Most future-of-jobs forecasts appear to disagree on the numbers, but all these statistics are variants that use the same underlying approach: they estimate task exposure and substitution, i.e., breaking jobs into tasks, estimating which tasks AI can automate, and inferring what happens to occupations as those tasks are reorganized. For example, at the global level, the IMF estimates ~40% of jobs are exposed to AI (higher exposure in developed countries due to the prevalence of cognitive-task-oriented jobs), the ILO places the figure closer to 24%, and Ernst and Young (EY) estimates the exposure could reach as high as 59%.[1]^,[2],[3]

This lens of task exposure is useful, but it treats the economy as a linear machine. In fact, if you read the fine print, the authors are careful – even humble – about what their models do not include. The ILO explicitly notes that its analysis does not account for entirely new occupations that may emerge. The IMF acknowledges that its framework is static; it does not model feedback effects, societal preference shifts, regulatory responses, infrastructure constraints, or the political economy of adoption.

But therein lies the problem: by holding constant the very forces that tend to move most dramatically during technological revolutions, we may be significantly myopic about the possibilities and directions of change.

In reality, when AI changes what a worker can do, it simultaneously causes a chain reaction of many things, including costs, pricing, consumer expectations, and bargaining power. These forces move together, amplify one another, and produce outcomes that cannot be understood by summing task substitutions alone.

Seen this way, task substitution, i.e. changes in what tasks a human does vs. AI, is only one of many chain reactions that the advent of AI sets off. There are at least two other chain reactions that set off simultaneously: value migration (i.e., where people see value and would be willing to pay a premium) and shifts in power structures (i.e., who controls standards and platforms, and determines where the value of productivity gains is captured).

First, as digital outputs become abundant and harder to distinguish, willingness to pay often shifts toward what remains scarce (i.e., value migration). History makes this clear. When quartz technology made accurate timekeeping cheap and widely accessible, the premium in watches increasingly moved toward heritage, craftsmanship, and brand meaning. Similarly, when streaming platforms made the consumption of recorded music inexpensive, artist revenue shifted toward live concerts, festivals, and merchandise: experiences that could not be infinitely replicated.[4]

Second, as new rules or platforms are introduced to regulate and/or intermediate the use of AI and digital outputs, they shape who benefits from productivity gains, i.e., whether productivity gains turn into higher wages or if these gains are concentrated in the hands of those who control the system (i.e., shifts in power structures). History again shows this pattern.  When containerization transformed global trade, it made shipping more efficient and also rewarded the actors who controlled ports, set standards, and coordinated rail and trucking networks.  Similarly, large digital platforms have managed to connect millions of buyers and sellers. However, as the owners of these platforms control access to users and set rules for sellers/creators, they significantly influence how much income flows to sellers/creators versus to the platform.

When we examine the future of work through this expanded lens of shifts in work, value, and power structures, many interesting conclusions become plausible. For example, here are two of many possibilities that the future may hold:

What if, in a world of infinite melodies, an economy of magical moments emerges?

Streaming platforms already host 100+ million songs, with ~100,000 new uploads every day.[5]^,[6] Generative AI could push that from millions to billions, overnight. When polished digital music becomes infinite, its marginal value may collapse. What becomes scarce instead is presence.

Taylor Swift’s Eras Tour sold ~10 million tickets, at an average price of about $200.[7]^,[8] Fans were not paying for access to her songs, they were paying for the experience of being there. And this could very well become the norm over time, not restricted to concerts by the world’s biggest stars.

Indeed, if half of the Eras tour ticket buyers decided to seek human, in-the-room experiences in an AI-saturated music world, and began attending just one local concert per quarter, at one-tenth the ticket price (≈ $20 per ticket), that alone would generate a spending of roughly $400 million per year. This would flow into local venues, local artists, and the surrounding ecosystem: sound engineers, lighting, stage crews, ticketing, security, promoters, hospitality, and small-scale cultural infrastructure. And that’s from one plausible behavioural shift, anchored in one fandom cohort.

Mainstream exposure models can’t “see” this kind of change, because it isn’t a task substitution story. Rather, it’s a value migration story: when digital perfection becomes abundant, the premium shifts to what remains scarce – human presence, locality, and shared experience.

What if, once diagnosis becomes cheap, “being a doctor” stops being the scarce resource, and AI-enabled nurses is what patients need?

Diagnosis is largely pattern recognition: matching symptoms, history, and tests to likely conditions. But what happens when even the most basic AI systems can recognize patterns across millions of cases, drawing on a wealth of global medical data far beyond the experience of any single doctor?

