Chaos and Order

💡 왼쪽 원문을 읽으면서 오른쪽에 따라 써보세요. Tab 키로 힌트를 받을 수 있습니다.

Introduction — What That Article Is Actually About
What Ericsson Actually Claimed, and What He Did Not
The Macnamara Meta-Analysis, and the 2018 Corrigendum
Rebuttal, Counter-Rebuttal, and a Replication Failure
The Most Uncomfortable Evidence, from the Genetics Side
So What Claim Honestly Survives
Optimizing Only the Part You Control
Closing — Admitting the Variables Outside Your Control Is What Makes the Ones Inside It Visible
References

Introduction — What That Article Is Actually About

On July 26, 2026, "The Illusion of Talent" went up on GeekNews and picked up 56 points and 12 comments. The original is a post on Jiuk Gwag's personal blog.

From the title it sounds like a piece about the psychological dispute surrounding talent and practice. It is not. The original is a personal essay about a junior developer's career anxiety, and it cites not a single study. The argument runs like this: many juniors define "development talent" solely as a particular type — someone who enjoys hard technical problems and writes good code — decide they do not fall into that category, and quit. But the developers who actually deliver work with different abilities: documentation, communication, domain understanding, the user's point of view, translation between job functions. The author explains this with a poker analogy, saying that not everyone is chasing the same royal flush; each of us is holding a different hand. As far as I could check, this article was never posted to Hacker News.

As an essay, it is a decent piece. The problem is when writing like this gets summarized into the much stronger claim that "talent is an illusion" and circulates in that form. The actual literature does not support that stronger claim. To put it precisely, the deliberate-practice research has never said that talent is irrelevant, and the data of the last twenty years points in the opposite direction if anything.

That does not mean the conclusion is "so talent is everything" either. Follow the literature all the way through and what survives is narrower, and therefore more useful. In this article I will go as far as that narrow claim.

What Ericsson Actually Claimed, and What He Did Not

The starting point is Ericsson, Krampe & Tesch-Römer (1993), Psychological Review, vol. 100, no. 3, pp. 363–406.

Study 1 is designed like this. Violin majors at the West Berlin music academy are divided by professors' nominations into three groups — the "best," who are candidates for international soloist careers; the "good," matched for sex and age; and a group headed for careers as music teachers, with relatively lower attainment. There are ten people in each group. Added to this, as a validation group, are ten middle-aged professional players from the Berlin Philharmonic and the radio symphony orchestra. Practice time was estimated by retrospective self-report and cross-validated against seven days of practice diaries.

The core numbers are the hours of solo practice accumulated by age 18.

Best: 7,410 hours
Good: 5,301 hours
Music teachers group: 3,420 hours

In Study 2, on piano majors, the 12 experts had 7,606 hours and the 12 amateurs had 1,606 hours.

Let me clear up one widespread misunderstanding before going on. The phrase "the 10,000-hour rule" was not coined by Ericsson. Across the entire 44-page paper, the number 10,000 appears exactly once, in the general discussion, in a sentence that passes over the amount of practice the best group had accumulated by age 20. He never called it a rule and never presented it as a threshold. The framing as a rule came fifteen years later, from Malcolm Gladwell's Outliers (2008).

One more thing. Ericsson's team never claimed that talent does not exist. Within the paper they acknowledge plainly heritable factors such as height, and they leave open the possibility that genetic individual differences in emotionality or activity level affect the capacity to sustain deliberate practice. Their claim was not "there is no talent" but that the major facts of expert performance can be explained without relying on the scarcity of talent. What disappeared in the process of popularization is exactly this carefulness.

And look at the sample size again. Ten people per group. That is not unusual for a sample in 1993 psychology, but it is not a weight this claim can bear thirty years later.

The Macnamara Meta-Analysis, and the 2018 Corrigendum

The most frequently cited counterargument is Macnamara, Hambrick & Oswald (2014), Psychological Science, vol. 25, no. 8, pp. 1608–1618. It covers 88 studies, 111 samples, 157 effect sizes, with a total N of 11,135.

There is an important fact here: the widely cited numbers were already corrected in a 2018 corrigendum. The authors discovered that they had misapplied the correction for dependent effect sizes and issued a corrigendum in Psychological Science, vol. 29, no. 7, pp. 1202–1204. Put the two versions side by side and they look like this.

