Your company can’t grow faster than its teams: how to develop the talent and capabilities to scale

The question isn’t just how many people a company has, but what the organization can do with them.

Key ideas

• Headcount, the volume of work a company can process and its organizational capability are three different things: growing the first doesn’t guarantee the other two.
• Before asking whom to hire, ask what the company needs to be able to do to execute its next strategy.
• Knowledge can exist inside an organization and never reach the people who decide: psychological safety and the design of coordination determine whether it flows.
• Many problems described as cultural are management capability gaps, and promotion systems can make them worse.
• Training isn’t capability building: training creates value when it changes behaviors that the organization lets people apply, practice and measure.
• AI is changing tasks before it changes jobs, and it opens a tension between today’s productivity and the development of tomorrow’s experts.

Growing headcount isn’t the same as scaling capability

It helps to separate three concepts that are often used interchangeably. Headcount is the number of people. Capacity, in the operational sense, is the volume of work the company can process. Organizational capability is what the company can do consistently enough to produce a result: closing complex sales, delivering projects with uniform quality, launching products or entering a new market. The three are related, but they don’t move together. A company can double its headcount without doubling its output, its speed, its quality or its ability to decide.

The reason is that an organizational capability doesn’t reside in any one person. It emerges from the combination of talent, knowledge, leadership, coordination, decision rights, systems, technology and learning. If one of those pieces doesn’t keep pace with growth, new people add cost and complexity before they add capability. That is why the thesis of this article, which we put forward as a strategic proposition rather than a proven law, is that a company scales when it manages to turn the talent and knowledge of its people into capabilities that can be reproduced, coordinated and renewed as complexity increases.

This changes the order of the questions. A company that needs to grow usually starts with the open position: whom does it need to hire? That question comes too early. The more useful sequence starts with strategy, continues with the capabilities that strategy requires, then the work and decisions that produce those capabilities, then the skills and knowledge needed, and only at the end reaches roles, teams, technology and the source of each capability. Before asking whom to hire, leadership needs to answer something else: what the company must be able to do to execute its next strategy.

A company can have talent and still work poorly

The paradox shows up in organizations of every size. There are companies with excellent professionals that coordinate poorly, that train a lot and develop few strategic capabilities, that hire managers and get more bureaucracy, or that decentralize decisions without distributing the information needed to make them. Individual talent is a necessary condition, but it isn’t the unit that produces an organization’s results.

Economist Gary Becker offers a useful distinction for thinking about which part of that talent can be bought and which part has to be built.¹ General human capital, such as programming, analyzing financial statements or negotiating, is valuable to many companies and can therefore be hired in the market. Firm-specific human capital (knowing the company’s customers, systems, exceptions and way of deciding) is valuable mainly inside that company, and it is almost always developed internally over time. A company that grows only by adding general human capital discovers that every new hire needs months to become fully productive, and that this period depends on people who are already overloaded.

The resource-based view of strategy adds a second layer: not every skill creates an advantage, and the most valuable capabilities are usually hard-to-replicate combinations of knowledge, relationships, experience, routines and coordination. We explore this in our article on sustainable expansion and internal capacity. Here, one practical consequence is enough: the capabilities that form part of the company’s strategic core deserve a deliberate decision about how they are built and protected, rather than being left to whoever happens to be hired at any given moment.

Knowledge that doesn’t scale

Almost every growing company has people without whom certain things don’t move: the founder who approves every major proposal, the sales director who knows the key accounts, the developer who understands the legacy system, the supervisor who knows what to do when the line fails. This key-person dependency has visible costs (bottlenecks, concentrated decisions, slow onboarding, continuity risk), but it isn’t solved by the usual recipe of documenting everything.

