Generative AI Doesn’t Transform Just Jobs: It Transforms Tasks
Why job descriptions are becoming insufficient, and what this means for HRIS and HR strategy
This article is the first part in a five-part series on skills-based organizations in the era of GenAI. It was inspired by a conversation with an HR partner on LinkedIn, and a few days later, by listening to episode 131 of IA Café from OBVIA, which explores the human and ethical implications of artificial intelligence at work.
The analyst who didn’t change her title
A few weeks ago, I was talking with an HR business partner in a mid-sized organization. She was describing the situation of an HR analyst on her team: same title for three years, same salary band, same job description.
Except her work had changed.
Before GenAI, this analyst spent about 40% of her time compiling data, formatting reports, drafting first versions of policies, and preparing presentations for committees. Today, these tasks are cut in half: some disappear entirely, others accelerate. She now spends that recovered time validating AI outputs, formulating more precise prompts, contextualizing recommendations for managers, and acting as a bridge between raw data and human decision-making.
Her title: HR Analyst. Her job description: unchanged. Her actual work: profoundly different.
This gap between the administrative container (the job) and operational reality (the tasks) is at the heart of what GenAI produces in organizations. And it is this gap that makes the traditional job description insufficient as a unit for managing work. This phenomenon is well documented: Selenko et al. (2022) show that workers’ professional identity is tested precisely because the title does not change while the tasks transform profoundly.
The job: a useful but static administrative container
The job description is not an arbitrary invention. It was designed to meet real needs: pay equity, clarity of responsibilities, legal compliance, workforce planning. In a world where work changes slowly, it serves these purposes well.
But the job description is a container. It describes a set of responsibilities, required competencies, and performance expectations at a given point in time. It is, by construction, static. It is typically reviewed once every two or three years in most organizations. And it is written in terms of what the job demands, not in terms of what work actually entails day-to-day.
This model worked because work itself changed at a manageable pace. A financial analyst in 2010 did essentially the same work as in 2005. Tools changed (Excel, then BI tools), but the fundamental tasks remained relatively stable: collect data, analyze, synthesize, present.
GenAI breaks with this pace. Not because it invents a new type of work, but because it intervenes directly at the task level, with a speed and reach that traditional HR cycles cannot absorb.
The World Economic Forum makes this clear in the Future of Jobs Report: 39% of workers’ core competencies will change or become obsolete by 2030. And 63% of employers identify skills shortages as their primary obstacle to growth. These numbers do not speak to jobs disappearing. They speak to tasks shifting, competencies moving, work recomposing — often without the job title or description changing at all.
Where GenAI actually acts: the tasks
To understand GenAI’s impact, you have to look one level below. Not the job. The tasks.
A task is the elementary unit of work: writing an email, analyzing a dataset, preparing a presentation, conducting an interview, validating an invoice, responding to a support ticket. Jobs are aggregates of tasks. And it is these aggregates that HR systems have been built around.
But GenAI does not operate at the aggregate level. It operates at the task level. And depending on each task’s characteristics, its effect varies. Dell’Acqua et al. (2023)’s work on the “jagged technological frontier” illustrates this precisely: GenAI excels spectacularly at certain tasks and is counterproductive at others. The boundary between what the tool does well and what it does poorly is not straightforward.
We can distinguish at least three distinct effects.
Reduction. Some tasks are accelerated to the point where they cease to be significant time constraints. Drafting a first version, compiling data from multiple sources, reformatting documents, generating summaries: these tasks do not disappear, they compress. A task that took two hours now takes twenty minutes, with equivalent or better quality output.
Amplification. Other tasks see their scope expand. An HR analyst who could previously review ten candidate profiles per hour can now review fifty, provided they maintain quality judgment on each recommendation. Capacity increases, but responsibility for judgment remains intact. Brynjolfsson et al. (2023) document that this amplification effect is strongly differentiated by initial skill level: less experienced workers benefit more in relative terms, while experts retain advantages that GenAI does not fully compensate for.
Creation. GenAI generates new tasks that did not exist before. Formulating effective prompts. Validating AI outputs (which requires understanding what AI does and what it might miss). Detecting hallucinations in specialized business contexts. Governing tool use within the team. These new tasks require competencies that job descriptions have not yet named.
This triptych — reduction, amplification, creation — occurs in the same job, often in the same week. And it produces a phenomenon that research is beginning to document: an employee can see overall performance increase thanks to GenAI while experiencing increased cognitive load related to validating and supervising outputs. Shao et al. (2025), in a daily longitudinal study published in the Journal of Management, show this empirically: on any given day, the same employee can benefit from substantial cognitive gains from AI AND suffer from information overload that degrades post-work recovery. Both effects coexist, on the same person, at the same time.
Standard adoption metrics (usage rates, time saved, tickets resolved) do not capture this duality. They measure the gain. They do not measure the invisible cost.
Competencies become more unstable, more contextual, and more hybrid
If tasks change at the speed of GenAI, the competencies needed to accomplish them change too.
GenAI accelerates an already-underway phenomenon: competencies become less stable, more contextual, and more hybrid.
Less stable. The useful lifespan of a specific technical competency shortens. Mastering a specific software counted for a lot ten years ago. Today, what matters more is the ability to quickly learn a new tool, understand its limitations, and integrate it into an existing workflow. Gobeil-Proulx (2021) already identified this in the Quebec context: competencies that endure are AI literacy, ability to work in interdisciplinary teams, creativity and complex problem-solving — not fixed technical competencies.
