Introduction

Working with AI has something particular about it: we not only have to keep learning, we also have to learn how to unlearn. The speed at which technology is evolving means that accumulating knowledge is no longer enough. We also need the adaptability to question what we know, let go of ways of working that no longer make sense, and learn again.

But there is one question I find increasingly difficult to ignore: Can we learn at the same pace as technology evolves?

AI needs good prompts that describe what we are trying to achieve, but it also requires something much harder to automate: knowing when to ask another question, what information is missing, which alternatives should be compared, and how to evaluate the result. This can make experienced professionals even more productive because their existing knowledge allows them to get more value from these tools.

And this is where a paradox emerges: if we want experienced professionals tomorrow, someone has to hire and develop the juniors of today. At the same time, we constantly talk about reskilling and upskilling as responses to the future of work, yet it is still not entirely clear who should be responsible for making that transition possible.

• Reskilling means developing a new set of skills to move into a different role or career.
• Upskilling means developing new capabilities within the work we already do.

So, rather than trying to predict which jobs will disappear or which skills we will need tomorrow, I wanted to look at what is actually happening. Not the hype. Not the panic. The data.

That leads me to three questions:

• What do the data actually tell us about the jobs AI is creating, changing and displacing?
• If companies need new skills faster than ever, why might entry-level roles be particularly vulnerable?
• Who should be responsible for preparing people for this transition: companies, governments, universities, or all of them?

What kind of impact can AI have on a role?

Before looking at what is happening to junior professionals, it is worth making one distinction: not every role exposed to AI will necessarily have the same future.

The first dimension comes from the AI Occupational Exposure Index (AIOE) [4], which measures the relationship between AI capabilities and the skills required across different occupations. Researchers at the International Monetary Fund (IMF) later introduced a complementarity dimension, helping distinguish between occupations where AI is more likely to enhance human work and those where it has greater potential to substitute for a significant share of tasks. PwC later incorporated both dimensions into its Global AI Jobs Barometer 2025 to analyse the potential effects of AI across occupations.

source:Source: Author’s adaptation based on PwC, The Fearless Future: 2025 Global AI Jobs Barometer.

PwC uses these two dimensions to distinguish between occupations with greater potential for augmentation and those with greater potential for automation.

In PwC’s analysis, an occupation is considered to have high AI exposure when its exposure index is above 0.5. From there, a complementarity score above 0.5 places the occupation in the augmented category, while a score below 0.5 places it in the automated category.

However, this matrix should not be interpreted as a definitive prediction about the future of a profession. It is a starting point for understanding how AI may affect a role, because technological exposure is not the only factor shaping what ultimately happens. For example:

• Regulation may limit which tasks can be performed by AI.
• Professional accountability may require certain decisions to remain in human hands.
• Sector characteristics, professional norms and levels of autonomy can influence how AI is adopted.
• The way an organization structures its processes can also significantly change the outcome.

As a result, two roles with a similar level of AI exposure may evolve in very different ways.

This is where the classification becomes particularly useful for thinking about talent. Instead of simply asking whether a role is “at risk” from AI, we can start asking which tasks are being automated, which are being augmented, and which increasingly depend on human knowledge, experience and judgement.

That distinction will be important for understanding what is happening to junior professionals.

2. The challenge facing junior professionals

If AI can significantly increase the productivity of an experienced professional, it is worth asking what happens to those who are still building that foundation. The technological transition is not affecting work evenly. Instead, it may be reshaping the threshold for entering the labour market.

Several recent sources point in this direction. In the Anthropic Economic Index survey, Cadences (June 2026), more than one third of respondents estimated that a more junior colleague had a greater than 60% chance of losing their job. A separate Anthropic analysis published in March 2026 also identified an approximately 14% slowdown in hiring among workers aged 22 to 25 in AI-exposed occupations following the launch of ChatGPT, without an equivalent slowdown among more experienced workers.

LinkedIn’s AI Labor Market Update from August 2026 points to a similar pattern. Entry-level hiring in AI-augmented occupations declined by between 3 and 10 percentage points more than overall hiring within those same occupations.

