He can do it perfectly in rehearsal.
The sequence has five steps. He knows them in order. He can demonstrate them to the group and explain why each one matters.
Then the customer becomes angry.
Step two receives an answer nobody used in the training example. The customer's voice is louder than expected. A colleague is listening. There is a queue behind them.
The five-step skill is still in his memory.
But the situation is no longer asking for recitation.
It is asking whether the underlying ability can adapt.
That is the practical territory between skill and capacity.
The terminology is not universal
Any serious page with this title must begin by refusing false precision.
Research and policy fields do not use “skill” and “capacity” in one globally standardized way.
The OECD Learning Compass, for example, describes skills as the ability and capacity to carry out processes and use knowledge responsibly toward a goal. In that framework, the words overlap.
Other fields distinguish skills from abilities, competencies, capabilities or capacities differently.
The Institute therefore uses the distinction on this page as an operational convention for its own Knowledge Library.
It is a tool for making claims more precise, not an attempt to correct the vocabulary of every discipline.
Working definitions
A skill is an ability to perform a process, behavior or task with some degree of proficiency.
A capacity, in the Institute's usage, is the broader availability of relevant abilities across the range of conditions in which they need to function.
A skill can be one component of a capacity.
The capacity claim becomes stronger when performance is retained, transfers, adapts to variation, survives relevant pressure and integrates with judgment about when and how the skill should be used.
This means capacity is not automatically “better” than skill.
Sometimes a narrow, precise skill is exactly what matters.
Closed skills and open demands
Training research often distinguishes relatively closed tasks from more open ones.
A closed skill has stable performance requirements. A procedure can be specified with high precision and repeated under similar conditions.
An open skill must adapt to changing context.
Interpersonal leadership, negotiation, teaching and many forms of judgment are open because the next move depends partly on what the environment does.
Blume and colleagues' meta-analysis of training transfer found that several relationships involving motivation and work environment were stronger when training concerned open rather than closed skills.
That finding does not prove the Institute's capacity terminology.
It supports the underlying reason the distinction is useful.
As a task becomes more context-sensitive, evaluation must move beyond whether the learner can reproduce one trained performance.
Skill demonstration can be genuine and still narrow
Imagine three people.
One can use a statistical package to run a model when the dataset is clean and the method has already been selected.
One can conduct the same analysis and also recognize that the chosen model violates assumptions in a new dataset.
One can do both, explain the implications to non-specialists and decide when the analysis should not be used to support a consequential claim.
All three may possess statistical skills.
The broader performances involve additional knowledge, metacognition, judgment and communication.
Calling the third person “more skilled” may be perfectly ordinary language.
The Institute sometimes uses “capacity” when it needs to make the wider system of abilities explicit.
The purpose is analytical clarity.
Skill acquisition does not guarantee transfer
This is where learning science matters.
Soderstrom and Bjork show why successful performance during training cannot be treated as a direct measure of lasting learning.
Blume and colleagues show why training transfer depends on more than what happened inside the programme.
A skill can be acquired and fail to appear later because:
- it was learned under cues that are absent in the real context;
- the learner never gets a chance to use it;
- the work environment discourages it;
- time pressure changes strategy;
- the learner cannot recognize when the skill applies;
- performance depended on feedback that is no longer present;
- the real task contains variation the training never represented.
Capacity language is useful when the desired outcome includes overcoming those boundaries.
A skill can be highly complex
The distinction should not imply that skills are simple mechanical routines.
Surgery is skilled.
Writing is skilled.
Musical performance is skilled.
Mathematical proof is skilled.
Therapeutic listening can be described as skilled.
Skilled action can involve perception, memory, emotion, judgment, tacit knowledge and adaptation.
In ordinary scientific and professional language, “skill” already covers very complex phenomena.
The Institute's use of “capacity” is therefore not a hierarchy of sophistication.
It is a way to emphasize reliability across conditions when that dimension matters to the claim.
Practice predicts performance, but not all of it
Deliberate practice is one of the best-known ideas in expertise research.
