When GenAI changes the task, what happens to the worker? (deep-dive)

Generative AI is changing the nature of work before it changes job titles. It enters organisations through everyday tasks: writing, coding, summarising, searching, classifying, analysing and supporting decisions. Each task may appear small in isolation, but together they alter where human value sits in the job.

This brief argues that deskilling and upskilling are often presented as opposites, when in practice they arrive together. Automation can remove repetitive effort and simplify execution, while raising the value of judgement, adaptability and work redesign. With Europe as the primary focus, the central question becomes: how can institutions and organisations use automation to simplify work while helping workers build the skills and know-how needed to move towards roles with lower exposure to routine automation?

The article explores what GenAI simplifies, why easier tools require stronger judgement, how the business operating model has to evolve, why the pressure to keep up is itself a workforce challenge, why agentic AI will intensify the transition, and how employers, educators and policymakers can help workers move with the work as it changes.

Keywords

#GenAI; #digital skills; #workforce transition; #AI literacy; #work redesign 

GenAI is changing tasks before it changes jobs  

Behind that shift in job titles lies a more granular reality. One in four jobs worldwide is exposed to some degree of transformation from generative AI (ILO and NASK, 2025). The important word here is transformation, not elimination. GenAI is already entering everyday work through writing, coding, summarising, searching, classifying and decision support. What matters most is how work itself changes when routine tasks become easier, faster or partly automated.

When a tool absorbs part of the task, the worker’s value moves. Less effort goes into producing the first version, and more goes into framing the question, judging the output, understanding the context and taking responsibility for the result. This is where the skills challenge begins: workers have to move with the work as it changes, while organisations, education systems and policymakers have to make that transition possible.

This is why the deskilling versus upskilling debate, though commonly framed as a choice, is a false binary. In practice, both happen at the same time. Automation simplifies some tasks, while it raises the value of judgement, adaptability and work redesign. The question worth asking, then, is how to make that simplification useful without leaving people behind.

For Europe, this matters more than it might first appear. Eurostat reported that 60% of EU citizens aged 16 to 74 had at least basic digital skills in 2025 (Eurostat, 2026a), against a Digital Decade target of 80% by 2030. That headline figure conceals sharper divides: only 33% of people aged 65 to 74 had at least basic digital skills in 2025, and the gap between people with higher and lower formal education remains wide (Eurostat, 2026a). Similarly, according to the 2026 State of the Digital Decade report of the European Commission, the current pace of up- and re-skilling appears to be insufficient to reach the 2030 targets of the EU Digital Decade both in terms of basic digital skills and number of experts in the Information and Communication Technology (ICT) sector. In 2025, the EU was just over halfway of the target of 20 million ICT experts in employment, counting 10.5 million specialists; while basic digital skills are forecasted to reach 68% by 2030, and 80% by 2036 (European Commission, 2026).  

Europe is therefore entering the GenAI transition while still building the foundations of the previous digital transition. That gap is not a side issue. It shapes who benefits, who adapts and who gets left at a standstill. 

What does GenAI actually simplify?

To understand where the transition actually bites, it helps to look at what GenAI can and cannot do. It is especially powerful in tasks that follow patterns: handling language, structuring information, creating first drafts, generating code, analysing text and producing summaries. That pattern-sensitive nature explains both where GenAI adds value and, just as importantly, where it reaches its limits.

The ILO-NASK 2025 global index finds that most jobs affected by GenAI are expected to be transformed rather than made redundant altogether, because human input remains necessary. In other words, this is a task-level transition: some parts of jobs become automated or assisted, while other parts remain deeply human. That distinction matters enormously for how we think about training, role design and workforce readiness.

It also points to something that is often misunderstood. When one task in a process can be automated, that does not mean the person performing it can be replaced. Out of the hundred tasks a person handles in a given role, few may become automated. That creates space: space to take on a new responsibility that opens value for the business, or space to focus more deeply on an existing task and raise its quality. This is precisely where training becomes strategic rather than defensive. The question shifts from "will this person still be needed?" to "what can this person now do better, or do next?" One useful way to understand the shift is through the history of programming languages. COBOL, Pascal, Perl, Java, Python and JavaScript all moved software creation progressively further away from machine-level syntax and closer to human reasoning. Each generation made programming more accessible, shifting value from mechanical instruction towards design, structure and problem-solving.