The contraction affected several areas:

• 🎨 Design and interface: hiring for UX Designer and UI Designer roles fell by between 34% and 49% year over year in the United States and India.
• 🔍 Analysis and management: entry-level hiring for Data Analyst roles fell by 27% in the United Kingdom, while hiring for Project Manager roles declined by 39% in France and 29% in Germany.
• 🛠️ Software development: entry-level Software Engineer roles fell by 32% in Germany, while Frontend Developer roles declined by 39% in India.

The pattern is also visible at a more local level. In Barcelona, the Digital Talent Overview 2026 shows that between 2023 and 2025, job postings for junior profiles fell by 26.93%, while demand increased for mid-level (+21.68%) and senior (+15.04%) profiles.

These figures do not prove that AI alone is responsible for the change, but they do suggest that entry-level workers are facing a different labour-market dynamic from more experienced professionals.

There is also another perspective that hiring data alone cannot capture.

From my experience working on software projects, I increasingly see the assumption that introducing AI automatically means accelerating implementation. And to some extent, that is true. We can generate code, build prototypes and automate certain tasks much faster.

But building software still requires much more than writing code: understanding requirements, designing the architecture, working with data, developing, deploying, testing, monitoring and maintaining the solution. Once AI is introduced, new requirements also emerge around evaluation, telemetry and observability to make sure the system behaves as expected.

AI can accelerate the construction of a solution, but it does not remove the need for engineering.

This creates another tension. If projects become shorter and teams are expected to deliver faster, it may become harder to integrate junior professionals precisely because a larger share of the remaining work requires autonomy, decision-making and contextual understanding from the outset.

At the same time, we are working with technologies that continue to evolve rapidly, meaning that everyone, including senior professionals, is still learning.

3. The experience paradox

The evidence above suggests that entry-level professionals may be facing increasing pressure. But the question does not end with hiring. If organisations increasingly value experience, autonomy and judgement, how are people supposed to develop those capabilities in the first place?

3.1. Does AI make us more productive, or does it change what productivity means?

The same sources discussed earlier also reveal another side of the story: AI is not only affecting who gets hired, but also how work is performed once someone is in the role.

The Digital Talent Overview 2026 helps illustrate this shift in the context of software development, where AI is already being used across several everyday tasks. Among the 372 software development and digital professionals surveyed in the Barcelona metropolitan area, 74.20% considered AI effective for generating documentation and 74.19% for explaining or understanding existing code.

Yet faster execution does not mean that the output can simply be accepted as it is. 87.80% of respondents said they always or almost always review AI-generated code before accepting or integrating it. In addition, 68.92% reported that these tools can produce code that appears correct at first glance but is not reliable, while 63.25% said that achieving high-quality results still requires substantial effort in prompting, correction and refinement.

These findings suggest that productivity gains do not eliminate the need for human judgement: AI can make certain tasks faster, but the output still needs to be understood, reviewed and validated.

The Anthropic Economic Index adds a broader perspective. Beyond the concerns about early-career employment discussed in the previous section, the Cadences survey also explored how active Claude users perceived AI to be changing their own work.

This reveals an important tension: AI can accelerate certain parts of the work, but that speed does not remove the need to understand, review and validate what it produces. In software development, generating code or documentation may take less time, but deciding whether a solution is correct, secure, maintainable and appropriate for a specific context still requires knowledge and judgement.

That ability to evaluate an output, identify potential issues, compare alternatives and understand their implications is precisely one of the capabilities that professional experience helps us develop.

3.2. What happens when the work that used to teach us starts to disappear?

For years, many professional careers followed a fairly natural progression. Junior professionals often started with relatively simple tasks such as implementing features, fixing bugs, writing documentation or working on well-defined problems. Over time, those tasks helped them understand systems, recognize patterns, learn why certain decisions had been made and gradually develop judgement.