Philosopher Michael Polanyi put forward an idea that remains useful: we know more than we can tell.² Some knowledge is explicit and can be codified relatively easily in a procedure, a manual or software. Some is tacit: it depends on experience, context, practice and judgment, and it is better passed on by working alongside the person who has it than by reading about it. The distinction shouldn’t be romanticized either, because much knowledge presented as non-transferable can in fact be taught or systematized. The useful question isn’t whether to document, but what to do with each type of knowledge: document what is explicit, frequent and stable; teach through practice and supervision what requires judgment; build into software or AI what is codifiable and repetitive; and protect, in the hands of experts, what is strategic and scarce.

Two hypothetical examples show the same problem in different settings. In a growing consulting firm, the partners still concentrate sales, solution design, quality control and client relationships. The firm hires more analysts, but every deliverable goes back through the same three people, and delivery capacity grows far less than payroll. In an industrial plant that increases output, two supervisors know how to solve the quality and maintenance problems that appear in no manual. Adding operators raises volume until the first complex failure; at that point the plant depends on those two people again. In both cases, hiring more doesn’t remove the constraint. What removes it is transferring knowledge, distributing decisions, developing supervisors and standardizing only what should be standardized, without losing the judgment that makes the organization valuable.

Knowledge that exists but never reaches decision-makers

There is a less visible form of knowledge that doesn’t scale: knowledge that exists inside the organization but doesn’t flow. As a company grows, information becomes distributed. Leaders stop seeing problems directly, customers interact with different people, risks surface at levels far from leadership, and many opportunities are spotted far from management. If the people who see those signals don’t communicate them, the company decides with incomplete information without knowing it.

Amy Edmondson, a professor at Harvard Business School, has studied this phenomenon for decades. In one of her early studies, on medication errors in units at two hospitals, she found a counterintuitive result: the units with better leadership and relationship-quality indicators recorded more detected errors, not fewer. Her interpretation was that people in those units were more willing to report them.³ The fact that no one reports problems doesn’t mean they don’t exist; it may mean the conditions to raise them don’t exist. This organizational silence is especially dangerous in growing companies, because leadership reads the absence of warnings as a sign that everything is working.

Edmondson later developed the concept of psychological safety: a shared belief within a team that it is safe to take interpersonal risks, such as asking questions, admitting mistakes, asking for help or disagreeing. In a field study of 51 teams at a manufacturing company, she found that psychological safety was associated with learning behaviors, and that those behaviors helped explain team performance.⁴ A later meta-analysis, combining 136 samples with more than 22,000 people and nearly 5,000 groups, consolidated the evidence on its antecedents and consequences, although the findings are mostly associations rather than demonstrated causal relationships.⁵

The concept is often misunderstood. Psychological safety doesn’t mean the absence of demands, accountability or consequences, nor that everyone agrees, nor that difficult conversations are avoided. Edmondson herself distinguishes it from comfort: learning happens when psychological safety is combined with high performance standards.⁶ Her prescription for leaders involves three practices: setting the stage, by explaining the uncertainty and interdependence of the work; inviting participation, with real mechanisms for questions and perspectives to surface; and responding productively, so that whoever brings a problem wants to do it again. None of this guarantees productivity, innovation or good decisions; technical capability, resources, clear goals, coordination and execution are also needed. But without the conditions for knowledge to flow, none of those other pieces works with the information it needs.

When no one reports problems, it doesn’t mean there aren’t any. It may mean there are no conditions to raise them.

How to keep working as a team once the company is no longer small

A small team has advantages nobody designed: trust, shared context, fast communication, informal coordination and the ability to adapt without asking permission. As it grows, functions, specialties, hierarchical levels and dependencies between areas appear. Each makes sense on its own, but together they fragment information and push more and more decisions across interfaces. In our article on operations and scalability we looked at how coordination channels grow as headcount increases. Here we are interested in the other side of the problem: what an organization does to avoid losing its ability to adapt.

In Team of Teams, Stanley McChrystal and his co-authors describe the transformation of a military task force that, despite having highly efficient elite teams, couldn’t keep up with a more decentralized and faster adversary.⁷ Their diagnosis was that the organization had been designed for efficiency, through hierarchy, specialization and centralized control, in an environment that demanded adaptability. The response they describe combined two ideas. The first, which they call shared consciousness, gave every team enough understanding of the context: goals, dependencies, constraints and the consequences of their decisions. One of its tools was a daily operations and intelligence briefing open to a very large number of participants. The second, empowered execution, moved decisions to the people who had the information to make them.