More contextual. The same competency can have vastly different value depending on organizational context. The ability to formulate effective prompts is useful everywhere, but its specific value depends on available tools, business processes, sector, and an organization’s regulatory constraints. In Quebec in 2024–2025, only 12.7% of enterprises used AI for production purposes, according to Statistics Quebec. This low figure does not mean AI competencies are unimportant: it means their value is very unevenly distributed across sectors and organizational maturity. Mehdi and Morissette (2024) refine this reading by estimating, through econometric analysis of the Canadian labour market, differentiated exposure levels across professional categories, with results that strongly nuance global projections.
More hybrid. GenAI values hybrid profiles — a compensation specialist who understands the implications of AI compensation algorithms, or a learning leader who knows how to design human content AND evaluate what an LLM can or cannot do in that context. These profiles do not fit neatly into a job family box.
What this demands of HRIS: moving from profile to workflow
This is where HRIS start to break down.
HRIS have been built to manage profiles. An employee has a job. That job has required competencies. The employee declares or acquires those competencies. We measure the gap. We prescribe training.
This model assumes work is relatively stable, that you can describe it once and manage it thereafter. It also assumes competencies are relatively stable attributes of individuals: you either have them or you do not, and you accumulate them progressively.
What organizations need is not a better competency profile. It is a capacity to dynamically read real work: understand which tasks are actually being accomplished, how they evolve, what competencies emerge as critical, and how those competencies distribute across the organization.
That’s a different kind of goal. It requires data beyond self-reporting, taxonomies that capture semantic relationships between competencies rather than simply classifying them, and governance that allows maintaining this reading in real-time.
Some HRIS vendors are advancing in this direction. Workday Skills Cloud, with its 47,000+ standardized competencies (Workday, 2022) and self-learning ontological approach, is probably the most documented example. The principle is this: instead of asking employees to fill out their profile, the system infers their probable competencies by analyzing their projects, training, assessments, and internal movements. It maps semantic relationships between competencies, similarly to how human neural networks function: multidimensional relationships rather than rigid hierarchy.
But an ontology is not a strategy. And a Skills Cloud deployed without governance, without a culture of competency sharing, and without real connection to development and mobility decisions, remains a sophisticated cataloging tool — not a transformation lever.
This will be the subject of Part 2.
Skills-based organization: a possible answer, not magic
The concept of skills-based organization (SBO) is not new. It has been circulating in HR circles for over a decade. What changes with GenAI is its urgency.
The core pitch of SBO is simple: if competencies are the unit of value in modern work — more than degrees, more than titles, more than tenure — then all talent management logic should be organized around them. Internal mobility, development, acquisition, workforce planning: all pivot on competency rather than job.
The data is encouraging. According to a 2024 Deloitte report synthesized by HR Grapevine, SBO organizations outperform peers across every measured indicator: talent placement (+107%), anticipating change (+57%), innovation (+52%), achieving objectives (+63%), and positive employee experience (+79%).
What is clear, however, is that SBO is an architectural response to work volatility: a way to build the organization so it can adapt faster than its own job descriptions. Facing GenAI, this adaptability has obvious value.
What is less clear is what a true SBO strategy actually demands: reliable data infrastructure, governance of competency signals, organizational culture that values mobility over retention, and rigor in measurement that exceeds adoption dashboards. Most organizations thinking they are doing SBO are actually doing competency cataloging with a nice name.
What if a list of competencies is not enough?
This first part has set the diagnosis: GenAI is shifting work at the task level, precisely where HRIS and job descriptions fall short. The logical question follows — if competencies are becoming the relevant unit of management, why do so few organizations manage to turn that into a real strategy?
That’s what we’ll explore in Part 2: why so many organizations have a skills catalog but so few have an actual skills-based approach — and what separates a taxonomy from an ontology.
References
Complementary references
Deloitte / HR Grapevine. (2024). The Clear Benefits of a Skills-Based Approach. HR Grapevine, 2024.
Jacob, S., Souissi, S., & Patenaude, N. (2022). Intelligence artificielle et transformation des métiers en gestion des ressources humaines. Chaire de recherche sur l’administration publique à l’ère numérique, Université Laval.
Statistics Quebec. (2024–2025). Adoption and use of artificial intelligence by Quebec enterprises in 2024 and 2025. Exploratory portrait. Government of Quebec. https://statistique.quebec.ca/fr/document/intelligence-artificielle-entreprises-quebec
World Economic Forum. (2025). Future of Jobs Report 2025. WEF.
Workday. (2022). How Workday Is Delivering Next-Generation Skills Technology at Scale. https://www.workday.com/content/blog/it/posts/2022/09/how-workday-delivering-next-generation-skills-technology-scale.html
Skillsoft. (n.d.). What’s a Skills Taxonomy (vs. Ontology)? And Why Having One Makes HR Easier. https://www.skillsoft.com/blog/whats-a-skills-taxonomy-vs-ontology-and-why-having-one-makes-hr-easier
Field interpretation
Observations on organizational practices and HRIS deployments in Quebec and Canada are drawn from GP-Nova’s field experience in its Workday consulting mandates. They do not constitute a systematic study and are not directly generalizable, but serve to illustrate and contextualize trends documented by research.