AI is beginning to intervene precisely in some of those early layers of work. If a tool can generate code, explain a function, create documentation or solve a task that was previously assigned to a junior professional, we gain speed. But we also change one of the traditional ways in which that professional used to learn.

This connects with another pattern in the sources analysed: AI tends to absorb tasks based on knowledge that can be codified more easily, while capabilities such as judgement, negotiation, contextual understanding and evaluating alternatives depend much more heavily on experience.

The hiring patterns discussed earlier add another dimension to this challenge. If organisations increasingly seek professionals who can contribute greater autonomy and experience from the outset, they may also need to think more deliberately about how early-career professionals are given opportunities to develop those capabilities.

That does not mean every junior task will disappear, nor that the only way to become senior is to follow exactly the same career path as before.

But it does raise an important question: if some of the tasks we used to learn through are disappearing or changing, how do we build that experience now?

4. If work is changing, we also need to rethink how we learn

Today, terms such as reskilling and upskilling are often used almost automatically whenever we talk about the impact of AI on employment. But what they really mean, when they apply and how they should be implemented are questions that come up frequently in conversations with colleagues and teams.

We have courses, training programmes, mentoring and many other ways of learning. The challenge is not simply having access to them, but knowing how to combine and structure them, because people do not all start from the same place or learn in the same way. The key is to understand what each person needs to learn, what knowledge they already have, and how they can build new capabilities while the nature of work itself is changing.

Before thinking about specific techniques, it is useful to distinguish the type of development we are trying to achieve:

1. 🌱 Upskilling. Developing new capabilities to improve or expand what we already do. For example, a lawyer using AI to analyse documents or prepare an initial contract review.
2. 🔄 Reskilling. Developing the capabilities needed to move into a different role. For example, someone working in operations preparing to transition into a role related to data analysis or automation.
3. 🔗 Cross-skilling. Building knowledge in an adjacent area that complements an existing profile. For example, a marketing professional learning the fundamentals of data analysis to better interpret campaigns and collaborate with technical teams.
4. ⚠️ Deskilling. The movement can also happen in the opposite direction. If technology systematically takes over certain tasks, some capabilities may be practised less often and gradually weaken. For example, if we consistently leave the initial assessment of a problem to AI, we may lose some of the practice that previously helped us develop judgement.

The following image is not intended to be a closed academic taxonomy. It brings together professional development concepts, learning mechanisms, methodologies and practices that can be used to design different learning journeys. They do not all belong to the same category, nor do they need to be used independently. Many of them complement each other and can form part of the same learning process.

One simple way to read the image is to think about which aspect of learning each block addresses:

• Alongside expertise [Who do you learn from?]: Learning alongside more experienced people through mentoring, coaching, shadowing or pair working.
• Progressive challenge [How do you stretch?]: Gradually increasing difficulty, autonomy or responsibility through stretch assignments, scaffolding or deliberate practice.
• Learning by doing [How do you build skill?]: Developing capabilities through projects, simulations and real-world situations.
• Feedback & reflection [How do you improve?]: Using feedback and reflection to understand what works and what needs to change.
• Learning in the flow of work [When does knowledge arrive?]: Accessing knowledge at the moment it is needed through microlearning, support resources or self-directed learning.
• Social learning [Who helps learning spread?]: Sharing experiences and solutions through peer learning, communities of practice or interest groups.

The important point is that we do not need to choose just one of these approaches. Different learning mechanisms can be combined depending on the person, the knowledge they already have and the capabilities they need to develop.

In this context, deskilling introduces another dimension. Some tasks may be practised less often because they become partially or fully automated. That is not necessarily negative. It can free up time and reduce repetitive work. But it does change which capabilities we continue to exercise and which ones stop being part of everyday learning.

And this opens up a broader question: if people start from different experiences, knowledge and ways of learning, should we begin to think about more personalised learning pathways?

AI itself could play a dual role here: not only transforming the skills we need, but also helping us adapt content, pace, practice and support to each person.

Could AI also help us personalise the way we learn and develop new capabilities?

