A job title describes what someone was hired to do. A skills taxonomy describes what they can actually do right now. Most companies are still managing a workforce that has outgrown its titles, using documents that were accurate the day they were written and increasingly wrong every day after.
This was the second question that came up in the chat during our CPO Roundtable session with HiBob, right behind the governance question. Someone asked how people are handling fluid role definitions against traditional job descriptions. It is the right question, because the honest answer is that most job descriptions stopped being accurate a while ago and nobody updated the paperwork.
A job description is a static document. Someone wrote it once, probably during a hiring cycle, and it sits in the ATS getting reused for the next opening in that role. It describes a title, a set of duties, and a handful of qualifications.
A skills taxonomy is a structured, living map of the actual capabilities your people have and the ones your work requires. It is not tied to a title. A skills taxonomy lets you see, at any point, who can do what, where the gaps sit, and how close someone is to being ready for a different role. A job description answers “what was this person hired for.” A taxonomy answers “what can this person actually do.”
Because the work is changing faster than the paperwork. Deloitte's research on skills-based organizations found that 63 percent of the work people currently do falls outside their core job description. That number was already high before generative AI started reshaping individual tasks inside almost every role.
LinkedIn's Workplace Learning Report puts a number on the pace: the skill sets required for a given job have shifted by roughly 25 percent over the past eight years, and that rate of change is expected to double by 2027. A document written once cannot track a target moving that fast. Even Deloitte has stopped trying. The firm restructured its own internal career progression this year, replacing the analyst-consultant-manager title ladder with skill families, a direct response to how thoroughly AI is reshaping the actual work inside professional services.
It sounds like a real capability, but it doesn't help you evaluate anyone. Ask a manager whether an employee has “AI literacy” and you will get a shrug or a guess. Ask whether that employee evaluates AI output for factual accuracy before acting on it, and you get something you can actually observe in a work sample.
That shift, from broad competency language to specific observable behavior, is the entire point of a good taxonomy. HiBob's AI Skills Framework, built on a survey of 1,200 AI decision-makers, organizes AI capability into three skill sets, seven competencies, and 29 observable behaviors. Nowhere in that structure does “AI literacy” stand alone as something you check off. It gets broken into behaviors a manager can actually watch for and score.
Three layers, going from broad to specific. This is the same shape HiBob used, and it is worth borrowing because it forces you to keep going past the abstraction until you land on something observable.
|
Layer |
What it covers |
Example |
|
Skill set |
The broad domain of capability |
Individual AI Usage |
|
Competency |
A specific capability within that domain |
Evaluating and Improving Output Quality |
|
Observable behavior |
A concrete, assessable action tied to real work |
Proactively reviews AI output for factual accuracy before acting on it |
A flat list of skill names skips the middle and bottom rows. It gives you a checklist, not a taxonomy. The value is in the third column: the thing you can actually see someone do.
Do not start with the whole organization. Start with two or three role families where AI is changing the work fastest, usually recruiting, customer-facing roles, or anything content- or analysis-heavy. Trying to taxonomize every job in the company at once is how these projects stall out and never ship.
Four steps for each role family:
First, pick the role family based on where the work has visibly changed in the last year, not where it might change someday. Second, define behaviors, not abstractions, using the same test as above: could a manager watch someone do this and score it. Third, validate the draft with the actual managers and practitioners in that role family before it goes anywhere near a hiring process or a performance form. Fourth, pilot it in one process, usually hiring or performance, before you try to run it everywhere at once.
If you have not already run a skills gap analysis, that is the natural first move. It tells you where the gaps already are, which tells you which role families to taxonomize first.
Four places, and none of them work well without it.
Hiring is the most immediate. Skills-based hiring only works if you know which skills actually matter for a role, and a taxonomy is what tells you that instead of a list of assumed qualifications. Performance is the second: you can evaluate someone against what they can actually do rather than a static list of duties nobody has updated since they started.
Internal mobility depends on it entirely. You cannot move someone into a new role based on their old title. You can move them based on a mapped set of capabilities that already overlaps with what the new role needs. And workforce planning, which we covered in more depth in Workforce Planning in an AI-First Environment, becomes a capability exercise instead of a headcount exercise once you can see what your people can do rather than just what they are called.
If this is the kind of structural work your organization needs help building, that is exactly where our Org Design and Workforce Planning practice starts: with the skills map, not the org chart.
A structured map of the capabilities your workforce actually has, organized by skill set, competency, and observable behavior, rather than a list of job titles and duties.
A skills taxonomy is the underlying structure. A skills-based organization is what you get when you use that structure to run hiring, performance, mobility, and planning instead of running them off job titles.
No. Start with two or three role families where AI is changing the work fastest, prove the model works, then expand it.
An observable behavior is something a manager can actually watch someone do and score, such as reviewing AI output for accuracy before acting on it. A competency label like “AI literacy” describes a category, not an action.
Most organizations can draft and validate a taxonomy for a single role family in four to six weeks if they start narrow and involve the actual managers in that role family early.