignore what’s happening.

You shouldn’t be training as a lawyer, accountant, financial analyst, actuary, writer, translator, or graphic designer right now without giving serious thought to the effects of AI.

Just as in 2015 it was possible to say that coding and data science were sectors with huge growth potential,⁴ it’s possible to make similar (if not certain) bets now about the skills that will put you in the best position over the next 10 years. That’s what we’ll explore here.

But to understand how automation will change the economy in the future, we first need to understand how it has transformed it in the past.

Reading time: 20 minutes

What people get wrong about automation

In the mid-1990s, ATMs started showing up in banks. At the time, people expected that these automated tellers would soon put human bank clerks out of the job.⁵ And indeed, they were almost right: the number of clerks per branch dropped from 21 to 13.

However, what they didn’t predict was that this also made it far cheaper to run each individual branch. In response, banks opened far more locations, which meant that overall employment of clerks actually increased for two decades, only they now spent their time talking to customers rather than counting money.

While it’s commonly assumed that automation decreases wages and employment, this example illustrates two ways that can be wrong. First, while automation decreased the value of the skill being automated (counting money), it increased the value of other skills (talking to customers).

Second, it shows that partial automation can even increase employment for people with the same job title, because it makes them more productive, which in turn makes employers want to hire more of them. In this case, fewer people could run a branch, providing better service to the same number of customers.

But there’s a final twist to the story: today, bank clerk employment is in decline.

While partial automation increased employment, the more dramatic automation made possible by online banking did indeed reduce it. This is a common pattern.

During the Industrial Revolution in Britain, textile production was automated significantly following the arrival of the spinning jenny, the water frame, and the spinning mule through the eighteenth century. Productivity increased greatly, while the industry remained one of the country’s biggest employers with over one million workers.

But the employment boom didn’t last forever. As automation went even further, for instance with the development of the power loom, and as foreign competition intensified, employment went into decline a generation later.⁶ A pattern of increasing then decreasing textile employment was also observed in the US in the nineteenth century.⁷

In the twenty-first century, some predicted that AI would soon replace radiologists. But so far, their employment numbers are up.⁸ This is probably because routine image analysis, the part AI is good at, is a small part of what radiologists actually do. Much of their time is spent talking to patients, conferring with other specialists, and dealing with edge cases.⁹ AI has made radiologists more productive rather than replaced them.

Sometimes automation means a smaller number of more highly paid workers using the new technology can replace a larger number of less skilled ones. There are signs this is happening in software engineering. Hiring of the most junior engineers shows signs of decreasing, but as of writing, senior employment is up, and average wages have increased.

Sometimes it has the opposite effect by making jobs easier to enter. The introduction of ride-sharing apps made it a lot easier to become a taxi driver. That reduced the wages of drivers by about 20% but tripled the number of people doing the job.¹⁰ Going forward, we might see things like nurses with AI advice able to do work that would have required a doctor before.

This doesn’t mean that radiologists, software engineers, and taxi drivers will never see declining employment in their fields. Employment of cashiers, secretaries, telemarketers, and special effects artists has been in decline for years as new technologies have automated their work. The point is that it’s hard to predict when automation will switch from increasing employment to decreasing it.

One intriguing possibility is that the pattern of rising then falling employment we saw with bank clerks could happen to the human race as a whole. As AI and robotics improve, they might initially make people more productive, meaning the world gets wealthier and wages as a whole increase, only to collapse later.

Economic models that allow for the possibility of full automation predict a rising-then-falling wage pattern. For example, the research institute Epoch AI modelled what would happen if an AI capable of doing 10% of all economically important tasks is created in 2026, and an AI capable of doing 100% of those tasks is created in the early 2030s. In their model, wages rise by 10 times until around 2037, creating a brief golden age for labour before they suddenly crash to zero after the final tasks are automated.

In contrast, if humans remain necessary for just 1% of tasks, the same model shows that wages increase indefinitely. The world gets richer and richer, and everyone works on the remaining 1% of tasks AI can’t do. The difference between 100% and 99% automation is enormous.

This second scenario can be compared to agricultural automation. Historically, almost everyone on earth worked in agriculture, and yet today in high-income countries only a couple of percent of people do. So we could say that the majority of jobs in these economies have already been automated.

And yet incomes in these countries are around 100 times higher now than they were back then. That means that on average people moved into much higher-paying and often newly created jobs. In some countries, such as South Korea, much of this transition was accomplished in just a single generation.