5. Redesigning early-career pathways in the age of AI

So far, we have looked at training, upskilling and the different ways people can learn within the flow of work. But there is a more structural question: what happens if the pathway through which someone enters a role, learns and builds experience is also changing?

The report Artificial Intelligence and the Future of Entry-Level Work, published by the World Economic Forum in collaboration with PwC in 2026, addresses precisely this challenge. Its starting point is that the impact of AI on entry-level roles will not depend only on what the technology can do, but also on the decisions we make about hiring, job design, talent development and education.

The report proposes four interconnected dimensions for thinking about those decisions: Job Access, Job Design, Talent Pipelines and Education System Alignment.

What I find particularly interesting about this framework is that it shifts the conversation from “which skills should we teach?” to a broader question: how do we build pathways that allow people to enter the workforce, learn, gain experience and progress professionally in an environment where AI can also perform part of the work?

n the image, I have summarised some of the concepts I consider most relevant within each dimension. They do not represent every recommendation in the report, but rather a selection of mechanisms that help illustrate how these pathways could evolve. [5]

5.1. Job Access: keeping the entry doors open

The challenge begins even before learning takes place: to build experience, people first need the opportunity to gain it.

Keeping entry points into the labour market open means deliberately considering how people access their first opportunities while AI changes entry-level roles and tasks. Some of the practices that can support this include:

• 🚪 Entry-level hiring. Embedding early-career hiring into workforce and talent planning, rather than allowing it to disappear as an unintended consequence of automation decisions.
• ⚙️ Work experience. Creating opportunities for people to gain practical experience and demonstrate their capabilities, particularly when “previous experience” itself becomes a barrier to entry.
• 🚧 Alternative pathways. Complementing traditional routes with internships, apprenticeships, placements and other mechanisms that provide different ways into the workforce.
• 🎓Access to AI tools. Providing access to AI tools and training so that the ability to experiment and develop these capabilities does not depend entirely on an individual’s personal resources.

Dropbox provides an interesting example of this approach. The company expanded its internship and new-graduate programmes by 25% and chose to reinvest part of the productivity gains associated with AI into higher-value work, while continuing to preserve entry routes for early-career talent.

5.2. Job Design: redesigning work so people can keep learning

Once someone gains their first opportunity, a second question emerges: what kind of work allows them to continue developing capabilities while AI takes on part of the workload?

Redesigning entry-level roles means deciding how work should be distributed between people and technology, and which experiences should be intentionally preserved because they remain important for developing judgement, autonomy and contextual understanding. Some of the practices that can support this include:

• Human–AI collaboration. Designing tasks and workflows around what people contribute, what AI can take on and how both can work together effectively.
• Automate vs retain. Deliberately deciding which tasks to automate and which to preserve because they remain important for developing professional judgement, domain knowledge or the ability to detect errors.
• Judgement & decision-making. Shifting part of the work away from mechanical execution and towards interpreting results, evaluating alternatives and making decisions.
• Learning by doing. Preserving opportunities to practise in real situations, particularly when AI begins to take over some of the tasks that traditionally helped people learn.

Hitachi provides a clear example of this approach. The company automates repetitive and transactional work, while intentionally preserving experiences that help build judgement, critical thinking and business understanding. In engineering, for example, early-career professionals continue to develop programming fundamentals before relying heavily on AI-generated code, so that they can validate and challenge what the technology produces.

5.3. Talent Pipelines: building new forms of progression

If entry-level roles are changing, and the tasks within them are changing as well, another challenge emerges: how do we build the experience people need to continue progressing professionally?

Rethinking talent pipelines means moving away from the assumption that development happens only through a traditional hierarchical sequence and considering more flexible pathways based on capabilities, exposure and mobility. Some of the practices that can support this include:

• Capability-based progression. Making progression depend more on demonstrated capabilities and impact, rather than only on tenure or job title.
• Non-linear movement. Allowing development to happen through movement across projects, functions or domains, rather than only by climbing a vertical career ladder.
• Cross-functional exposure. Exposing people to different teams, problems and contexts to broaden their understanding of the business and develop complementary capabilities.
• Mentoring & mobility. Combining support from more experienced professionals with internal mobility opportunities that allow people to build knowledge across different contexts.

