Contents
Much of what’s written on AI-driven job architecture predicts the same outcome: jobs dissolve, skills take their place, and the org chart becomes a relic.
The organizations doing the work are more cautious. In a recent analysis of 87 organizations running more than 28 skills strategies, Deloitte found that many keep their traditional structures of work and map skills onto them. It calls a clean job architecture foundational to every business outcome it studied.
That finding matters more now than it would have a few years ago, because an AI-driven job architecture is under pressure from two directions.
The EU Pay Transparency Directive, whose transposition deadline passed on 7 June 2026, depends on pay structures that are stable and defensible. Meanwhile, the skills inside roles keep changing.
Treating this as a choice between freezing the architecture and dissolving it misreads both pressures. A more workable approach has three parts:
- Keep the structure stable on purpose. Job families, levels, and pay ranges move only through documented reviews.
- Keep the skills mapped to it current. The skills and proficiency expectations attached to each role update as the work changes.
- Decide in advance what triggers a change. Written rules set how much skills change is enough to reopen a role’s level.
AI has real work to do in keeping the skills layer current. Deciding when the structure moves is a job for people.
Skills make an AI-driven job architecture work harder
Job architecture tells you how roles are organized. Skills tell you what those roles actually require today. The companies getting value from skills-based talent management are the ones connecting the two.
The shift is real, just not the version that was pitched
Interest isn’t in question. In a Workday survey of 2,300 business leaders, 55% said they’d already started moving to a skills-based talent model, and another 23% planned to start within a year. 81% said a skills-based approach increases an organization’s potential for growth.
What’s faded is the idea of throwing out jobs altogether. In Deloitte’s analysis, the organizations getting value usually start with a business outcome, such as agility or productivity, and work back to the skills practices that support it.
Deloitte describes mapping skills to jobs as a “translation layer” between the two. The all-or-nothing version stays rare. Deloitte also cites a Gartner poll of 80 HR leaders in which only 2% said their organizations had adopted skills-based approaches across all their processes.
Jobs set the structure, skills keep it honest
The two layers do different jobs, and organizations need both.
Jobs and levels carry the structure a company runs on:
- Pay. Salary ranges are often tied to job levels, and market benchmarking tends to follow the same structure.
- Org design. Reporting lines and headcount budgets are built around positions.
- Employment terms. Contracts and, where they apply, collective agreements often refer to job titles or grades.
- Pay reporting. The Pay Transparency Directive groups workers into categories, and most employers will take those categories from their job structure.
Skills tell you whether that structure still matches the work. They show which roles have changed, where people could move, and whether a level still means what it meant when it was defined.
The hard part is keeping them connected
Building the connection once is well documented. AIHR’s reference framework walks through it step by step: define roles, group them into families, level them, benchmark pay, and align titles.
Keeping it current is where most guidance goes quiet. When HR reviews a framework only every few years, it slowly drifts away from the work it describes, because skills keep shifting inside roles.
Pay transparency sharpens the problem. It makes the structure harder to change just as the skills underneath it keep moving.
Pay transparency makes the structure harder to move
The EU Pay Transparency Directive never uses the term “job architecture.” In practice, though, it’s hard to comply with without one.
What the directive asks employers to do
A handful of requirements do most of the work. All of them are in the directive text:
- Define work of equal value. Pay structures have to let employers assess whether workers are doing work of equal value, using objective, gender-neutral criteria. Those criteria “shall include skills, effort, responsibility and working conditions” (Article 4).
- Explain how pay is set. Workers must be able to see the criteria used to decide their pay, pay levels, and pay progression (Article 6). Member states can exempt employers with fewer than 50 workers from the progression part.
- Share pay data by category. Workers can ask for average pay levels, broken down by sex, for colleagues doing the same work or work of equal value (Article 7).
- Report gaps on a schedule. Employers with 250 or more workers report every year, starting by 7 June 2027. Those with 150 to 249 report every three years from the same date, and those with 100 to 149 from 7 June 2031 (Article 9).
- Fix gaps that can’t be explained. If the pay report shows a gap of 5% or more in any category, employers must justify it on objective, gender-neutral grounds or fix it within six months of submitting the report. If they don’t, they must carry out a joint pay assessment with workers’ representatives (Article 10).
Why it all comes back to an AI-driven job architecture
Every one of those obligations depends on the same thing: sorting workers into groups who do the same work or work of equal value.
For most employers, those groups come straight from the job architecture, meaning the job family, the level, and the criteria behind the level.
That has two consequences:
- Consistency matters more. If similar roles sit at different levels in different business units, the categories they produce are hard to defend, whether a worker asks or a gap shows up in the report.
- Every change needs a reason. Moving a role to a new level changes the comparison group for everyone in that role. Employers will need to show why they made the move, using objective, gender-neutral criteria.
Where things stand
The transposition deadline passed on 7 June 2026, but national rules are still catching up. In mid-April, L&E Global reported that no member state had fully transposed the directive.