Regardless of whether wages continue to rise or eventually fall, the coming decades are likely to be very disruptive for the people whose jobs are automated. My purpose here isn’t to figure out the ideal societal response, but to help you navigate the years and decades ahead.

If wages on average are going to increase initially, then even if the value of some skills are going down, the value of others must be going up. Your best strategy is to focus on learning those skills increasing in value, maximising your contribution (and wages) in the time between now and full automation (if that ever happens).

Four types of skills likely to increase in value in an age of AI

We’ve reviewed all the economics literature and recent data we could find about AI automation, and combined that with some theory about how AI works, to create the following four categories of skills more likely to increase in value. Which skills to focus on is a moving target, but you can use these categories to evaluate your options at the time. You’ll want to look for skills that are valuable today, and that satisfy at least two of the four categories.

1. Skills that are hard to automate

The best way to build an intuition for what AI can and can’t do is to try to use the most cutting-edge tools to do real work. But we can also understand AI’s limitations by looking at how these systems are trained.

LLMs like ChatGPT are created by training them to predict patterns in (primarily) internet data. This makes them excellent at tasks that involve pattern matching, recall, and recombination of that data. That turns out to be a lot of tasks.

In 2017, experts assumed social skills and creativity would most resist automation.¹¹ Today, therapy is a popular use of AI, and it’s also surprisingly good at certain types of idea generation, creating visuals in novel combinations of artistic styles, or writing poetry most can’t distinguish from the human authored stuff.¹²

The internet contains a lot of data that’s relevant to many routine knowledge work tasks, things like fact-finding, translation, outlining basic context, and giving advice. And if a job is done on a computer, it’s also easy to track how it’s being done and gather even more data.

This is why most research finds the tasks most likely to be affected by this wave of AI are the more routine aspects of white-collar jobs, especially those in the 70th–90th percentile of income, which is those earning $100,000–200,000 in the US.¹³

On the other hand, AI remains bad at tasks totally missing from their training data. The internet contains huge amounts of text, but there’s no equivalent store of data describing physical movement. Creating this data will require building physical robots with cutting-edge sensors and having them move around the real world.¹⁴ So, unlike previous waves of automation, white-collar workers are likely to be affected before many blue-collar ones.

The physical tasks that will resist automation the longest will be the ones that require the most complex movement, that are done in unpredictable environments, and that require the highest reliability, because those factors increase how much data is needed to describe the task.¹⁵ Driving seems close to cracked, with Waymo operating fully self-driving cars in multiple cities and growing over 10% per month,¹⁶ but something like surgery or plumbing seems significantly further away.

Does this mean you should retrain as a plumber? Probably not. While wages in blue collar jobs might grow faster than many white collar ones, they’re starting from a lower base. Plumbing also doesn’t look attractive on the other three criteria to come. In addition, some white collar knowledge work skills will also prove hard to automate, which could make their value go up a lot. Which skills are those?

Since 2024, leading AI models have become much more than simple LLMs. The LLM is used as a base model, but it’s also taught to reason and complete tasks using ‘reinforcement learning.’ This is a bit like learning through trial and error. The model tries to solve a problem (like a maths puzzle), is graded on the accuracy of its output, and is then adjusted in a way expected to make it more accurate next time.

Reinforcement learning has led to dramatic progress in maths, coding, and answering known scientific questions — areas in which pure LLMs are weak — because these domains have objective answers that can be verified immediately.

But consider a task like building a company: it involves making judgement calls with no obviously correct answers, and success is determined over many years. It’s much harder to get reinforcement learning to work for a task like this. Other examples might be things like starting a new community, directing a frontier research project, having a conceptual insight, or setting organisational strategy. All of these require skills that aren’t clearly described within internet data, but are also:

• Messy — they lack clearly defined instructions and measurable outcomes.
• Long-horizon — they take time to implement and measure success.

This is why AI can solve maths problems most PhDs would struggle with but, as of 2025, is still worse than most 10-year-olds at playing Pokémon,¹⁷ a game which is open-ended and unfolds over days.

Most real jobs consist of projects that take weeks, are open-ended, and require coordination within large teams. Something most humans would find easy, like getting a set of shelves installed in an office, still seems significantly beyond what current models are able to do. This is why so few jobs have been automated to date.

This is likely to change, as AI is becoming better at longer and more open-ended tasks. But it’s where, absent a major paradigm change, humans will likely continue to add value the longest. Given this, it’s plausible many knowledge work organisations will become more top heavy, with an expanded class of managers overseeing AI-enabled teams who do the more routine work.