Dentsu Japan shows one way of making this type of progression visible. The organisation created four internal levels of AI capability development, AI Basic, AI Facilitator, AI Master and Chief AI Master, to recognise progression beyond job title alone.

The report also highlights Allianz, which uses intergenerational mentoring tandems to combine the digital fluency of younger professionals with the experience, business context and professional judgement of more senior colleagues.

5.4. Education Alignment: bringing education closer to real work

The final dimension extends the challenge beyond organisations. If the capabilities required by the labour market are changing faster, education and training also need to remain connected to the evolving nature of work.

Aligning education with the labour market means creating more continuous mechanisms of collaboration between education providers, employers and other actors in the ecosystem to reduce the distance between what people learn and what they eventually need to apply. Some of the practices that can support this include:

• Work-integrated learning. Embedding projects, practical experience and real-world problems into education so that learning is not separated from application.
• Employer partnerships. Building more continuous relationships between employers and education providers to share how roles, technologies and capability requirements are evolving.
• Co-designed curricula. Designing parts of educational programmes with input from employers and industry specialists to reduce the gap between education and work.
• Skills frameworks. Using shared frameworks to describe required capabilities and provide employers, educators and individuals with a common language.

Singapore provides a particularly interesting example of this approach. Its ecosystem combines labour-market data, National Skills Frameworks and skills-based job matching tools to connect employers, education providers and individuals.

The report also highlights examples such as Merck, which works with universities to define required capabilities and connect educational programmes with hiring opportunities, and Guanghua School of Management, which has incorporated AI, applied projects, internships and employer collaboration into its educational model.

6. Conclusions: if work is changing, the way we build experience needs to change too

After looking at how AI is affecting roles, entry-level positions and traditional ways of learning in different ways, one idea runs through the entire article: the transformation is not happening only in the tasks we perform, but also in the way we build the experience needed to perform them well.

🔄 Are we changing what it means to be junior or senior?

For a long time, many career paths followed a relatively predictable logic. Junior professionals started with more limited and repetitive tasks that allowed them to practise, make mistakes and gradually understand a system, while more experienced professionals took on decision-making, coordination and supervision.

AI is beginning to change that distribution. Some of the coding, documentation, initial analysis and information gathering that used to form part of the learning ground can now be completed much faster. At the same time, approaches such as vibe coding make it possible to build solutions at a speed that would have been difficult to imagine only a few years ago. Yet they still require us to understand what we are trying to build, provide the right context, review the output and recognise when something that looks correct actually is not.

This creates an interesting paradox: experienced professionals may begin performing tasks directly again that they previously delegated, while people at the beginning of their careers may encounter supervision and validation tasks earlier, even though they are still developing the judgement required to perform them well.

The traditional boundaries between what we consider a junior role and a senior role therefore begin to look less clear. If we automate part of the work through which experience was traditionally acquired, we will need to find new ways to build that experience.

🧠 Does learning also mean unlearning?

Years ago, a manager I worked with used to say that one of the biggest challenges technology creates for experienced professionals is learning how to unlearn. Over time, that idea has become increasingly relevant to me.

Experience remains valuable, but not because it allows us to repeat indefinitely what we already know how to do. Its value also lies in recognising which knowledge remains useful, which practices need to change and when a way of working needs to be reconsidered.

People at the beginning of their careers face a different situation. They are often building their foundational knowledge at the same time as these new tools become part of the normal way of working. That does not mean one group is better prepared than the other. Their starting points and learning needs are different.

In this context, deskilling also becomes increasingly relevant. Automating a task can free up time, reduce repetitive work and allow people to focus on higher-value problems. At the same time, however, it can mean that certain capabilities are practised less often. The question should not necessarily be how to avoid that process altogether, but rather which capabilities we can afford to practise less and which ones we need to preserve because they will remain essential for supervision, questioning and decision-making.

