The Value of Work in the Age of AI

Follow-up to the keynote “Becoming Human Again”, Futuromundo EDU, June 2026

This summary is intended for everyone who attended the keynote, and for everyone engaging with the question of what the value of work will be based on in the future. It does not reproduce the argument of the talk. Instead, it places it in context and makes accessible the evidence on which it rests. Where the state of research is inconclusive, that is stated explicitly.

The starting thesis

For two centuries, the value of work was formed through a clear chain: education produces knowledge, knowledge is consolidated through experience, and experience is rewarded through seniority. This chain is the silent foundation of almost all compensation systems, career paths and organisational charts with which companies work today.

The thesis of the keynote is that this chain no longer holds because its first link is losing its scarcity. And most organisations respond by making the old model of value creation more efficient instead of building the new one.

What the research says

On the devaluation of professional knowledge

The idea that the half-life of knowledge is declining is widespread and is often supported by striking figures. The robustness of those figures varies.

The most frequently cited framework comes from innovation research and differentiates by type of knowledge: school knowledge about twenty years, university knowledge about ten years, professional knowledge about five years, technical knowledge about three years, and IT knowledge one to two years (Vahs/Brem, Innovationsmanagement, 2013).

An important counterposition is that some education researchers consider the term misleading. Existing knowledge is not decaying faster; rather, the amount of newly produced knowledge is growing. The practical effect for individuals is similar, but the conclusion is different. The issue is less one of decay than of the ability to remain connected to current knowledge.

For the keynote’s argument, the distinction is secondary. What matters is not how quickly knowledge becomes obsolete, but that it is losing its exclusivity. A knowledge advantage that anyone can access at any time is worthless as a source of value, regardless of how long the knowledge remains valid.

On the success rate of transformations

The frequently cited claim that seventy percent of all transformations fail should be used with caution. It can be traced back to an estimate by John Kotter from 1993 that was not itself based on empirical research. Later citations often refer back to one another in a circle.

The major consultancies’ own surveys provide a more robust basis:

• McKinsey, “Losing from Day One” (2021): Fewer than one third of surveyed companies report transformations that both improved performance and sustained that improvement over time. The rate has remained stable across fifteen years of research.
• BCG, “Flipping the Odds of Digital Transformation Success” (2020): Thirty percent fully successful, forty-four percent creating some value, and twenty-six percent producing no significant outcome.

The core statement therefore remains valid, but becomes more precise: seven out of ten initiatives do not fail completely. But fewer than one third achieve what they set out to do and sustain it over time.

On collective intelligence

The scientific basis for the assumption that group performance is a distinct, measurable quantity goes back to Woolley and colleagues (Science, 2010). Their finding: there is a factor c that predicts a group’s performance across different types of tasks, and this factor correlates only weakly with the average individual intelligence of its members.

Three variables proved particularly relevant: members’ social perceptiveness, an even distribution of conversational turn-taking, and the proportion of women in the group.

The finding has been influential and has also been critically examined. Several replication attempts produced divergent results, particularly concerning how much variance the factor actually explains. A meta-analysis of twenty-two studies (Riedl, Kim, Gupta, Malone, Woolley, PNAS 2021) confirms the existence of the factor while specifying more precisely the conditions under which it appears.

For practice, the decisive insight holds independently of the debate over effect sizes: the quality of collective thinking depends on conditions that can be designed, not on selecting the smartest individuals. Speaking time, social perception and composition are design questions.

On the role of AI: augmentation rather than substitution

In 2022, Erik Brynjolfsson coined the term “Turing Trap”. His argument: when AI is designed primarily to imitate and replace human performance, workers lose bargaining power and value becomes concentrated among those who control the technology. When AI is instead designed to augment human capabilities, new products and services emerge and the value created is distributed more broadly.

His central point for leaders is that incentives structurally favour automation because personnel costs are the most visible line item. The choice between substitution and augmentation is therefore not a technical decision, but a deliberate leadership decision.

A secondary calculation in the same paper is noteworthy: for every dollar invested in machine-learning technology, an estimated nine dollars must be invested in intangible human capital — in skills, processes and organisation.

On quiet non-use: the competence penalty

This is the best-supported empirical finding and one of the most underestimated in practice.

A study of almost 29,000 software developers at a large technology company (Gai, Hou, Tu, 2025) found that although the company had internally demonstrated a thirty percent productivity increase and actively promoted the tool, only forty-one percent of developers used it at all during the first year.