Now comes vibe coding: natural language as an interface to software creation. If high-level languages made code easier to write, this next step makes the language itself increasingly optional. Anyone can now prompt their way to a working piece of code. But that does not mean anyone can produce well-structured, secure, maintainable software through vibe coding. The output still has to meet standards. It still needs to be fast, secure, scalable and correctly architected. Understanding those touch points, knowing what good looks like and being able to set the guardrails: that knowledge does not disappear with the language barrier. It becomes more critical, because the barrier to generating code has dropped while the consequences of getting it wrong have not. As a result, the value of the developer shifts again: further from writing syntax, and further towards setting structure, defining standards and taking responsibility for what the code actually does.

This shift raises three questions that organisations and institutions have not yet answered clearly. First, what does the role of a junior developer look like going forward? It may well evolve away from writing boilerplate code and towards setting structure, defining guidelines and guardrails, identifying the touch points a system must cover and validating AI-generated output against those standards. Second, how could organisations rethink their training approach to ensure junior developers are genuinely equipped with these new skills, rather than being left to figure it out through trial and error? Third, how will universities, engineering schools and training centres adapt their curricula to reflect this shift? Teaching students to write syntax in isolation is becoming less relevant. Teaching them to think about architecture, security, performance and accountability is becoming essential. The same logic plays out across business functions. In finance, automation can support reconciliation, reporting drafts and anomaly detection, freeing professionals for interpretation, scenario thinking and decision support. In HR, GenAI can help prepare job descriptions, policy drafts and learning content, while judgement remains essential for culture, fairness and sensitive employee decisions. In customer service, simple queries can be automated, while complex cases still demand empathy, escalation and problem-solving. 

Why easier tools require stronger judgement

A common assumption says that easier tools reduce the need for learning. GenAI points to a more demanding picture. When tools become easier to access, more people use them. As usage spreads, more people need the skills to use them well. Simplicity at the surface does not mean simplicity in practice.

Cedefop’s 2024 AI skills survey, published in 2025, maps AI use in Europe and its impact on jobs and skills. Its findings show that AI is already reshaping workplace capability needs, while training models and support structures are still catching up (Cedefop, 2025). In the meantime, many workers are learning through experimentation in their daily work. That creates fast adoption, but it also creates uneven quality, unclear accountability and inconsistent practices across teams and organisations.

AI literacy is therefore becoming a baseline workplace capability, and it goes well beyond knowing which tool to open. Workers need to understand what AI systems can do, where they produce weak or biased outputs, how data and prompts influence results, and when human review becomes essential. Equally important is domain judgement, because AI output gains value only when it is connected to the business, legal, operational or human context of a decision.

The EU AI Act reinforces this direction (European Union, 2024). Article 4 requires providers and deployers of AI systems to take measures, to their best extent, to ensure a sufficient level of AI literacy among staff and other people dealing with AI systems on their behalf (European Commission, 2025a). This creates a practical bridge between regulation and capability-building. For employers, AI literacy becomes part of responsible deployment, not an optional add-on. For workers, employability increasingly depends on the ability to combine tool fluency with critical thinking and professional responsibility. The importance of AI literacy is recognised by DigComp 3.0, the EU’s main digital competence framework, where AI literacy is not outlined as a separate competence area, but rather embeds AI literacy transversally across the entire framework (Cosgrove and Cachia, 2025).

The skills that rise in value fall into five families.

First, digital confidence: the ability to use tools, data and platforms with ease. Second, AI literacy: the ability to understand AI outputs, limitations and risks. Third, domain expertise: the ability to judge whether an output fits the problem. Fourth, human and organisational skill: communication, collaboration, empathy and the flexibility to change navigation. Fifth, accountability: the ability to explain, validate and take responsibility for decisions, supported by automation.

Taken together, these skills show why simplification and capability-building have to advance in parallel. Even if automation makes a task easier, the worker still needs to understand the purpose, the risk and the consequence. Value shifts from performing the routine towards knowing how to use, question and improve the automated result. 

When tasks change, the business operating model has to follow

GenAI changes individual tasks, but it also changes the way work is organised around those tasks. Once a tool can draft, summarise, classify, search, compare or recommend, the question quickly moves from “can this task be automated?” to “how should the work be redesigned now?”

That shift matters because skills do not exist in isolation. They live inside a business operating model, shaped by processes, roles, decision rights, governance, performance measures and learning routines. A worker may learn to use a GenAI tool, but value appears only when the surrounding process also changes: who reviews the output, who owns the decision, where human judgement is required, how quality is checked, and how exceptions are escalated.

A better approach starts with the work itself. Organisations need to map which tasks are becoming easier, which carry higher risk, which roles are shifting, and which capabilities workers need next. The Organisation for Economic Cooperation and Development (OECD) has highlighted that AI adoption is creating emerging divides between firms, sectors and countries that have adapted their operating models and those that have not (OECD, 2025). Crucially, that divide is not primarily about technology access. It is about operating model readiness.