Domain	2014 original paper	2018 corrigendum
Overall	12%	14%
Games	26%	24%
Music	21%	23%
Sports	18%	20%
Education	4%	5%
Professions	Less than 1% (not significant)	1%

The authors stated that this correction does not affect the substance of the conclusions, and that is in fact the case. Still, the "12 percent" that circulates on the internet is technically a superseded number, so if you are going to cite it, 14 percent is the accurate figure to use.

There are two things to read out of this table.

First, variance explained of 14 percent means the remaining 86 percent is other factors. That 86 percent contains genetics, starting age, quality of coaching, opportunity, and measurement error, all of it. It does not mean practice is unimportant; it means practice alone cannot explain most of the individual differences.

Second, the variation across domains matters more than the headline. In domains like games and music, where the rules are stable and feedback is immediate, practice explains something in the twenties, but in education and the professions it drops to 5 percent and 1 percent. Which side is software engineering closer to? It is a domain where the rules keep changing, feedback is delayed, and the definition of performance differs from organization to organization. It is closer to the professions than to music. This does not mean practice is meaningless; it means that in this domain, what you choose to practice on decides the outcome more than how much you practice.

And there is one more coldly sobering number. If you look only at the subset where practice volume was measured objectively — through diaries or computer logs rather than retrospective memory — the explanatory power drops to 5 percent. That is a figure showing just how generous retrospective self-report is.

Rebuttal, Counter-Rebuttal, and a Replication Failure

Ericsson pushed back. In Ericsson (2016), Perspectives on Psychological Science, vol. 11, no. 3, pp. 351–354, he argues that the meta-analysis coded "deliberate practice" far too loosely. His examples are that group activities, watching matches on TV, and play and competition were all summed together as long as they were domain-related. Deliberate practice as he defined it is an individual activity designed to target weaknesses with immediate feedback, and that is not the same thing as "total time spent on the domain." This is a legitimate objection.

The Macnamara team's counter-rebuttal (2016, same issue, pp. 355–358) went two ways. Recomputing after excluding the composite measures Ericsson objected to moved the explanatory power in the sports domain from 18 percent to 17 percent, which is barely a movement at all; and Ericsson's own earlier papers had not restricted deliberate practice to teacher-designed activities either.

But the heaviest result in this dispute was not the dispute itself; it was an attempt at replication. Macnamara & Maitra (2019), Royal Society Open Science, vol. 6, no. 8, 190327 redid Ericsson's 1993 study under a double-blind procedure. It used 39 violinists, 13 in each group.

Here is the result. The core finding that accumulated deliberate practice corresponds to skill tier did not replicate. If anything, many of the best players had accumulated less solo practice than the average of the "good" group. The ordering of the original paper was inverted. The relationship between solo practice and performance came out at an eta squared of 0.26, that is, 26 percent of the variance, and while the authors acknowledged that this value is by no means small, they wrote that it does not support the original claim that differences in level of performance are largely explained by differences in amount of practice.

Summarizing the state of the literature here: deliberate practice has a real effect, and its size runs from roughly 5 percent to 26 percent depending on the domain. That is a considerable effect, but not a decisive factor. Neither "practice is everything" nor "talent is everything" is supported by the data.

The Most Uncomfortable Evidence, from the Genetics Side

The result that hurts the pro-practice position most came from Mosing et al. (2014), Psychological Science, vol. 25, no. 9, pp. 1795–1803. It is a sample of 10,500 Swedish twins.

Two things came out of it. First, the amount of practice is itself 40 percent to 70 percent heritable. How much you practice is not a pure product of will but to a substantial degree an inherited disposition. Second, and more decisively, when they compared monozygotic twin pairs who differed in how much they practiced, the twin who practiced more was not reliably better. In the strongest design available, one that effectively controls for genes and family environment, no causal effect of practice appeared.

There is a legitimate objection about the limits of this study, and Ericsson's point above applies directly. The sample consists of population-based twins, not experts at conservatory level. There is no guarantee that a null causal effect in the amateur range generalizes to the expert range. This is not a forced defense but a genuine restriction of scope.

Even so, an implication survives. The very propensity to practice that "just practice more" advice is addressed to is itself a product of individual differences. Plans built on the assumption that it is a variable controllable by willpower fail often, precisely because of that assumption.

So What Claim Honestly Survives

Once you get this far, the proposition that "talent is an illusion" cannot be sustained. What remains instead is a narrower and more defensible claim. Small early differences compound through the paths of opportunity and feedback rather than through ability itself.

