In 2020 David Deming and Kadeem Noray published in the Quarterly Journal of Economics a study that follows US graduates’ careers over time, not just at first hire. Technology graduates start with the highest entry salaries of anyone. Then something happens that degree rankings never mention: the advantage erodes over the first decade, shrinking by roughly half.
Not because those people get worse. Because the specific skills they were hired for get replaced, and whoever enters later already has the new ones.
That’s the figure to keep in mind whenever someone offers a list of skills to bet on for the next five years. The right question isn’t which skills will be worth most, it’s which ones depreciate fastest. And that one is measurable.
The statistic everyone cites and nobody can check
Two numbers have circulated for years. First: the half-life of a skill is supposedly five years, two and a half for technical ones. Second: somewhere between 40% and 50% of required skills will change within a few years.
The first number has no traceable primary source: it gets repeated from report to report without the methodology ever appearing. The second, which shows up in the World Economic Forum’s future-of-jobs reports, does have a clear source, and that’s precisely why it needs care. It’s a survey of HR leaders at large companies stating what they expect: a snapshot of employer expectations, not a measurement of what happened.
That isn’t pedantry. A stated forecast and an observed depreciation are different objects, and only the second one has historical series behind it.
What actually depreciates
Deming and Noray’s mechanism is precise and worth understanding, because it applies to anyone, not just to programmers.
Someone entering a technology-intensive occupation is paid well because they hold scarce, newly arrived skills. But those same skills are tied to tools and paradigms that get superseded. A few years later the fresh graduate has learned them at university, and the head start disappears. The counter-intuitive consequence is that exit rates from technology occupations into other jobs are high: not because the work is unpleasant, but because the economic advantage gets used up.
There’s a second way to see the same thing, from Edward Lazear. In his framing almost no skill is genuinely firm-specific: skills are general, but each firm combines them with different weights. What you lose when you change jobs isn’t the skill, it’s the combination. That explains why someone excellent in a narrow setting can be worth little outside it, without having become incompetent.
And there’s a third mechanism, the most banal and most underrated: decay from disuse. Winfred Arthur and colleagues aggregated in 1998 nearly two hundred datasets on skill retention after training. The finding: after about a year without practice, performance drops by more than nine tenths of a standard deviation, with the sharpest drop on complex cognitive tasks rather than motor ones.
Translated: a huge share of what you call an “obsolete skill” isn’t obsolete. It’s simply unpractised.
What is growing, according to the data
Here there’s a solid historical series. David Deming reconstructed, also in the Quarterly Journal of Economics, how demand by social-skill intensity moved in the US labour market from 1980 onwards.
The result: social-skill-intensive occupations grew by around twelve percentage points as a share of total employment, and wages grew in them too. But the detail that matters most is different: the strongest growth is neither in highly mathematical occupations nor in highly social ones taken alone. It’s in those requiring both at once.
| Skill profile | 1980-2012 trend |
|---|---|
| High maths + high social skills | Strongest growth in employment and wages |
| High maths, low social skills | Modest growth, less impressive wages |
| Low maths, high social skills | Employment growth, more contained wages |
| Low on both | Contracting |
Deming’s theoretical model explains why: social skills lower the cost of coordinating, which lets people specialise more. They’re a multiplier on technical work, not an alternative to it. I’ve written elsewhere about why the hard versus soft skills opposition is badly posed: this is the evidence that shows it best.
The useful question isn’t “which skill should I learn”. It’s “how fast does this skill depreciate, and what does it cost to keep it alive”.
The directions supported by structural data
With the same caution that applies to any forecast, these are the trends resting on documented mechanisms rather than on expectation surveys.
| Direction | Why it holds | Expected depreciation |
|---|---|---|
| Relational and coordination skills | Deming’s series, forty years of growth | Very slow |
| Basic quantitative reasoning | Cross-sector, independent of tools | Slow |
| Judgement under ambiguous data | Not reducible to writable rules | Slow |
| AI tool literacy | Growing demand but unstable paradigms | Fast on tools, slow on concepts |
| Mastery of one specific software or platform | Useful now, replaceable | Very fast |
| Sector regulatory knowledge | High rent while the rule stands | Discontinuous: changes all at once |
The second-to-last row is the important one. It doesn’t mean a tool isn’t worth learning: it means the return on that hour is concentrated in the first few years and should be accounted for that way.
How to choose what to learn
- Separate concepts from tools. Before studying something, ask whether you’re learning a principle or an interface. Principles transfer, interfaces have to be relearned at every version.