The relationship between the two ideas is what makes the model useful for a company. Autonomy without context produces misalignment: each team decides well based on its own information and poorly based on the whole. Context without autonomy produces slowness: everyone understands what is happening, but every decision still travels to the top. In this view, leadership stops being the job of deciding everything and becomes the job of creating the conditions for others to decide well: distributing information, clarifying purpose, setting boundaries, building trust and removing obstacles.

The book should be read with care. It is its protagonists’ account of a military experience, not experimental evidence, and it doesn’t show that hierarchies no longer work or that every company should be organized as a network. The right structure depends on the type of work, risk, regulation and degree of interdependence. What it does offer is a question every growing company should ask: how much of the information its teams need to decide actually reaches them, and how many decisions still travel upward only because nobody delegated them. The same tension repeats in other dimensions: specialization increases efficiency and creates silos; processes enable scale and reduce adaptability. None of these tensions is resolved once and for all. They are managed.

The management capability gap

When a company grows, the sequence is usually similar: more sales, more people, more specialization, more teams, more interdependencies and, therefore, more intermediate decisions. Those decisions fall to managers. If those managers haven’t developed the ability to delegate, prioritize, coordinate, give feedback, manage performance, resolve conflict, allocate resources and design other people’s work, the result looks a lot like a cultural problem: disengaged teams, poor communication, lack of commitment. Often it is something else: a management capability gap.

Part of that gap is manufactured by the promotion system. The most common practice is to promote the best specialist: the top salesperson ends up leading the sales team, the best engineer ends up managing the engineers. Alan Benson, Danielle Li and Kelly Shue studied this practice using data on salespeople at 214 firms and more than 1,500 promotions into supervisory roles.⁸ They found that companies prioritized performance in the current role over other characteristics that better predicted performance as a manager. According to the NBER summary, doubling a salesperson’s sales increased their relative probability of promotion by 14.3%, and doubling a new manager’s pre-promotion sales was associated with a 7.5% decline in the sales of each of their subordinates. Those with collaborative work experience, by contrast, did better as managers. The result comes from one specific setting, sales, but it illustrates precisely that technical performance and management capability are different capabilities.

That doesn’t mean specialists can’t become good managers. It means management should be treated as a capability that requires its own selection, development, practice, feedback and evaluation. It also calls for a review of career architecture: if the only way to grow professionally is to manage people, the company loses some of its best specialists in the attempt to turn them into bosses, and gains managers who never chose to be one. Parallel technical tracks and project or expert leadership roles are alternatives each company should assess in its own context; there is no universal model. Diagnosis, for its part, starts by observing behavior rather than measuring climate: how long decisions take, how many are escalated to leadership, how many people each manager leads, how the teams of newly promoted managers perform, and whether anyone prepared them for the role.

External signals suggest the pressure on managers is high. According to Gallup’s State of the Global Workplace 2026 report, global employee engagement fell to 20% in 2025, and manager engagement dropped to 22%, nine points lower than in 2022.¹⁷ These are survey data and don’t allow the decline to be attributed to any particular cause. But they raise a relevant question for any growing company: what happens when organizational complexity grows faster than management capability.

The good news is that this capability can be built, and there is evidence of it in Latin America. Miriam Bruhn, Dean Karlan and Antoinette Schoar analyzed a program run by the government of the state of Puebla, Mexico, that randomly assigned management consulting services to small and medium-sized enterprises.⁹ A year later, the firms that received the support showed improvements in productivity and return on assets relative to the control group. Five years later, they had grown their workforce by around 57% and their total wage bill by 72%, while the control firms stayed roughly the same size. It is an experiment with a limited sample in a specific context, and it should be read without triumphalism. What it shows isn’t that any consulting works, but that management practices aren’t a fixed trait of a company: they can be learned, and their effect can compound over time.

