New notes on skills, managers and pay transparency.New notes from the team.Read the latest
All insights

Skills became infrastructure. Judgement didn't.

In October 2026 the two largest HR suites moved skills and task data into the core of their platforms. Inferred skills are now a commodity. The evidence says the bottleneck was never the taxonomy; it was what anyone did with it.

Note
Perspective № 19
Published
Reading time
4 min read
Filed under
Skills, Industry

A head of talent opens her HR system on a Monday in 2027 and finds that every employee now has a skills profile, inferred from their job history, their training records and the tickets they closed. The vendor announcement called it a foundation for agents. She asks the obvious question: which of the fourteen open senior roles can we fill from inside? The system lists matches. Nobody can say what “match” means, who set the threshold, or whether the mobility policy's rule about internal candidates was applied.

What changed in October

On 6 October 2026 a major HR suite announced it would acquire a Belgian skills-inference company, describing its skills and work graph as “an excellent grounding layer for agent queries regarding work and skills planning”. A week later another suite showed task-level “work intelligence” that decomposes jobs into tasks and maps them to skills, and launched agents it calls autopilots. Skills data stops being a product you buy beside the HR system and becomes part of the system.

That is good news for anyone who has maintained a job catalogue by hand. It also settles a question the market spent five years on: inferred skills are infrastructure now, like org structures and cost centres. Nobody will win on having them.

What the evidence says skills data did so far

The most careful study of skills-based hiring, from the Burning Glass Institute and Harvard Business School in February 2024, followed 11,300 roles at large firms from 2014 to 2023. Only 3.6% of roles dropped a degree requirement. Those that did raised non-degree hiring by about 3.5 percentage points. Net effect across the economy: 0.14% of hiring, roughly 97,000 workers out of 77 million hires, “not even 1 in 700”. Where firms meant it, the results were real: non-degree hires at the leading firms stayed longer (58% two-year retention against 48%) and earned about 25% more. The gap between the announcement and the practice was the finding.

Skills-based hiring, 11,300 roles, 2014 to 2023

  • Roles that dropped a degree requirement3.6%
  • Share of all hiring that changed0.14%
  • Two-year retention, non-degree hires at committed firmsagainst 48%58%
Burning Glass Institute and Harvard Business School, February 2024. Fewer than one hire in 700.

The data behind the profiles is not in better shape. A May 2026 survey of 512 HR leaders found that only 20.9% update skills data continuously, that 90% describe their skills data as complete or accurate, and that 76% of those same leaders admit less than three-quarters of skills are captured. A third had failed to identify internal talent for a critical role. A March 2026 survey of 1,500 workers found 90% of managers but 69% of employees believed skills were understood, and only 18% of organisations used a central skills database at all.

A respected analyst wrote in June 2025 that the skills-based organisation “was a strong concept, but it proved too static”, and in October 2026 that a long list of tasks “may be out of date in a few months”. Both are right. A profile is a snapshot; a decision is made on a date.

What a skill in the record cannot do

  1. It cannot say what counts as a match. “Has the certification” is a fact. “Every required certification and the minimum years for the role” is a rule someone wrote, and it differs by site.
  2. It cannot apply your policy. The mobility policy's clause on internal candidates lives in a document, not in a taxonomy.
  3. It cannot carry a threshold. “Thin bench” and “critical role” each have a number, and the number is in an HR director's head until it is written down and signed.
  4. It cannot be accountable. A list of matches has no owner. A decision does.

Where the work moves

If the suites now supply the facts, the work that remains is the judgement: the rules a company actually runs on, written with the people who hold them, tested on real cases, signed, and applied with the source shown. That is what we build in Atelier, and it is deliberately indifferent to where the skills come from. A Coworker reads whatever skills your system holds, applies the rule your HR director signed, and shows which record and which clause produced each line of the answer.

The useful question to ask any vendor in 2027 is therefore not “do you have skills intelligence?” Everyone will. It is: whose rule decided this, when was it signed, and can I see the record it was applied to?

Sources

  1. Sigelman, Fuller and Martin, “Skills-Based Hiring: The Long Road from Pronouncements to Practice”, Burning Glass Institute and Harvard Business School, February 2024
  2. Betterworks, 2026 Talent Intelligence Survey, May 2026
  3. TalentLMS, Skills Visibility Report, March 2026
  4. Josh Bersin, “What is task or work intelligence?”, 19 October 2026

See PeopleHub on one of your own workflows.

Forty-five minutes, and a clear account of what PeopleHub can and cannot do for it.

Book a working session