- Estimate the half-life before you start. A skill tied to one commercial product has a short horizon by construction. That’s fine, as long as you know it when you decide.
- Build pairs, not peaks. Deming’s data says the value sits in the combination of technical skill and the ability to coordinate. The second half is rarely studied on purpose.
- Apply what you learn within a few weeks. Decay from disuse is documented and brutal: a course without application is spending with a return close to zero.
- Review at increasing intervals, not in bursts. Spreading practice over time beats the same amount of study crammed together: it’s one of the most replicated effects in learning psychology.
- Check how much of your value is tied to one context. If the answer is “a lot”, you’re accumulating firm weights, not skills. The cost of finding this out late is quantified in reskilling.
- Treat AI as a conceptual skill. The tools will change every few months; understanding how they work and where they fail won’t. The distinction is explained in generative AI.
My take: reduce the risk instead of chasing the peak
I think the problem with lists of future skills isn’t that they pick the wrong items, it’s that they optimise the wrong variable. They maximise the expected value of a bet, when the rational thing under high uncertainty is to reduce variance: prefer slowly depreciating skills even if the peak payoff is lower.
I also think the Deming and Noray result is the most uncomfortable in the whole literature, because it contradicts the dominant advice of the last fifteen years. “Study a technical subject and you’ll be fine” is true for the first salary and becomes progressively less true afterwards. Anyone telling that to an eighteen-year-old should at least tell them the second half too.
If I had to bet, I’d say this: the most valuable skill of the next five years will be how quickly a person can become mediocre in a new field. Not excellent. Mediocre, fast, many times over.
Frequently asked questions
Are technical skills no longer useful? Very much so: they’re the part that gets you into the market. The point is that their premium gets used up, while the premium on coordination skills doesn’t.
How long does a skill really last? It depends on how tightly it’s bound to a specific tool. A statistical principle lasts decades; mastery of one software version lasts as long as that version.
Should I learn to code? The programming logic, yes, almost always. The specific language matters less than it seems, and it’s the part that depreciates first.
How do I avoid forgetting what I study? By applying it early and reviewing at spaced intervals. Decay from disuse is the most frequent cause of skills that look obsolete and aren’t.
Can social skills be studied? Partly yes, if you treat them as concrete tasks (running a meeting, giving feedback, negotiating) rather than as character traits. A starting point is critical thinking.
Sources
- Deming, D. J., & Noray, K. (2020). Earnings Dynamics, Changing Job Skills, and STEM Careers. Quarterly Journal of Economics, 135(4), 1965-2005. The erosion of the initial wage premium in technology occupations.
- Deming, D. J. (2017). The Growing Importance of Social Skills in the Labor Market. Quarterly Journal of Economics, 132(4), 1593-1640. The growth of social-skill-intensive occupations and the effect of combining them with mathematical ones.
- Arthur, W., Bennett, W., Stanush, P. L., & McNelly, T. L. (1998). Factors That Influence Skill Decay and Retention: A Quantitative Review and Analysis. Human Performance, 11(1), 57-101. The meta-analysis on decay from disuse.
- Lazear, E. P. (2009). Firm-Specific Human Capital: A Skill-Weights Approach. Journal of Political Economy, 117(5), 914-940. General skills with firm weights.
- Autor, D. H., Levy, F., & Murnane, R. J. (2003). The Skill Content of Recent Technological Change: An Empirical Exploration. Quarterly Journal of Economics, 118(4), 1279-1333.
- Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed Practice in Verbal Recall Tasks: A Review and Quantitative Synthesis. Psychological Bulletin, 132(3), 354-380. The effect of spacing practice over time.
- Neuman, S., & Weiss, A. (1995). On the Effects of Schooling Vintage on Experience-Earnings Profiles. European Economic Review, 39(5), 943-955. Faster depreciation of education in high-technology sectors.
- Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting Meta-Analytic Estimates of Validity in Personnel Selection. Journal of Applied Psychology, 107(11), 2040-2068.
- Bessen, J. (2015). Learning by Doing: The Real Connection Between Innovation, Wages, and Wealth. Yale University Press. On-the-job learning as the main source of the skills actually used.
- World Economic Forum. Future of Jobs Report. Periodic series. Source of the estimates on changing skill requirements: an employer survey, not a retrospective measurement.
This article was written following the process described in how we write. If you find a figure reported wrongly, write to me: I’ll correct it and note the correction at the bottom of the article.