Details will vary by country. The reliance on equal-value categories comes from the directive itself, so it applies everywhere. Nestor’s guide to preparing for the EU Pay Transparency Directive covers the wider compliance picture.
That stability is exactly what’s hard to hold onto, because the skills inside roles aren’t standing still.
The skills inside roles are moving anyway
Keeping a job architecture steady would be easy if the work inside each role stayed the same. It doesn’t.
In the World Economic Forum’s Future of Jobs Report 2025, employers expect 39% of workers’ core skills to change by 2030.
That’s down from 44% in the 2023 edition, and the report suggests the drop may reflect more companies investing in continuous learning. Even so, it means close to two in five core skills shifting within five years.
Upskilling in place is where drift comes from
For an AI-driven job architecture, the more useful figures show how employers expect to handle that change. For every 100 workers, they anticipate:
- 29 upskilled in their current roles
- 19 reskilled and moved into new roles internally
- 11 who will need training but won’t get it
- 41 who won’t need significant training
Redeployment is easy for an AI-driven job architecture to see. Someone moves to a new role, and the system records the move.
Upskilling in place is harder to spot. The title and level stay the same while the skills the role needs shift underneath them.
Over a few years, that adds up to drift. The level description still reads the same, but it no longer matches what people at that level actually do. This is the change skills data picks up and job titles miss.
Out-of-date skills data means an out-of-date equal-value assessment
Under the Pay Transparency Directive, drift is also a compliance issue.
Skills are one of the four criteria Article 4 lists for judging whether two jobs are of equal value. When the skills a role needs change significantly, its place in an equal-value category may need to be reassessed.
So a skills layer that goes stale eventually takes the equal-value assessment with it.
There’s a useful flip side. Current skills data gives employers something concrete to point to when they explain why roles sit where they do.
The directive also says that “relevant soft skills shall not be undervalued.” That’s a reason to describe skills like communication or stakeholder management as carefully as technical ones. Nestor’s guide to building a soft skills taxonomy is a good starting point.
The structure needs to stay stable, and the skills layer needs to keep moving. The practical question is how to run both.
Keep the structure stable and let the skills layer move
The way through is to stop treating job architecture as one thing. It works better as three layers, each changing at its own pace and owned by different people:
- Structure. Job families, levels, leveling criteria, and pay ranges. It changes only through scheduled, documented reviews, owned by HR and Total Rewards, with worker representatives involved where they exist.
- Skills. The skills and proficiency levels mapped to each role, plus employee skills profiles. It changes continuously as the work changes. Managers validate it, HR maintains the taxonomy, and AI suggests updates.
- Review triggers. The rules for when skills changes justify reopening a level. HR and Total Rewards leadership set them in advance and revisit them occasionally.
Most organizations already have some version of the first two layers. The third is usually missing, and it’s what keeps the other two connected.
Triggers stop the layers drifting apart
Without written triggers, the setup can go wrong in two opposite directions:
- Too rigid. Skills changes pile up until the next full rebuild, and levels fall out of step with the work.
- Too loose. Levels get reopened whenever a manager pushes for it, and categories stop being consistent.
Triggers give the structure a clear, defensible reason to move. Depending on how you level roles, a trigger might look for:
- Shifts in what your leveling method measures. A role’s required skills change in the areas your method scores, such as scope, complexity, or accountability.
- New skills with no home. Skills appear across a job family that no current level description covers.
- Mismatched moves. People move between levels without comparable skills profiles.
The thresholds are each organization’s call. Write them down, because under pay transparency they become part of the explanation for any category change.
Start with the roles under most pressure
Few organizations can build all three layers across the whole company at once. Josh Bersin argues for starting skills work with specific business problems rather than an enterprise-wide rebuild, and the same logic applies here.
Two good places to start:
- Job families where the work is changing fastest, often the ones most exposed to AI.
- Job families most likely to face questions once pay-gap reporting begins.
If reliable skills data doesn’t exist yet, a focused skills audit is a more realistic first step than a full taxonomy. Nestor’s guide to running a skills audit in 90 days shows how to scope one.
Of the three layers, skills is the one that needs constant attention, and that’s where AI earns its place.
Where AI helps, and where it shouldn’t make the call
Keeping skills current across thousands of roles and employees is a big part of why job architecture gets rebuilt every few years instead of maintained. It’s also the work AI is best suited to.
What AI can take on
In the skills layer, AI can handle work that would otherwise need a large team:
- Inferring skills from role data, job descriptions, and work history.
- Mapping skills to roles across the whole job architecture, not just the families someone had time to review.
- Keeping profiles current as people take on new work, projects, or training.
- Flagging mismatches when a role’s skills no longer fit its level description, so the case can go through your review triggers.
This is the part that scales with software rather than headcount. Without it, the skills layer only gets updated when someone finds the time.
What should stay with people
The structure layer is different. Employers need to be able to explain every decision about levels, equal-value categories, and pay to an employee, a works council, or a regulator.
The EU AI Act points the same way. It classifies AI systems used “to make decisions affecting terms of work-related relationships,” along with promotion decisions, as high-risk (Annex III, point 4).