Training isn’t the same as building capability

The reflexive response to a capability gap is training. A company can offer courses, increase its team’s individual knowledge and still fail to change behavior, improve coordination or move results. The problem isn’t training itself, but the question from which it is decided. The traditional question is which courses to offer. A better one is which skills people need. The most useful is what the organization must be able to do that it can’t do today. Only the third connects learning to strategy.

From that question, training starts with the business outcome it needs to enable and works backward: which capability produces it, which concrete behaviors it requires, which skills and knowledge support them, which learning experience develops them and, above all, which practice, application and feedback turn them into habit. The control questions are simple and almost never answered: which behavior should change after this training, where will it be practiced, who will reinforce it, how will it be measured, and which organizational conditions allow it to be applied. A manager who learns to delegate in a workshop and returns to a structure where every decision requires their signature hasn’t developed a capability; they’ve collected a certificate.

Chris Argyris and Donald Schön contributed a distinction that reveals a deeper limit.¹⁰ Single-loop learning corrects actions without questioning the assumptions that produce them; double-loop learning revisits those assumptions, rules or goals. Many companies train their people to solve, over and over, problems created by a poorly designed system: they train the service team to handle complaints that exist because the delivery process fails. Sometimes the right question isn’t how to train better, but why the problem still exists.

The pressure to update skills is real. In the World Economic Forum’s Future of Jobs Report 2025, based on more than 1,000 employers representing more than 14 million workers across 55 economies, companies expect that, on average, 39% of current skills will be transformed or become outdated between 2025 and 2030. Sixty-three percent identify skills gaps as the main barrier to their transformation, and 85% plan to prioritize upskilling their workforce.¹⁸ These are expectations declared by employers, not measurements of skills that have actually disappeared, and the 39% doesn’t mean that share of people will lose their jobs. What they do indicate is that the ability to learn is becoming an organizational capability in its own right.

Artificial intelligence changes tasks before it changes jobs

Labor economics offers a more precise way to think about the impact of technology than asking which jobs will disappear. David Autor, Frank Levy and Richard Murnane showed that jobs aren’t indivisible units but bundles of tasks, and that technology can substitute for some of those tasks and complement others.¹¹ The relevant sequence is a cascade: AI changes certain tasks; changing tasks alters the skills required; changing skills alters roles; changing roles alters teams; and changing teams can alter structure. None of these transitions is automatic, and at each one the company has decisions to make.

The available evidence shows effects that seem contradictory and that, read together, are more revealing than any of them alone. Inside a company, AI can accelerate the learning of less experienced workers. Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied 5,172 customer-support agents and found that access to a generative assistant increased issues resolved per hour by 15% on average, with much larger gains among workers with less experience and lower initial performance.¹² The authors interpret that the tool spread the practices of the most skilled workers to newer ones. It is a specific implementation, not a universal benchmark, but it suggests a relevant possibility: AI as infrastructure for distributing know-how that used to be concentrated in experts. That reading is Advisent®’s interpretation, not a stated conclusion of the study.

In the labor market, by contrast, the entry door for young workers seems to be narrowing. The International Labour Organization estimates that one in four workers worldwide is in an occupation with some degree of exposure to generative AI, and that the highest-exposure category accounts for 3.3% of global employment; the ILO itself notes that exposure doesn’t equal automation or job loss.¹⁹ Using ADP payroll data through June 2026, Brynjolfsson, Bharat Chandar and Ruyu Chen found that employment of 22- to 25-year-olds in the occupations most exposed to AI was about 19% below where it would be had it kept pace with peers in less-exposed occupations, up from the 15% they had observed a year earlier.²⁰ The gap worked mainly through less hiring rather than more separations, and it didn’t appear among experienced workers. The authors stress that these are descriptive patterns, not causal estimates: the gap shrinks when education is taken into account, and some trends appear before the spread of generative AI. The study doesn’t show that AI destroyed 19% of youth employment. It is an early signal no company should ignore.