In fields like finance, this would be a continuation of long-term trends in which a smaller number of (often higher-paid) quants have replaced a larger number of discretionary traders.¹⁸ Likewise, it’s plausible that client-facing, direction-setting lawyers and consultants will see the value of their skills go up, while the need for junior analysts shrinks.

Next, in some cases, even if AI can perform a task proficiently, it might not be allowed to. The following are some categories of task suggested by economists who study automation in which people might demand a ‘human-in-the-loop’:

• Legal liability: Someone needs to be legally responsible for important decisions, such as with chartered engineers or court lawyers.

• High reliability required: AI systems make weird mistakes, so people will want human oversight, such as historians checking AI-generated research, control of expensive industrial machinery, or military leadership.

• Professional protections: Doctors and lawyers control certifications and form powerful lobbies that could block AI applications.

• Preference for human touch: Many people will prefer humans for certain services, perhaps as a luxury, such as with nannies, artists with compelling stories, and religious leaders.

• Physical presence needed: Many roles require someone physically present, and in some cases people might not be willing to fully hand these responsibilities to robots for a long time, such as police or teachers.

• Institutional inertia: Many organisations will be slow to adopt AI, keeping humans in important jobs for longer than otherwise makes sense, perhaps including in government or in monopolistic companies.

These factors could work to protect certain oversight positions in white-collar professions, as well as protecting certain social jobs, at least for a while.

In addition, AI automation will be slowed by a lack of physical infrastructure. Even if robotics suddenly started working perfectly, robot production today is only in the millions, so it would take many years to build the billion or so robots required to do all physical jobs.

Even in the digital world, lack of physical infrastructure might well slow down adoption of AI. If ‘AGI’ were created tomorrow, it would probably take a ton of computing power to run, making it prohibitively expensive to use to automate lower-paid jobs. That would mean its deployment would initially be limited to the highest-value applications, like assisting with R&D.

2. Skills needed for AI development

In 2014, Facebook bought WhatsApp for $19 billion back when it had 450 million users but only 55 employees. The declining cost of software has made it easier and easier for a small group of people to rapidly build a product that can serve millions of people. Now this trend seems to be entering a new phase.

Generative AI startups are growing fast even compared to the previous generation of software as a service (SaaS) companies, themselves famous for their speedy growth:

As AIs move from being reactive chatbots into proactive agents, the skills required to direct these ‘digital workers’ will become incredibly valuable.

You will be able to use these workers to rapidly build and iterate on new products, but you could just as easily have them help you set up a nonprofit on a shoestring budget, search for new cancer drugs, or provide expert medical advice 24/7 anywhere in the world. Key skills could include:

• Spotting problems and deciding what to focus on
• Understanding different AI models and designing around their weaknesses
• Writing clear project specifications for AI agents to follow
• Understanding what users really want
• Designing systems to manage AI workers, including error-checking
• Coordinating with people and bearing responsibility for results

These skills are similar to managing humans today, and there’s already evidence that competent human managers are better at managing AI teams.

What’s crucial is that these kinds of skills are complementary to AI. As AI gets better, they become more needed, rather than less.

A tailor of bespoke Neapolitan suits, descended from a long line of tailors, is not something AI can easily replicate, so it satisfies the first category — it’s hard to automate. But tailoring is not complementary to AI, so while it might retain its relative value, it doesn’t increase.

To give another example, as AI becomes better at generating solutions to scientific problems, the key question becomes what problems to direct those AIs towards. This skill in ‘research taste’ is also messy and long-horizon, so seems harder for AI to learn. For both reasons, it likely increases in value as AI gets better.

Other complementary skills involve knowledge of the fields needed for AI scale-up, such as AI software and hardware development; robotics development, manufacture and maintenance; the construction of datacentres and factories required for these; and cybersecurity.

3. Skills that produce things people could use far more of

I only need to file my taxes once a year, so if AI can do it more cheaply, I will spend less on my accountant. I’m not going to start filing taxes more often with the money saved. In contrast, as we saw, the arrival of ride-sharing platforms made taxis much cheaper and easier to use, and as a result, people took way more rides.

There are many other areas where, as what’s on offer gets better and cheaper, people use them more and more. As people get richer they spend more of their income on healthcare, suggesting that even if technology made healthcare more efficient, people would simply use far more of it.¹⁹

Likewise, if luxury experiences like exotic travel suddenly became affordable, people would choose to travel a lot more. This is in contrast to jobs that exist to satisfy legal requirements, like licensing or certification. If they can be done much more efficiently using AI, people will simply hire fewer human workers.