In finance, GenAI can help prepare reporting drafts, detect anomalies and summarise performance movements. As this happens, the finance role moves towards interpretation, scenario thinking, business advice and decision support. In customer service, GenAI can handle standard answers, classify requests and suggest responses. The human role shifts towards complex cases, emotional judgement, escalation and service recovery. In HR, GenAI can support job descriptions, policy drafts and interview preparation. The human value sits in fairness, culture, trust, sensitive conversations and organisational impact.

Across all of these examples, the same pattern holds. When routine execution becomes easier, workers need stronger judgement, context awareness, process understanding and accountability. Upskilling works best when it is tied directly to operating model change.  

The pressure no one talks about: keeping up is exhausting

One dimension of this transition rarely appears in policy documents, yet workers feel it every day. The pace of (Gen)AI change is, for many people, genuinely overwhelming.

Every week brings new model releases, new features, new claims about what is now possible or what is now at risk. Workers are asked to learn new tools, follow constant product updates, understand changing risks, protect their employability and continue delivering their existing work, all at once. This is a significant cognitive load to carry alongside an ordinary job, and it does not spread evenly. People with stronger digital confidence can filter the noise. Those already stretched by routine work are far more likely to fall behind, simply from the weight of it all.

The OECD’s work on emerging divides in the AI transition confirms that readiness gaps are widening between workers with higher and lower levels of education, between large firms and Small and Medium Sized Enterprises (SMEs), and between regions with stronger and weaker digital infrastructure (OECD, 2025). The pressure to keep up, in that light, is not a personal failure. It is a structural challenge that operates at scale.

There is also a subtler problem: distraction. When every week brings a new AI release, workers can struggle to identify what deserves their attention. Some chase every new tool. Others shut down and disengage entirely. Neither response builds durable capability. What workers need, more than anything, is clarity on which capabilities matter, which tools are relevant to their work, and what change is worth absorbing versus what is simply noise.

That clarity is a leadership responsibility. Without it, the pace of AI change creates anxiety rather than capability, and the stress of keeping up becomes a silent drag on performance, morale and retention. Recognising this as a structural issue, not a personal one, is the first step to addressing it properly. 

From GenAI to agentic AI: the transition will intensify

Although this brief focuses on GenAI, the same transition will intensify as AI systems become more agentic. Where GenAI mainly changes how workers generate, interpret and refine outputs, agentic AI goes further by supporting, coordinating or executing workflows with a degree of autonomy.

The OECD describes AI agents as systems that perceive and act upon their environment, using tools to achieve goals and adapt to changing inputs and contexts (OECD, 2026). This is not a distant prospect. Organisations are already beginning to treat AI agents as part of their available capacity. Some are exploring what it means to assign agents to specific tasks or workflows, the way they would deploy a contractor or a team. The language is still developing, but the direction is visible.

This matters for skills and jobs because the focus shifts from using a tool to supervising a system of action. When AI suggests a text, a human reviews the output. When AI executes a workflow, however, a human needs to define the boundaries, monitor the process, validate exceptions and understand where responsibility sits. The skill requirement moves accordingly: from prompt use towards oversight, process understanding, escalation design and accountability.

New responsibilities will emerge around agent orchestration, workflow oversight, quality control and governance. Organisations will need people who can design the boundaries within which agents operate, monitor what agents do, and take responsibility when things go wrong. Importantly, these are not technical roles in the narrow sense. They require judgement, context and accountability just as much as they require technical familiarity.

For Europe, this forward-looking dimension matters because institutions usually move more slowly than technology. Education systems, training providers, workplace agreements and regulatory practices all need time to adapt. The practical challenge is to build skills models that work for today’s (Gen)AI tools while also preparing for more autonomous systems. The encouraging point is that the core skill set remains consistent: AI literacy, judgement, domain understanding, process awareness, communication and accountability. These are not disposable skills that expire with each new model release. They are the foundations of durable professional value. 

How workers can move towards more resilient roles

The topic of transition often stays abstract. In practice, workers move best through adjacent pathways. A person rarely jumps from one occupation into a completely different one. More often, they move by building on existing experience and layering in new capabilities that fit emerging needs.

A practical transition map could look like this:

These pathways require training, but also visibility: people need to see where they can go before pressure forces them to move. Generic upskilling programmes too often remain isolated from the work itself. What helps is learning that connects directly to a worker’s role, tasks and next possible move.

The most resilient roles will combine three elements: human context, domain knowledge and responsible use of technology. Occupations that require trust, care, physical presence, negotiation, complex judgement, social intelligence or accountability have lower exposure to full automation. Yet, many roles can become more resilient by shifting their task mix. The aim is to help workers move from routine execution towards judgement-rich contribution, before automation removes the routine and leaves nothing to stand on.