Productivity today, experts tomorrow

The tension between those findings has a direct organizational consequence. Many professions develop their experts through entry-level tasks: research, drafting, data analysis, coding, document review, customer service or diagnosing simple problems. These are precisely the tasks AI can accelerate the most. The benefit is productivity. The risk is fewer opportunities to learn through practice. If we automate the tasks through which junior professionals used to learn, how will we develop the senior experts of the future?

There is evidence the risk isn’t theoretical, although it comes from other fields. Matt Beane, of the University of California, Santa Barbara, spent more than two years studying the introduction of robotic surgery in hospitals. He found that, because the lead surgeon could operate without assistance, residents practiced far less (ten to twenty times less, by his estimate), and that those who managed to build the skill did so through informal paths he called shadow learning, with little expert supervision.¹³ In a different setting, Hamsa Bastani and her co-authors ran an experiment with nearly a thousand high school math students in Turkey. Those who practiced with unrestricted GPT-4 scored 48% better during practice but 17% worse on the subsequent exam without the tool than students who had never used it. A version designed as a tutor, with safeguards that required students to reason, improved practice by 127% and largely eliminated the negative effect.¹⁴

None of these studies concerns companies, and they don’t support the conclusion that AI will prevent the development of experts. But they show something that does generalize: the same technology can accelerate learning or erode it depending on how its use is designed. Decades ago, James March described the tension between exploiting what an organization already knows how to do and exploring what it doesn’t yet know: too much exploitation produces efficiency today and rigidity tomorrow.¹⁵ Using AI only to produce faster is a form of exploitation. Using it to develop people as well is an investment in future capability.

To think about that design, we propose a set of principles we call AI-Enabled Apprenticeship by Advisent®. It isn’t a validated methodology but a design proposal with four conditions: that whoever works with AI understands the reasoning and assumptions behind the output, not just the output; that there is deliberate practice without the tool, to build independent judgment; that human experts give feedback on AI-assisted work; and that autonomy grows as competence is demonstrated, with final accountability always assigned to a person. The question that organizes it all is how to use AI to improve today’s productivity without eroding the capacity to develop tomorrow’s talent.

What AI can accelerate

• Research and summarizing
• Drafting
• Data analysis
• Coding
• Answering frequent questions

What should be preserved as practice

• Understanding the reasoning behind the output
• Solving problems without the tool
• Getting feedback from experts
• Gaining autonomy progressively
• Being accountable for the final result

+48% / −17%Practice with unrestricted GPT-4: better during practice, worse on the exam without the tool

10 to 20 timesless practice for robotic-surgery residents, according to Beane

+15%issues resolved per hour with an AI assistant, with larger gains among the least experienced

Conceptual representation by Advisent®. The figures come from different contexts (high school education, medicine and customer support) and aren’t comparable with one another. Sources: Bastani et al. (2025), Beane (2019), Brynjolfsson, Li and Raymond (2025).

Integrating AI into teams isn’t an automatic sum either. Using it as an assistant, a knowledge interface, a tutor, a workflow participant or decision support are different things, and in no case does decision support replace the responsibility of deciding. Each use raises different questions about who verifies, who is accountable and what the team learns along the way. We look at this in more detail in our article on artificial intelligence and competitive advantage.

Hiring is only one of five possible responses

When a capability gap appears, the first reflex is to open a search, and in 2026 the second is to ask whether AI can solve it. In Lead the Work, published in 2015, before generative AI, John Boudreau, Ravin Jesuthasan and David Creelman proposed a distinction that is especially useful today: work, jobs and employees are different things.¹⁶ Work is the activity needed to produce an outcome. A job is one particular way of bundling tasks and responsibilities. An employee is one of the possible sources for getting that work done. Separating them makes it possible to ask first what outcome is needed, what work produces it, what capabilities it requires and what the best way to access them is, which may or may not be a hire.