The cleanest case is the relative age effect in sports. Within the same selection year, a child born in January and a child born in December can differ by nearly a full year of development in the youth stage; that gap gets reflected in selection; the selected child gets better coaches and more playing time; and as a result actually becomes better, and is selected again. Talent did not produce the outcome — the selection system produced something that looks like talent. There is something to disclose honestly here. The existence of this phenomenon began with Canadian junior ice hockey and has been confirmed repeatedly across several sports, but in writing this article I was not able to verify the specific month-by-month distribution figures of the 1985 original paper from a primary source. So here I will cite only the direction and will not use the numbers.

The case where I did verify the numbers lies elsewhere. Jean Côté's Developmental Model of Sport Participation (DMSP) divides ages 6 to 12 into the sampling years, 13 to 15 into the specializing years, and 16 and up into the investment years. And Wall & Côté (2007), Physical Education and Sport Pedagogy, vol. 12, no. 1, pp. 77–87 tracked 12 elite youth hockey players and produced an interesting result. The model's simple prediction — that the dropouts would have sampled fewer sports and had less enjoyable play time — was not confirmed. The players who continued and the players who quit did not differ in breadth of sport experience or in play time.

There was exactly one item that differed significantly. The dropout group had begun unenjoyable, performance-only off-ice training (weights, running) at a far earlier age — a mean of 11.75 years versus 13.8 years, with a p value below 0.01. It is a study with a sample of 12, so it should not be loaded with too much weight, but the direction is clear. The problem is not early specialization in itself but when training with the enjoyment stripped out was introduced.

Optimizing Only the Part You Control

The action item that comes out of the literature above is not "practice more." As we just saw, that is substantially outside your control. What is within your control is four things.

Choose what you practice on. One reason the explanatory power of practice is low in the professional domains is that most people repeat what they already know how to do. List what you have built over the past six months, and count how many of those you did not know how to build when you started. If that ratio is low, your career accumulates but your skill stalls.

Reduce the feedback delay. The reason games and music have high explanatory power is that when you do it wrong you find out immediately. The ways to create that condition artificially in software are all well known already: tests, short deploy cycles, code review, and the habit of recording your own predictions and checking them later. Writing one line about "how is this choice going to hurt six months from now" when you make a design decision, and actually reading it six months later, is by itself enough to close the loop.

Design the opportunity side. What the relative age effect tells us is that placement, not ability, is what compounds. At the individual level this is the question of which team, which project, and which reviewer you stand in front of. The gap between two people of similar skill three years later usually comes not from how much they practiced but from what problems they had the opportunity to touch over those three years. At the organizational level, it is worth checking whether you have a structure in which the good projects are repeatedly assigned only to the people who are already good.

Do not introduce enjoyment-stripped training too early. This is what the hockey study suggests. An intensity that is not sustainable does not increase the total; it decreases it. If the amount of practice is substantially a product of disposition, then not exhausting that disposition is the core of managing the total.

The part the original essay gets right also meets us here. Defining "talent" as a developer by the single ability to enjoy hard algorithm problems is to ignore the distinction between domains we saw in the table above. Software engineering is not a narrow game with fixed rules; it is a broad domain in which the definition of performance shifts with context. In a domain like that, a single-axis concept of talent is inaccurate as a measuring instrument.

Closing — Admitting the Variables Outside Your Control Is What Makes the Ones Inside It Visible

To summarize:

Ericsson's 1993 study has a sample of ten per group, never claimed a 10,000-hour rule, and never denied the existence of talent.
According to the Macnamara meta-analysis (N of 11,135), the variance in performance explained by deliberate practice is 14 percent overall, in the twenties for games and music, and 5 percent and 1 percent for education and the professions. Measure practice volume objectively and it drops to 5 percent.
The double-blind replication attempt in 2019 failed to replicate the core finding of the original study.
In the study of 10,500 Swedish twins, the amount of practice was itself 40 percent to 70 percent heritable, and within monozygotic twin pairs the one who practiced more was not reliably better.
The honest claim, therefore, is not that "talent is an illusion" but that early differences compound through opportunity and feedback.

If this conclusion feels deflating, that is because we assume it has to be either "effort is enough" or "you have to be born with it." The data says it is neither. And that is not only bad news. A plan that mistakes an uncontrollable variable for a controllable one fails and leaves self-blame behind as well, whereas knowing the scope accurately lets you concentrate your force on the levers that remain. The remaining lever is not the total volume of practice but its structure — what to practice, how fast you can find out you were wrong, and which problems you will be standing in front of.