Other jobs, like nurses in the UK’s National Health Service, have employment effectively set by the government, so even if their efficiency increases, wages and employment might not. Likewise, the number of academic jobs is not mainly set by market forces, which means their employment could remain steady, even as AI makes researchers more productive.

AI automation will also likely create a huge amount of wealth, potentially concentrated in the top 1% who own most capital, and therefore who own the AI systems that are doing the automation. Increased income inequality will spike demand for luxury goods. Something like providing personalised tea-tasting events in San Francisco would be hard for AI to do and would likely also see increasing demand.

4. Specialised skills that are hard for others to learn

Consider a job like being a server at a fancy restaurant. People often eat out more as they get wealthier, and this is the sort of physical, social-skills-heavy job where people might retain a strong preference for a human touch, so I expect there to be plenty of jobs like this going forward.

However, it’s also a job that other people can enter relatively quickly. The skills that will see the largest increases in value will be those where it takes time for the labour market to respond.

This favours learning specialised skills that match your unique strengths. For example, construction work is a complex physical job that will be harder to automate. However, a construction worker who learns a trade like electrician, and specialises in data centres, is more likely to become a critical bottleneck, and see the value of their skills increase a lot faster. In fact, electrician wages in Virginia, where many datacentres are being built, are already 37% above the national average.²⁰ So which specific skills will be most valuable in the future?

Which skills will be most valuable in the future?

Let’s combine this discussion with our findings from the last couple of sections to make a list of valuable work skills. This list isn’t comprehensive, but should give you some ideas about where to focus.

Broadly speaking, the most valuable skills will be those that are most needed in the impactful categories of jobs we discussed in part six, are more likely to increase in value given automation (satisfying at least two of the four categories we’ve just covered) and are faster to learn relative to their value, since that means each hour spent learning yields a higher return. The latter could be of increasing importance as the job market starts to change faster. Spending eight years in medical school, for instance, is a much riskier proposition today.

Another quality that makes skills more valuable is how transferable they are to different jobs and problems. Management is a relatively transferable skill, because basically every organisation needs managers, so even if one industry disappears, or you decide you want to switch into a totally different area, your skills are more likely to remain useful. Become an expert in early Roman portraiture, however, and your options are more limited.

Some skills are so transferable they’re useful in basically every career. Here are some that seem likely to increase in value in the future and which we consider especially useful for impactful career paths.

Although they’re not all the classic ‘work skills’ you’d list on your LinkedIn profile, they’re often more worth learning, because they’ll make you more effective at everything else you want to do. You can start practicing them in whatever job you’re doing now.

1. Deploying AI to solve real world problems: If I had just one piece of advice right now, it would be to learn to use AI to accomplish real work. This advice has already become a bit of a cliche, but is still often not applied. Some universities even try to ban their students from using AI. As AI becomes more powerful, the skills involved in effectively directing AI systems increase in value, in particular those that address the messy gaps where people want a human-in-the-loop. Having concrete AI expertise also makes you more credible when working on its social impact.

2. Taking care of yourself: Ambitious people often don’t take care of themselves. This can make them burn out and ultimately be less successful. Taking care of yourself starts with the basics: getting enough sleep, exercising regularly, eating well, and maintaining your close friendships. Building these habits can have a big impact on your day-to-day energy and happiness.

   In addition, about 30% of people in their twenties have some kind of mental health issue.²¹ If you’re one of them, learning to cope with it better is usually one of the best investments you can ever make — both for your own sake and your ability to help others. This often involves some form of therapy, such as cognitive behavioural therapy (CBT) and its variants, which are practical, solution focused, and have the clearest evidence base. You should speak to a psychiatrist about other treatment options, too. Trying to tackle the world’s biggest problems will be challenging at times, and making yourself more resilient is well worthwhile.

3. Personal effectiveness and agency: It’s often possible to significantly increase your productivity by developing good habits, like setting daily goals and learning to run meetings. These skills are relatively quick to improve and help you in every job. On a more advanced level, if you can become the type of person who can work on difficult, ill-defined problems over many years, that’ll be of huge value. And this is something AI is likely to continue to struggle with.

4. Prioritisation, forecasting, and decision-making: In 2017, we did an analysis of which skills were most commonly required in the most attractive jobs, and found that judgement and decision-making came out on top. Since then, the proportion of jobs that require decision-making skills has been expanding, and these jobs have seen faster wage growth than average.