Today, work can be distributed across employees, internal specialists, freelancers, consultants, partners, vendors, software, automation, AI systems and combinations of people and AI. To structure that decision, we propose The Capability Sourcing Model by Advisent®. It starts from a scheme widely used in talent management, build, buy, borrow, and adds two options: automating or augmenting with technology, and redesigning the work. The extension is Advisent®’s; the underlying scheme isn’t. The prior step is to diagnose the nature of the gap, because a shortage of people is different from a shortage of skills, knowledge, coordination, management capability, clear decisions, technology or reasonable work design. A coordination gap isn’t solved by hiring, and a management gap isn’t solved with software.

Five responses to a capability gap. They can be combined; the prior step is diagnosing what kind of gap it is.

Response	When it makes sense	What risk it creates
Build	The capability is strategic, will be needed for a long time, and company-specific knowledge creates value	Takes time and requires someone to teach and room to practice
Buy	Stable capacity is needed, talent is available in the market and demand justifies the cost	Fixed cost and slow ramp-up when company-specific knowledge matters
Borrow	The need is temporary, niche expertise is required or time is critical	Dependency and less internal learning if the capability is strategic
Automate or augment	The task is technically feasible, suitable data exist and the economics work	New dependencies, oversight costs and loss of formative practice
Redesign	The problem lies in how the work is organized, not in how many people do it	Requires coordinating changes across areas and managing resistance

Conceptual model by Advisent®, not a validated instrument. It extends the build, buy, borrow scheme widely used in talent management with the options to automate and redesign. Source: Advisent®.

The choice depends on observable variables: strategic importance, frequency, urgency, cost, talent availability, risk, quality, internal learning, technological feasibility and flexibility. There is also a criterion that tends to be left out: what capability the company will have three years from now as a result of this decision. Outsourcing or automating can cut costs or add speed today and, at the same time, create dependency, externalize strategic knowledge or eliminate the tasks through which the next generations were trained. The redesign option deserves special attention, because sometimes the question isn’t whom to hire, but whether the company needs more people or a different way of organizing the work.

The capability stack: reversing the order of the questions

The traditional way of thinking about talent follows a sequence: strategy, org chart, jobs, people and, finally, training. Its problem is that the org chart appears before anyone knows which capabilities the strategy requires, and training arrives at the end to make up for what the design didn’t solve. We propose reversing that order with The Organizational Capability Stack by Advisent®, a seven-layer conceptual framework rather than a validated instrument.

The usual sequence

1. Strategy
2. Org chart
3. Jobs
4. People
5. Training

The org chart comes before capabilities, and training compensates at the end.

The Organizational Capability Stack

1. 01StrategyWhat do we want to achieve?
2. 02Required capabilitiesWhat must we be able to do consistently?
3. 03Work and decisionsWhich activities and decisions produce them, and who decides?
4. 04Skills and knowledgeWhat do we need to know and know how to do?
5. 05Capability sourcesBuild, buy, borrow, automate or redesign?
6. 06Systems and technologyWhich processes, data and tools enable the work?
7. 07Learning and renewalHow are capabilities updated when strategy changes?

↺ What is learned changes what the strategy can aim for

Conceptual framework by Advisent®, neither a linear process nor a validated instrument. Source: Advisent®.

Each layer answers a question. Strategy defines what the company wants to achieve. Required capabilities define what it must be able to do consistently to get there. Work and decisions describe which activities and decisions produce those capabilities, and who has the right to make them. Skills and knowledge define what people need to know and know how to do. Capability sources determine how each one is obtained. Systems and technology enable that work. And learning and renewal ensure that capabilities are updated when strategy changes. It isn’t a linear process: what the company learns in the lower layers changes what the strategy can aim for. That last layer connects with what the strategy literature calls dynamic capabilities, the ability to renew and reconfigure resources when the environment changes, which we discuss in our article on internationalization and competitiveness. It is the difference between optimizing the capabilities the company needs today and developing the ones it will need in its next stage.