In practice, the split looks like this:
- AI suggests. It proposes skills updates and flags roles that may need a review.
- People decide. HR and Total Rewards confirm whether a level or category changes.
- The reason is recorded. Each decision is documented against your objective, gender-neutral criteria.
That keeps AI doing the heavy lifting, and keeps the decisions employees care most about in human hands.
How Nestor keeps the skills layer current
Nestor helps with the two layers that take the most work to build and maintain, and leaves the decisions to your team.
For the structure and skills layers, Nestor covers:
- Job architecture. Nestor’s AI helps build AI-driven job architectures, so the structure and the skills attached to it are set up together rather than stitched together later.
- Skills mapped to roles. Skills are mapped to the roles in the architecture, which gives each level a clear picture of what the work requires.
- Dynamic skills profiles. Employee skills profiles are kept up to date inside the platform, not rebuilt for each review.
- Gaps and development. It identifies capability gaps and recommends development actions, which matters most for people being upskilled in their current roles.
- Your existing systems. The platform integrates with the HR and business systems you already use.
On the decision side, Nestor provides governed, role-based access to its AI-generated recommendations, workforce insights, and reports. Human oversight stays central to consequential decisions about employees, which is where level and pay decisions belong.
You can see how skills mapping works on the AI Skills Management page, and how Nestor approaches the directive on its pay transparency page.
The architecture that lasts moves on purpose
The question AI raised about job architecture was whether it would survive. The evidence so far says it will, and pay transparency makes it more central, not less.
The more useful question is how much of it should move, how often, and who decides.
Organizations that settle those choices in advance are in a stronger position. They have:
- a structure that changes only for documented reasons
- a skills layer that stays current as the work changes
- written triggers that connect the two
When a worker asks why their role sits where it does, or a pay-gap report raises a question, they can answer with current data and a clear rationale.
Without that setup, the risk is that one of the two layers is out of date at exactly the moment someone asks.
Frequently asked questions about AI-Driven Job Architecture
What is job architecture?
Job architecture is the structure an organization uses to group its roles into job families and levels, with titles, pay ranges, and career paths attached. It gives HR a consistent way to compare roles, set pay, and show employees how they can progress.
What is AI-driven job architecture?
AI-driven job architecture is a job structure where AI helps build and maintain the skills each role requires, while people keep control of job families, levels, and pay ranges. AI maps skills to roles and keeps skills profiles current. Changes to the structure itself go through documented reviews.
What is the difference between job architecture and job leveling?
Job leveling is one part of job architecture and job architecture is the whole structure of job families, levels, and roles. Job leveling is the process of deciding which level a role belongs to, based on criteria such as scope, complexity, and accountability.
What is the difference between job architecture and a skills framework?
Job architecture is the structure of job families, levels, and roles. A skills framework is the skills taxonomy plus proficiency definitions, role mappings, and governance. The skills framework describes what each role in the job architecture requires, and the two change at different speeds.
Will skills-based organizations replace job architecture?
Skills-based organizations are not replacing job architecture. Many are adding skills to their existing job structures instead. In Deloitte’s April 2026 research, the organizations getting value from skills typically map them onto existing jobs. In a Gartner poll of 80 HR leaders, only 2% said skills-based approaches ran across all their processes.
How often should job architecture be updated?
The structure (families, levels, and pay ranges) should change through scheduled, documented reviews. Organizations should update the skills layer continuously as the work changes. Between reviews, written triggers decide when skills changes are big enough to reopen a level.
What are job architecture review triggers?
Review triggers are written rules that decide when changes in a role’s skills justify reviewing its level or category. Examples include shifts in what the leveling method scores, new skills that no level description covers, and moves between levels without comparable skills profiles. Under pay transparency, triggers also document why a change was made.
Does the EU Pay Transparency Directive require a job architecture?
The EU Pay Transparency Directive does not require an AI-driven job architecture by name, but most employers will need one to comply. It requires pay structures that assess work of equal value using objective, gender-neutral criteria, along with pay information and reporting by categories of workers. Those categories usually come from job families and levels.
What counts as work of equal value under the EU Pay Transparency Directive?
Under Article 4 of Directive (EU) 2023/970, employers assess work of equal value using objective, gender-neutral criteria that include skills, effort, responsibility, and working conditions. Employers should agree these criteria with workers’ representatives where they exist, and must not undervalue relevant soft skills.
When does EU pay transparency reporting start?
Employers with 250 or more workers must report annually, starting by 7 June 2027. Employers with 150 to 249 workers report every three years from 7 June 2027, and those with 100 to 149 workers every three years from 7 June 2031. National laws set the exact requirements in each country.
Can AI decide job levels or pay grades?
AI can suggest changes, but decisions about job levels and pay grades should stay with people. The EU AI Act classifies AI systems used to make decisions affecting terms of work-related relationships or promotion as high-risk, and the Pay Transparency Directive requires objective, gender-neutral criteria. In practice, AI flags and proposes, and people decide and record why.