   If you want to do good, these skills are probably even more important, because some projects have much more impact than others, and these skills are needed to identify what’s best. Looking forward, as technology makes it easier to get stuff done, the question of what’s worth doing in the first place will become ever more important.

5. Social skills: Every job involves working with others at least to some degree, and the importance of social skills has also been increasing over time. Looking forward, even though AI is good at certain types of social interaction, people will still want relationships with other people. Smaller teams of humans managing powerful AI agents will need to coordinate extremely well because their leverage will be so much greater. Social skills are also a core component of leadership and management, and vital in many of the most impactful paths, such as policy and communications roles.

6. Learning how to learn: Being able to learn things quickly is useful almost no matter what you want to do, but as technology changes faster, being able to rapidly retrain becomes ever more valuable. AI has also meant you can get basically free 24/7 private tutoring on almost any skill, which many people say is enabling them to learn a lot faster than before.

It’s often worth spending some months taking the lowest-hanging-fruit ways to improve each of the skills listed above (explaining how would take a whole chapter, so we’ve put the best resources in our guide to being more successful).

But these skills are not typically what people make the main focus of their career. Here are some skills that are still highly transferable, seem relatively future-proof, and can become such a focus:

1. Leadership and management skills: These include people management, project management, entrepreneurship, and organisational strategy. Leadership and management skills have always been among the most valuable because they help coordinate and direct teams of people, compounding your impact.²² Their importance could increase going forward as AI automates more routine knowledge work, allowing small groups of leaders to get more done than before, unlocking moonshot projects that were previously too expensive.

2. Operations management: Every organisation needs people to actually run things. This isn’t about big-picture strategy, but rather all the day-to-day realities all organisations must handle, like recruitment, setting up financial systems, and office management. Some aspects of this skill, like more routine admin tasks, are being automated by AI, but humans are still crucial for more complex tasks that require good judgement. It’s plausible operations teams will get smaller, but more organisations will be founded, and they’ll grow faster than before. Among organisations we think are impactful, these skills are highly in demand today.²³

3. Policy and political skills: We’ve already seen the huge influence government has on many of the world’s most pressing problems, and how individual government workers can have a surprising amount of influence on it. Looking forward, even if much routine analysis and bureaucracy gets automated, citizens will want decision-makers to be real people for a while longer. That means the nebulous, long-horizon skills of understanding how the system works and getting things done in government will likely remain valuable.

4. Communications: Every organisation needs skilled communicators, both internally and externally, to explain their mission and offering, and they’re especially in demand at the organisations we think are most impactful right now.²⁴ You can also use these skills to help spread important ideas. As it becomes easier to create large volumes of reasonable-seeming content with AI, it’s possible the value of creating content will decrease, but the skill of figuring out what to produce in the first place will increase in importance, as will the value of having a sense of taste. Likewise, trust and an authentic relationship with your audience become more important. More practically, this could look like mastering social media, building a Substack following, acquiring PR expertise, or becoming good at running in-person events.

These four skills can be applied to almost any global problem, but many of the highest-impact roles also require specific, in-depth expertise. Next we’ve made a list of more specialist skills to consider learning. We chose them because they seem the most useful given our views about pressing problems and the likely trajectory of AI. While almost everyone should be able to agree on the skills listed above, the list below could look quite different depending on your views about what matters.

Some of the following skills also take longer and are harder to learn, but true expertise seems likely to be one of the last things that will get automated,²⁵ and experts of many kinds will be required to provide oversight of AI systems.

1. Machine learning research and engineering: This is the technical side of getting AI systems to work. As AI becomes more useful, the value of making it 1% more efficient goes up. This means that even as much of ML engineering is itself getting automated, the remaining parts that only humans can do are becoming more valuable (as reflected by rapidly increasing salaries).

Within this skill, you could focus on the more technical aspects of AI safety research, which we believe is one of the most valuable fields of research in the world today. Many promising research ideas in that field are bottlenecked by a lack of engineers able to implement them. Another option would be to use this skill to apply AI to important global problems, such as building AI tools to improve government decision making. Having this skill also gives you much more credibility if you decide to work in AI policy or communications.

Finally, it commands huge salaries, which you can use to earn to give. One risk is that working on frontier AI development accelerates AI progress without improving society’s capacity to adapt, so if you learn this skill, we recommend focusing on applying it to AI safety, or the application of existing AI technology to important social problems.