Macnamara, Hambrick and Oswald's 2014 meta-analysis found that deliberate practice was associated with performance across games, music, sports, education and professions, but explained very different proportions of variance across those domains.
That is useful for two reasons.
First, skill development often requires structured practice.
Second, practice is not an omnipotent explanation for performance.
Domain, opportunity, individual differences, resources and other conditions matter.
Capacity language should never become a moralized claim that enough practice makes everyone equally capable.
From execution to selection
One difference between a narrow skill and broader capacity often appears before the action itself.
The person has to know which skill to use.
A good negotiator does not deploy the same technique in every conversation.
A good emotion regulator does not reappraise everything.
A good teacher does not give the same kind of feedback to every learner.
A good AI user does not verify every low-stakes sentence with the same intensity, nor accept every high-stakes answer because verification would take too long.
Selection requires a model of the situation.
That model depends on perception, knowledge, goals and consequence.
Capacity becomes visible in the fit between repertoire and context.
From technique to timing
A skill can be technically correct and badly timed.
A person may know how to ask an open question but ask it when the other person first needs a direct answer.
They may know how to challenge an assumption but do so in a way that humiliates a colleague and reduces future candor.
They may know a grounding technique but use it to avoid information that genuinely requires action.
Capacity includes timing because real environments contain more than the target behavior.
They contain relationships, goals, power and competing needs.
Pressure can expose the difference, but pressure should not become theatre
Organizations sometimes respond to transfer problems by making training dramatically stressful.
That is not automatically good practice.
The relevant pressure should resemble the conditions that genuinely change performance, and it should be introduced proportionately.
A safety procedure may need testing under realistic time demand.
A leadership behavior may need testing when disagreement is directed at the learner rather than someone else.
A boundary-setting skill may need practice with some interpersonal resistance.
But inducing distress for its own sake does not prove resilience.
Capacity assessment should be ecologically relevant, not performatively harsh.
Habit and skill are related but different
Some skilled components become automatic with repetition.
Automaticity can free attention for higher-level decisions.
Lally and colleagues' habit-formation study illustrates how repeated behavior in a stable context can gradually become more automatic, with substantial variation across individuals and behaviors.
But automaticity can also make behavior less sensitive to changing conditions.
This is why capacity often combines automatic components with flexible supervision.
A skilled driver does not consciously calculate every steering correction. They still need active judgment when weather, traffic or road conditions change.
The most mature performance is not necessarily the one with the most conscious effort or the one with the least.
It is the one that allocates attention well.
Skill inventories are useful but incomplete
Organizations often map skills because skills are easier to name and train.
Can use software X.
Can facilitate a meeting.
Can build a financial model.
Can write a prompt.
Can deliver feedback.
These inventories help with staffing and learning design.
They become incomplete when the organization assumes the listed skill will remain reliable under every context that matters.
A more capacity-oriented map might add:
Can recognize when software X should not be used.
Can facilitate disagreement when status is unequal.
Can challenge assumptions in the financial model when the business wants a particular answer.
Can detect when prompting is the wrong intervention because the underlying problem is missing evidence.
Can give feedback while personally implicated in the issue.
The skill has not disappeared.
The context has entered the description.
AI creates new skills and new capacity questions
Prompting is an obvious example.
A person can learn techniques for obtaining more useful model output. That is a skill.
But responsible use may require a wider capacity:
- knowing when the task is suitable for the model;
- preserving confidential information;
- evaluating sources;
- detecting hallucinated certainty;
- recognizing when domain expertise is insufficient;
- deciding when output must be independently verified;
- understanding whose interests or data may be absent;
- taking responsibility for the downstream decision.
AI literacy frameworks increasingly include ethical, human-centred and contextual dimensions for exactly this reason.
The operational skill matters.
So does the surrounding human system.
Measurement should not reward the vocabulary
If an organization says it has developed “capacity,” the evidence should exceed exposure and confidence.
A useful evaluation sequence might ask:
Can the person perform the component skill?
Can they retain it after delay?
Can they recognize when it applies?
Can they adapt it when superficial features change?
Can they choose among alternatives?