The conversations we have today around the number of doctors per population would shift dramatically. For example, the recommended doctor-to-population ratio to deliver essential health services is 1 per 1,000 people.^[9] In many African countries, however, the ratio is significantly higher, with a single doctor serving nearly 4,900 people.^[10],[11] This is a seemingly insurmountable gap to close.

Yet, if AI turns diagnosis into a low-cost “first step” that nurses, community health workers, and patients can reliably use, what would happen? If each of the 1.3 million nurses in Africa could be AI-enabled to perform reliable diagnostics and act as a doctor, this insurmountable workforce gap could shift dramatically to 960 people per doctor, well within the recommended threshold.^[12]

Further, the distinction between doctors and nurses may blur. Their role may shift from identifying diseases from symptoms to interpreting AI outputs, validating results, and helping patients understand and emotionally process the personal implications of an AI-generated report. Care becomes less about ‘who can name the disease?’ and more about ‘who can guide the decision, ensure adherence, manage risk, and deliver treatment in messy real life?’ – including emotional processing, trust, follow-up, and system navigation.

And then, the power question then becomes decisive: do these expanded roles and broader access translate into higher incomes and greater security for frontline healthcare providers? Or does value concentrate with those who design the algorithms, control the tech platforms, and own the medical data infrastructure, ultimately keeping affordable healthcare out of reach for millions?

The future of work, then, will be shaped less by what AI can do and more by where value migrates and who steers where it lands.

So, what does this mean in practice?

If task substitution is only one part of the story, then preparing for the future of work cannot be reduced to scaling reskilling programs based on occupation forecasts alone. Instead, the priority for all actors should be building the capacity to read and respond to signals of value and power reshuffles. Specifically:

For policymakers: (i) move beyond rigid workforce plans and build labour systems that adapt in real time and (ii) shape who sets AI standards, controls data, and defines compliance, because these governance choices determine whether productivity gains broaden or concentrate power and income.

For education and skilling institutions: (i)shift from training for specific job titles to building capabilities that remain valuable even as the system evolves, (ii) offer multiple entry and exit points for learners, and (iii) shorten feedback loops with employers so that programs evolve quickly as demand changes.

For employers: (i) anticipate where AI will commoditize offerings and where willingness to pay will shift, and (ii) consider how productivity gains are shared to sustain trust and performance within the workforce.

For individuals: (i) Focus less on chasing “hot jobs” and more on tracking where premiums emerge (e.g., wherehuman judgment, trust and presence become newly scarce and valued), and (ii) build durable, uniquely human capabilities that remain valuable.

In a world of simultaneous changes in task composition, value, and power structures, nimbleness becomes the decisive asset. The future of work debate must therefore mature. Counting jobs still matters, but understanding how value shifts and who captures it matters more. Instead of asking whether AI will create or destroy jobs, we should ask whether our policies, institutions, and career strategies are flexible enough to cope with the chain reactions that AI is triggering.

^{[1] IMF, Gen-AI: Artificial Intelligence and the Future of Work, 2024}

^{[2] ILO, Generative AI and Jobs: A Refined Global Index of Occupational Exposure, 2025}

^{[3] EY, The impact of GenAI on the labor market, 2024}

^{[4] Arjun Sharma, Streaming vs. Live Music: How Changing Consumer Preferences Are Reshaping Revenue Models in the Music Industry (A Field-Based Study), 2025}

^{[5] Murray Stasses, Music streaming platforms now host quarter of a BILLION tracks. Where does it end?, 2026}

^{[6] Tim Ingham, It’s happened: 100,000 tracks are now being uploaded to streaming services like Spotify each day, 2022}

^{[7] Time, A Look Back at Taylor Swift’s Record-Breaking Eras Tour, 2024}

^{[8] Eddie Fu, Taylor Swift’s “Eras Tour” Concludes with Record-Breaking $2 Billion in Ticket Sales, 2024}

^{[9] Esther E. A. et al., Assessment of the preparedness of the Ugandan health care system to tackle more COVID-19 cases, 2020}

^{[10] Worldometer, Africa Population, accessed February 2026}

^{[11] A. Ahmat et al., The health workforce status in the WHO African Region: findings of a cross-sectional study, 2022}

^{[12] A. Ahmat et al., The health workforce status in the WHO African Region: findings of a cross-sectional study, 2022}

<< BACK TO THE DALBERG OBSERVATORY