This approach also matters for inclusion. Eurostat data shows wide differences in basic digital skills across age, education level and Member States (Eurostat, 2026a). People with higher formal education have significantly stronger digital skills than those with lower formal education. Left unaddressed, AI adoption will naturally flow towards those already best positioned to use it. Deliberate effort is therefore needed to extend transition support to mid-career workers, older workers, lower-skilled workers and workers in SMEs. 

What needs to happen: employers, educators and policymakers

Europe already has strong policy foundations. The Digital Decade sets the digital skills ambition. The Union of Skills frames capability development as a competitiveness priority and promotes upskilling, reskilling and better use of talent across the EU (European Commission, 2025b). The AI Act introduces AI literacy as a responsibility for organisations deploying AI systems. Together, these signals point in the right direction. The gap, however, is not in ambition. It is in execution inside workplaces and learning systems.

Employers need to treat (Gen)AI adoption as a workforce transition programme, not a technology deployment. In practice, that means mapping tasks, defining new responsibilities, identifying adjacent roles and giving workers time to practise. Training should be connected to applied use cases and supported by managers, because people learn faster when they can apply new skills in the flow of work, with feedback and with a clear reason to change. Equally, it means being honest about where roles are changing and providing clarity, not just access to tools.

Educators and training providers need to update programmes around applied AI literacy, data confidence and problem-solving. The deeper responsibility, however, is longer-term. Students entering the labour market will find AI already embedded in many tools and workflows. They need more than tool familiarity. They need learning habits, critical thinking, digital confidence, ethical awareness and the ability to keep adapting as systems evolve. The aim is to prepare people for continuous transition, not for a fixed set of tools.

Policymakers should focus support where transition capacity is weakest. SMEs need practical guidance, shared resources and access to training that does not assume a large HR department or a dedicated learning budget. Workers in routine-heavy roles need visible pathways before pressure builds. Regions and sectors with slower digital adoption need targeted support so that the AI transition strengthens European competitiveness across the whole economy, not just in the strongest parts of it. Public funding, social dialogue, vocational education and lifelong learning systems can all play a role in turning AI adoption into workforce progression.

Eurostat data shows that AI use reached 55% among large EU enterprises in 2025, but only 17% among small enterprises (Eurostat, 2026b). That gap does not just reflect differences in technology access. It reflects differences in advisory capacity, training capacity, governance capacity and management bandwidth. Closing it is therefore a policy priority with direct consequences for European competitiveness.

Conclusion: simplify tasks, strengthen people

Deskilling and upskilling belong in the same conversation, because they happen at the same time. Automation simplifies tasks, reduces friction and makes work easier, while it also raises the value of judgement, adaptability and work redesign. GenAI accelerates this shift and broadens it, because it reaches many knowledge tasks at once.

The challenge is transition: helping workers move towards roles where human strengths matter, while employers, educators and policymakers close readiness gaps before they become permanent.

There is also a human dimension that no policy framework fully captures. Many workers are navigating this shift while already stretched, uncertain and watching the tools around them change faster than they can follow. That experience deserves acknowledgement, not just activation measures. Transition works better when it is structured, when expectations are clear, and when people have a credible path forward, rather than a constant, shapeless pressure to keep up.

Europe can treat GenAI as another wave of disruption to manage, or it can use it as an opportunity to build a stronger and more equitable skills model. The better path is clear: simplify the work where automation creates value, strengthen the people where human judgement creates trust, and build transition pathways before pressure makes them urgent. That is how deskilling becomes productive simplification, and upskilling becomes a practical route into the next phase of work.

View this paper in full, together with its references via this link.

About the author

Joseph Antoun works across data, AI, digital strategy and business change. He helps leaders understand what technology means in practice: how it affects decisions, operating models, skills and the way people work. His perspective is shaped by regular conversations with senior technology and business leaders, as well as by hands-on experience in translating complex digital change into clear organisational choices.

He believes that AI adoption should be approached with both ambition and responsibility. The value of technology does not come from using new tools, but from knowing where they help, where they create risk, and what people need in order to work with them well. His work focuses on that transition: helping organisations build the judgement, confidence and structures required to turn digital change into lasting capability. 

About the organisation

Knowledge Sphere is an advisory and capability-building practice focused on data, AI, digital transformation and the future of work. It helps organisations connect technology adoption with business value, operating model change and workforce readiness through advisory, training and talent-related initiatives. Its work focuses on helping people and organisations adapt with confidence as digital technologies reshape work, roles and decision-making. More information is available on Knowledge Sphere’s website.