Can they use it under relevant pressure?
Can they explain the limits of the skill?
Can they refrain from using it when it would be inappropriate?
The last question matters.
Capacity includes not deploying a skill merely because one has it.
Capacity can also be collective
The distinction becomes even more important when no single person is expected to carry the whole capability.
A surgical team, newsroom, emergency service or AI-governance group may distribute skills across roles. One person recognizes a technical anomaly, another understands the legal consequence, another has authority to pause the process. The relevant capacity can therefore belong partly to the coordinated system rather than to one heroic individual.
This does not make individual skill irrelevant. Collective capacity depends on people actually possessing the component skills and on the interfaces between them working under pressure. Information has to travel. Authority has to be clear. Someone has to notice when the handoff failed.
For organizations, this is a reason not to turn every capacity problem into a training problem. Sometimes the missing intervention is role design, staffing, escalation, time or coordination. A person can be highly skilled and still work inside a system that cannot mobilize the skill when it matters.
What the evidence does not show
Current evidence does not show that:
- skill and capacity have one universally agreed scientific distinction;
- skills are simple while capacities are complex;
- capacity is always superior to skill;
- successful training performance proves transfer;
- more deliberate practice explains all expertise;
- open skills can be assessed with one standardized context;
- stress testing automatically reveals “true” capacity;
- automaticity is always desirable or always dangerous;
- people who fail to transfer a skill necessarily failed to learn it;
- every organization needs to rename its skill framework as a capacity framework.
The distinction earns its keep only when it improves design or measurement.
Institute operational convention
Throughout the Knowledge Library, “skill” should generally be used when the claim concerns performing a defined process or task.
“Capacity” should be preferred when the claim concerns broader reliable availability across conditions, especially where perception, strategy selection, regulation, judgment, adaptation or social context are integral to the outcome.
This convention prevents capacity from becoming a prestige word.
If a narrow skill is what is actually being trained, call it a skill.
If a programme claims capacity, it should be prepared to show more than successful rehearsal.
The angry customer again
The five steps were useful.
They gave the employee something to practice.
But when the customer answered unexpectedly, he needed more than memory of the sequence.
He needed to hear what changed.
He needed enough regulation to stay present.
He needed to choose which part of the framework still fit.
He needed to know when to abandon the script.
Later, he needed feedback about the moment he stopped listening and began defending himself.
The skill was real.
Capacity was the larger developmental claim.
That distinction is modest.
It is also enough to change how training is designed, how performance is measured and what institutions mean when they say a human ability has actually been developed.
References
- OECD. What Students Learn Matters: What does research say? Skills, knowledge, attitudes and values in the OECD Learning Compass 2030. 2020. https://www.oecd.org/en/publications/what-students-learn-matters_d86d4d9a-en/full-report/what-does-research-say_816c97c8.html
- OECD. Future of Education and Skills 2030/2040. https://www.oecd.org/en/about/projects/future-of-education-and-skills-2030.html
- Soderstrom NC, Bjork RA. Learning Versus Performance: An Integrative Review. Perspectives on Psychological Science. 2015;10(2):176-199. https://doi.org/10.1177/1745691615569000
- Blume BD, Ford JK, Baldwin TT, Huang JL. Transfer of Training: A Meta-Analytic Review. Journal of Management. 2010;36(4):1065-1105. https://doi.org/10.1177/0149206309352880
- Macnamara BN, Hambrick DZ, Oswald FL. Deliberate Practice and Performance in Music, Games, Sports, Education, and Professions: A Meta-Analysis. Psychological Science. 2014;25(8):1608-1618. https://doi.org/10.1177/0956797614535810
- Lally P, van Jaarsveld CHM, Potts HWW, Wardle J. How are habits formed: Modelling habit formation in the real world. European Journal of Social Psychology. 2010;40:998-1009. https://doi.org/10.1002/ejsp.674
- UNESCO. Artificial intelligence and the Futures of Learning: AI Competency Frameworks for Students and Teachers. Frameworks launched during Digital Learning Week 2024. https://www.unesco.org/en/digital-education/ai-future-learning