🎯 Will we all continue learning in the same way?

These different starting points also challenge a fairly common practice: designing large-scale training programmes that are essentially the same for everyone.

From my experience as both a professional and an educator, it is becoming increasingly clear that different groups have different needs, even when they work within the same organisation. Their prior knowledge differs, their experience differs, the problems they solve differ, and so does the way they develop new capabilities.

AI could play a particularly interesting role here. The same technology that is accelerating the need to learn can also help us personalise content, exercises, pace, feedback and support according to what each person already knows and what they still need to develop.

Bill Gates has highlighted this potential when discussing systems capable of understanding a person’s interests and learning style, measuring their comprehension, and adapting content and feedback accordingly. The opportunity is not simply to digitise the training we already have, but to begin thinking about much more personalised learning pathways.

🌍 What if AI could also expand opportunity?

After discussing automation, junior roles, deskilling and changing career pathways, it would be easy to end this analysis by focusing only on the risks. Yet the same technology creating these tensions can also expand what is possible.

Yann LeCun captured this perspective in a simple phrase: “AI will amplify human intelligence.” The idea matters because it shifts the conversation away from replacement and towards the expansion of human capabilities.

That potential may be especially significant in areas where resources remain scarce. Education and healthcare are two clear examples. Bill Gates has argued that AI can play an important role in sectors where there is a major shortage of professionals, helping expand access to support, knowledge and learning opportunities that are not currently available equally to everyone.

This does not remove the challenges explored throughout the article, but it allows us to close the discussion from a different perspective. We have seen that not every job will be affected in the same way, that junior positions are facing particular pressures, that experience remains important for providing context and judgement, and that some of the tasks traditionally used to build that experience are changing.

That is why responding with more courses, upskilling or reskilling may not be enough. We may also need to rethink how people learn through work, how roles are designed, how career pathways are built and how education and the labour market remain connected.

AI can transform roles, accelerate work and expand our ability to learn. It can also force us to question structures that we have taken for granted for decades.

Technology can expand our possibilities... The question is how we choose to use them.

References

• [1] Digital Applied Team. (2026, July 1). AI won’t take your job, but it will reshape it. Digital Applied. https://www.digitalapplied.com/blog/anthropic-economic-index-jobs-change-not-vanish-2026

• [2] World Economic Forum. (2025, January 7). Future of Jobs Report 2025: 78 million new job opportunities by 2030 but urgent upskilling needed to prepare workforces. https://www.weforum.org/press/2025/01/future-of-jobs-report-2025-78-million-new-job-opportunities-by-2030-but-urgent-upskilling-needed-to-prepare-workforces/

• [3] LinkedIn Economic Graph. (2026, August). AI Labor Market Update: August 2026. LinkedIn. https://economicgraph.linkedin.com/workforce-data/publications

• [4] PwC. (2025). The fearless future: Global AI Jobs Barometer 2025: Methodology appendix. https://www.pwc.com/gx/en/issues/artificial-intelligence/job-barometer/2025/methodology-report.pdf

• [5] World Economic Forum. (2026, June 22). Artificial intelligence and the future of entry-level work: A framework for safeguarding and reinventing early career pathways. In collaboration with PwC. https://www.weforum.org/publications/artificial-intelligence-and-the-future-of-entry-level-work-a-framework-for-safeguarding-and-reinventing-early-career-pathways/

• [6] United Nations Security Council. (2024, December 19). Security Council, seventy-ninth year: 9821st meeting, Thursday, 19 December 2024, New York (S/PV.9821). United Nations. https://digitallibrary.un.org/record/4070651

• [7] Gates, B. (2023, March 21). The age of AI has begun. Gates Notes. https://www.gatesnotes.com/the-age-of-ai-has-begun

• [8] Mobile World Capital Barcelona. (2026, May 27). Digital Talent Overview 2026. https://mobileworldcapital.com/en/trends/digital-talent-overview-2026/