1. AI governance and strategy: This means studying the potential impacts of AI, and how government, philanthropists, and other key actors can best respond. This is a diverse area which overlaps with many other fields including economics, law, policy, ethics, forecasting, and macrostrategy. Given the risks and rapid rate of change, there’s currently a huge need for more expertise of this kind, but not many people have more than a couple of years’ experience, so it’s possible to get to the forefront relatively quickly.

2. Cyber and information security: As the world becomes more digital and robotics becomes more integrated into our lives, it becomes more important to guarantee these systems are secure (after all, no-one wants to be kidnapped by their robot butler). At 80,000 Hours, we’re especially concerned about the theft of new biotechnologies and advanced AI systems themselves. It’s hard to see how the coming years go well if these technologies can be immediately stolen by bad actors. This skill is also extremely in demand commercially.

3. Expertise in an emerging power: China is likely to play a crucial role in shaping the coming decades, but decision-makers in the West tend to understand China poorly and often fail to coordinate with it to solve global problems. This means expertise in China-West relations, and especially coordination around AI and other catastrophic risks, seems highly valuable.

More broadly, the world is transitioning from an era of US dominance to a multipolar system with several great powers, including countries like India, Russia, and Saudi Arabia, expanding their spheres of influence. These dynamics are going to affect the development of many issues going forward.

1. Expertise in AI hardware: AI chips may turn out to be the oil of the next century. Whichever country has the best and the most chips will be able to deploy the most effective AI systems, and could therefore wield the most economic power. If this turns out to be the case, understanding how these chips work and how they are made could become one of the most valuable skills to possess. In particular, one avenue for the governance of AI systems focuses on the design and availability of chips. Among people working on reducing the risks associated with AI, there is vastly more expertise in software than hardware.

Here are several other valuable areas of specialist expertise, though they’re less targeted at the most crucial gaps, and in some cases less transferable:

• Economics, especially the understanding of growth and automation, though also its general toolkit
• International relationship, security studies, and public policy
• Applied mathematics, since those skills are useful in many other fields
• Robotics development and maintenance (as well as the construction of datacentres and factories) since they often become huge bottlenecks, making them valuable skills, especially for earning to give
• Engineering, especially for pandemic prevention
• Synthetic biology and bioengineering
• Certain branches of philosophy, such as moral philosophy and philosophy of mind
• Certain areas of history, such as economic history, macrohistory, and the history of social change and important ideas

We’ll come back to how to pick what to study in the next part of the guide, which asks which jobs best advance your career.

Finally, the value of skills used in jobs that require a physical presence or involve complex physical manipulation — such as those used in law enforcement, construction, teaching, and surgery — and the skills used to provide luxury services or where a human touch is highly valued — like mediator, conference organiser, artist, or bespoke craftsperson — will probably keep pace with the rest of the economy, though won’t obviously outpace it.

Which combinations of skills are best?

Now you’ve gained a sense of which skills will be most valuable in general, it’s worth thinking a little about the combinations that could prove especially useful. In a widely cited paper, the economist David Deming shows that outside of computer technology, jobs requiring mathematical skills have represented a decreasing share of employment, and have had below-average wage growth for decades, while jobs involving social skills have grown more than average.²⁶

However, he also shows that the biggest growth has been for jobs that require both technical and social skills. This could be because they allow you to act as a bridge between more widely deployed technical systems and real-world applications.²⁷ In particular, we think there’s a huge need for people who understand emerging technologies like AI and synthetic biology, but also have skills in policy and communication.

You should also think about whether to focus on one main skill, or build a portfolio of several. In some areas, success is more a matter of being exceptional at one thing — for example, a salesperson is mainly judged based on how many sales they bring in — and so in those areas you should focus most of your efforts on that core skill. Having one impressive achievement usually also opens more doors than having several moderate ones.

In other areas it’s useful to have an unusual combination of skills, becoming the best person in that niche. Many organisation-building roles require generalists with some knowledge of management, operations, and marketing, communications, as well as sufficient understanding of what the organisation actually does.

But in every case, remember what the history of automation tells us about its likely future. While most people assume automation drives down wages, that’s only true when it’s nearly 100%. Until then, automation drives down the value of some skills, but drives up the value of others. The aim isn’t to find a single job that will never be automated, but rather to ride the wave, moving into the skills that become most valuable at each time.

So which jobs will let you learn these skills the fastest?

Take a break

Read next:  Part 9: Which jobs best advance your career?

Continue →

Or see an overview of the whole guide.

Get the whole guide as a book

If you find this guide helpful, preordering the book (especially from a physical retailer!) is a great way to support us.

Preorder the book