The explanation was not a lack of competence or access, but anticipated social judgement. In a controlled experiment, identical code received lower ratings when evaluators believed it had been produced with AI assistance. Perceived competence fell by around nine percent on average, with a substantially stronger effect for women than for men.

An independent study of 276 clinicians (Johns Hopkins, published in npj Digital Medicine) reached a comparable result: doctors who visibly incorporated AI into decision-making were judged by colleagues to be less professionally competent. The effect was weaker when AI was clearly used as a safeguard rather than as the primary decision-maker, but it did not disappear.

The consequence for leadership is substantial. When use is associated with a loss of status, the result is not open resistance but quiet non-use. Licence counts, training completion rates and acceptance surveys do not capture this. It only becomes visible in the usage rate after six to twelve months.

What follows from this

Taken together, the findings lead to four conclusions that extend beyond the keynote.

First. The bottleneck is not an organisation’s stock of knowledge, but its ability to turn the knowledge already present into robust shared judgement. The quality of judgement is not an individual but a collective quantity.

Second. AI adoption is not an IT task. It touches professional identity and social status. Programmes that fail to address this level produce usage rates far below what the business case and licence volume suggest.

Third. The choice between substitution and augmentation is the real strategic lever. It is not made in IT but in leadership, and it is usually made implicitly.

Fourth. When the source of value shifts, the systems that represent value must follow: compensation logic, career paths, capability development and organisational design. This is the uncomfortable part, and it is omitted from most transformation agendas.

Our own research

The magnitude cited in the keynote comes from our own survey of business transformation in the DACH region, initiated jointly with Prof. Dr. Katharina-Maria Rehfeld of International University Berlin. Completed transformation initiatives were examined according to whether they had systematically used collective intelligence, artificial intelligence, both or neither.

The finding: initiatives that combine both achieve significantly higher success scores than those that use one or neither. At the upper end, this translates into an up to fivefold higher probability of success compared with the standard approach.

The survey will continue in 2026, with a stronger focus on how transformation capability can be systematically developed and measured.

Sources and further reading

Collective intelligence

Woolley, A. W., Chabris, C. F., Pentland, A., Hashmi, N., Malone, T. W. (2010): Evidence for a Collective Intelligence Factor in the Performance of Human Groups. Science 330, 686–688.
https://www.science.org/doi/10.1126/science.1193147

Riedl, C., Kim, Y. J., Gupta, P., Malone, T. W., Woolley, A. W. (2021): Quantifying collective intelligence in human groups. PNAS 118.
https://www.pnas.org/doi/10.1073/pnas.2005737118

Malone, T. W. (2018): Superminds. The Surprising Power of People and Computers Thinking Together.

Augmentation rather than substitution

Brynjolfsson, E. (2022): The Turing Trap. The Promise and Peril of Human-Like Artificial Intelligence. Daedalus 151(2), 272–287.
https://www.amacad.org/sites/default/files/publication/downloads/Daedalus_Sp22_19_Brynjolfsson.pdf
Preprint: https://arxiv.org/abs/2201.04200

Competence penalty and adoption barriers

Gai, P. J., Hou, J., Tu, Y. (2025): Competence Penalty Is a Barrier to the Adoption of New Technology. Study of 28,696 software developers.
Summary: https://ssti.org/blog/employee-use-and-perceived-impacts-their-competence-may-be-behind-slow-ai-adoption-workplace

Yang, H. et al. (2025): Peer perceptions of physicians using generative AI. npj Digital Medicine, Johns Hopkins University.
Summary: https://www.futurity.org/doctors-artificial-intelligence-peers-3304862/

Transformation success rates

McKinsey & Company (2021): Losing from Day One. Why Even Successful Transformations Fall Short.
https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/successful-transformations

Boston Consulting Group (2020): Flipping the Odds of Digital Transformation Success.

For a critical discussion of the seventy-percent figure:
https://strategyu.co/do-70-percent-of-change-initiatives-really-fail/

Half-life of knowledge

Vahs, D., Brem, A. (2013): Innovationsmanagement. Von der Idee zur erfolgreichen Vermarktung. Schäffer-Poeschel.

For a critical discussion of the term:
https://www.bibb.de/dienst/publikationen/download/16571

Ricardo dos Santos Miquelino is Co-CEO of ... and dos Santos and a partner at the Swiss Future Institute. For more than fifteen years, he has worked at the intersection of collective and artificial intelligence with organisations including ZEISS, Deutsche Bahn and Coca-Cola.