Knowing whether a company is building capabilities requires measuring more than training hours, completion rates or satisfaction surveys. We suggest reading four levels together, because no single metric is enough: cutting onboarding time is useless if it erodes quality, productivity per employee can reflect efficiency or overload, and low turnover doesn’t necessarily mean a healthy organization.

If all of this had to be translated into a single decision for the next day, it would be this: before approving the next open position, ask the leadership team to answer in writing which concrete capability the next stage requires that doesn’t exist today, whether the gap is about people, management, decisions, knowledge or work design, and whether hiring is really the best response compared with building, borrowing, automating or redesigning. And, in parallel, review who was promoted to manager in the past year and what preparation they received for the role.

From how many people to which capabilities

A company doesn’t scale by simply adding more people. It scales when it manages to turn talent, knowledge, technology, leadership and coordination into capabilities that keep working as complexity increases. That requires changing the question used to plan growth: not whether the company has enough people, but whether it has the capabilities its next stage demands.

In a context where AI is changing tasks and skills faster than talent structures change, the advantage won’t depend only on having the right capabilities today. It will depend on the ability to learn, reconfigure and renew, and to do so without sacrificing the development of the people who will have to sustain the company ten years from now.

Which capabilities do we need to build, which to buy, which to source from third parties, which to augment with technology, and what work should we stop doing or redesign?

Hiring adds people. Building capabilities is what makes growth possible.

Frequently asked questions about talent and organizational capabilities

What are organizational capabilities?

They are what a company can do consistently enough to produce a result: closing complex sales, delivering projects with uniform quality, launching products or entering a new market. They don’t reside in any one person but in the combination of talent, knowledge, leadership, coordination, decision rights, systems, technology and learning.

Why doesn’t hiring more people necessarily increase productivity?

Because every new person also adds coordination, interfaces and intermediate decisions, and because company-specific knowledge takes time to transfer. If management, decision rights, systems and knowledge transfer don’t keep pace with growth, headcount grows before capability does. That is why it pays to ask first what the company must be able to do, and only then whom to hire.

What is the difference between training and capability building?

Training is an event or a program; capability building means the organization can do something it couldn’t do before. Training contributes to capability when it starts from a business outcome, defines which behaviors must change, includes practice and feedback, and takes place in an organization that lets people use what they learned. Course hours and completion rates don’t measure capability.

Why isn’t the best specialist always the best manager?

Because technical performance and management capability are different capabilities. A study by Benson, Li and Shue using data on salespeople at 214 firms found that companies promoted their best salespeople even though other characteristics better predicted managerial performance, and that those managers’ teams sold less. Specialists can become good managers, but the role requires its own selection, development and practice.

What is psychological safety, and what isn’t it?

It is a shared belief within a team that it is safe to take interpersonal risks, such as asking questions, admitting mistakes, asking for help or disagreeing. It doesn’t mean the absence of demands, accountability or consequences, or that everyone must agree. Amy Edmondson’s research associates it with team learning, especially when combined with high performance standards, but on its own it doesn’t guarantee productivity or good decisions.

How does AI affect the development of junior professionals?

In two ways that can coexist. Inside companies, AI can speed up the learning curve of less experienced workers, as a customer-support study showed. But it can also reduce the practice through which experts are formed, and according to a descriptive Stanford analysis with data through June 2026, employment of 22- to 25-year-olds in the most exposed occupations grew less than that of their peers, mainly through less hiring. The outcome depends on how the use of the technology is designed.

When should a company hire, outsource, automate or redesign the work?

First, diagnose whether the gap is about people, skills, knowledge, coordination, management, decisions, technology or work design. Building internally makes sense when the capability is strategic and long-lasting; hiring, when stable capacity is needed and talent is available; borrowing, when the need is temporary or highly specialized; automating, when the task is feasible and the economics work; and redesigning, when the problem lies in how the work is organized. The options can be combined.

How can you measure whether a company is building capabilities?

