What is deliberate practice?

Deliberate practice is effortful work at the edge of current ability, directed at a specific weakness, with immediate feedback and repeated correction. It is not repetition, not time served, and not enjoyable. People avoid it by practising what they can already do, which feels productive and is not.

This research leans on the mechanism constantly, in the argument that removing repetitions removes the capability they built. The mechanism deserves stating properly, including the part where the strong version of it did not survive re-analysis.

The four conditions

Ericsson, Krampe and Tesch-Romer's 1993 study of musicians made accumulated deliberate practice central to theories of expert performance, and specified what counts. Graded entry. Four things have to hold together:

The task sits at the edge of current ability, so that failure is frequent. It is aimed at a specific weakness rather than at the whole activity. There is immediate feedback on what went wrong. And the attempt is repeated with correction rather than moved past.

Remove any one and the rest stops working. Practising a whole piece end to end is repetition. Practising the bar you keep fumbling, slowly, with someone telling you what your hand is doing, is deliberate practice. Ericsson and Pool's Peak (2016) is the accessible statement.

The claim that did not hold

The version that reached popular culture, that practice explains most of the difference between performers, is stronger than the evidence supports.

Macnamara, Hambrick and Oswald pooled 88 studies and 157 effect sizes. Accumulated deliberate practice explained 12 per cent of the variance in performance, 95% CI [9%, 15%], leaving 88 per cent unexplained. By domain: 26 per cent for games, 21 per cent for music, 18 per cent for sports, 4 per cent for education. Graded entry. Macnamara and Maitra later re-ran the 1993 study directly and reached a compatible conclusion. Graded entry.

Practice is necessary and it is not sufficient, and the defensible position sits between Ericsson and the meta-analysis rather than on either. The case for protecting repetitions does not require that repetitions explain everything, only that removing them removes something.

The figure everyone quotes, and what it actually covers

There is a fifth domain in that list, and it gets quoted more than the other four combined: under 1 per cent for the professions. It is used as proof that practice does not build professional expertise. Read at source, it will not carry that.

The professions estimate rests on 7 effect sizes. It is not statistically significant, at p = .62, so the honest reading is that this meta-analysis did not measure the professions rather than that it found nothing there. And the occupations sampled were computer programming, military aircraft piloting, soccer refereeing and insurance selling.

Which is to say the number is not evidence about law, medicine, consulting, analysis or any of the work this research is usually discussing. The authors themselves suggest deliberate practice may simply be less well defined in these domains. Anyone citing the figure against professional practice is citing four unrelated occupations and a null result.

Kind and wicked environments

David Epstein's Range (2019) supplies the sharper objection. The domains that produced the deliberate-practice literature, chess and classical music, are unusually kind learning environments: the rules are stable, the goal is fixed, and feedback is immediate and unambiguous. Most professional work is wicked. The rules shift, the goal is contested, and feedback arrives late, filtered or not at all.

In wicked domains Epstein argues that breadth, delayed specialisation and analogical thinking outperform early specialisation. He is arguing a case and selects for it, and the underlying distinction is sound.

It is also the same boundary Kahneman and Klein drew for intuition: recognitional expertise is trustworthy where the environment is predictable and the individual had the chance to learn its regularities. Graded entry. Kind environments build reliable expertise through practice. Wicked ones build confidence without necessarily building accuracy. Hence the experienced professional who is certain and wrong, a combination that takes years to produce.

The AI question, stated precisely

The mechanism above says nothing about AI. What it does is make the AI question exact: which of those difficult, feedback-rich repetitions disappear when the machine makes the first attempt?

Bastani and colleagues have the closest thing to a direct test. Students with unrestricted GPT-4 scored 17 per cent below a control group once access was withdrawn, while a hints-only tutor that preserved the attempt largely removed that harm. Graded entry. The tutor kept the first condition, work at the edge of ability, and the unrestricted tool removed it.

Sankaranarayanan's scaffolded condition did the same thing in adult programming, with 39 per cent failure against 77 for unrestricted use once the tool was gone. Graded entry.

Both interfaces preserved the attempt and the correction. Neither preserved the difficulty by accident: it was designed in.

What this does not settle

How much of professional expertise is built by deliberate practice at all is contested, and this page does not resolve it. The mechanism is real, its explanatory share is smaller than the popular version claims, and its applicability outside kind environments is unproven. That is as far as the evidence reaches in either direction, which is why the professions figure above is not repeated here as a finding.

Nor does anything here establish a dose. Nobody knows how many unaided attempts a junior lawyer or analyst needs, or over what period, and the school and laboratory results should not be converted into a workplace prescription.

And Ericsson's own framework requires a coach. Most professional deliberate practice has never had one, which means the feedback condition was already weakly met before AI arrived.

Key sources

• Ericsson, K. A., Krampe, R. T. and Tesch-Romer, C. (1993). The Role of Deliberate Practice in the Acquisition of Expert Performance. Psychological Review, 100(3). Graded entry.
• Macnamara, B. N., Hambrick, D. Z. and Oswald, F. L. (2014). Deliberate Practice and Performance in Music, Games, Sports, Education, and Professions: A Meta-Analysis. Psychological Science, 25(8), 1608-1618. Graded entry.
• Macnamara, B. N. and Maitra, M. (2019). The role of deliberate practice in expert performance: revisiting Ericsson, Krampe and Tesch-Romer (1993). Royal Society Open Science, 6. Graded entry.
• Ericsson, K. A. and Pool, R. (2016). Peak: Secrets from the New Science of Expertise. Houghton Mifflin Harcourt. In the essential works.
• Epstein, D. (2019). Range: Why Generalists Triumph in a Specialized World. Riverhead. In the essential works.
• Bastani, H. et al. (2025). Generative AI Without Guardrails Can Harm Learning. Graded entry.

Related SuperSkills research

The mechanisms alongside it: productive struggle, retrieval practice and desirable difficulty. What removing the reps accumulates as: capability debt, the missed reps and the missing rungs. On the faculty it builds, judgement. On how fast it goes when practice stops, how fast do skills decay. For the person at the start, should juniors use AI at all.

About this research

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. Deliberate practice is Ericsson's concept and the critique is Macnamara and Maitra's. Neither is a SuperSkills coinage, and both are on the page because the argument is stronger with the challenge included.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.