Read four levels together: learning (proficiency demonstrated in real situations), behavior (observable changes in practice), operations (time to proficiency, time to decision, quality consistency, key-person dependency) and business (margin, delivery capacity, customer retention). No single metric is enough, and all of them should be compared against a baseline.

Sources and further reading

Academic foundations and books

1. Becker, G. S. (1993). Human capital: A theoretical and empirical analysis, with special reference to education (3rd ed.). University of Chicago Press. First edition: 1964.
2. Polanyi, M. (1966). The tacit dimension. Doubleday.
3. Edmondson, A. C. (1996). Learning from mistakes is easier said than done: Group and organizational influences on the detection and correction of human error. The Journal of Applied Behavioral Science, 32(1), 5–28. doi.org/10.1177/0021886396321001
4. Edmondson, A. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly, 44(2), 350–383. doi.org/10.2307/2666999
5. Frazier, M. L., Fainshmidt, S., Klinger, R. L., Pezeshkan, A., & Vracheva, V. (2017). Psychological safety: A meta-analytic review and extension. Personnel Psychology, 70(1), 113–165. doi.org/10.1111/peps.12183
6. Edmondson, A. C. (2019). The fearless organization: Creating psychological safety in the workplace for learning, innovation, and growth. Wiley.
7. McChrystal, S., Collins, T., Silverman, D., & Fussell, C. (2015). Team of teams: New rules of engagement for a complex world. Portfolio.
8. Benson, A., Li, D., & Shue, K. (2019). Promotions and the Peter principle. The Quarterly Journal of Economics, 134(4), 2085–2134. Summary: NBER Digest
9. Bruhn, M., Karlan, D., & Schoar, A. (2018). The impact of consulting services on small and medium enterprises: Evidence from a randomized trial in Mexico. Journal of Political Economy, 126(2), 635–687. Summary: Kellogg Insight
10. Argyris, C., & Schön, D. A. (1978). Organizational learning: A theory of action perspective. Addison-Wesley.
11. Autor, D. H., Levy, F., & Murnane, R. J. (2003). The skill content of recent technological change: An empirical exploration. The Quarterly Journal of Economics, 118(4), 1279–1333.
12. Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. doi.org/10.1093/qje/qjae044
13. Beane, M. (2019). Shadow learning: Building robotic surgical skill when approved means fail. Administrative Science Quarterly. Summary: UC Santa Barbara
14. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. doi.org/10.1073/pnas.2422633122
15. March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2(1), 71–87.
16. Boudreau, J. W., Jesuthasan, R., & Creelman, D. (2015). Lead the work: Navigating a world beyond employment. Wiley.

Reports and institutional research

17. Gallup. (2026). State of the global workplace 2026. gallup.com
18. World Economic Forum. (2025). The future of jobs report 2025. weforum.org
19. Gmyrek, P., Berg, J., et al. (2025). Generative AI and jobs: A refined global index of occupational exposure (ILO Working Paper 140). International Labour Organization. ilo.org
20. Brynjolfsson, E., Chandar, B., & Chen, R. (2026, August revision). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford Digital Economy Lab. digitaleconomy.stanford.edu

The Organizational Capability Stack by Advisent®, The Capability Sourcing Model by Advisent® and AI-Enabled Apprenticeship by Advisent® are proprietary conceptual developments, not validated instruments. The Capability Sourcing Model extends the build, buy, borrow scheme widely used in talent management. The concepts of psychological safety and its leadership practices (Edmondson), shared consciousness and empowered execution (McChrystal and co-authors), the distinction between work, jobs and employees (Boudreau, Jesuthasan and Creelman), general and firm-specific human capital (Becker), tacit knowledge (Polanyi), single- and double-loop learning (Argyris and Schön), exploration and exploitation (March) and the task-based approach (Autor, Levy and Murnane) belong to their authors. Reading AI as knowledge-distribution infrastructure is Advisent®’s interpretation. The surveys cited reflect declared perceptions and expectations; the Stanford study is descriptive and doesn’t establish causality; the consulting-firm and industrial-plant examples are hypothetical.

Last updated: October 5, 2026