On September 1, workforce intelligence firm TalentNeuron published The Great Reallocation, a study of how seven large enterprises have moved headcount over two years: Salesforce, Klarna, Wells Fargo, Google, Microsoft, Citi, BT Group.
One number in it is worth stopping on.
In 2026 TalentNeuron re-ran a Fortune 100 manufacturer’s already-drawn elimination list at task granularity. 34% of the flagged roles contained tasks requiring human judgment, and those tasks were central to the transformation strategy the cuts were supposed to fund. Executed as drawn, the list would have removed the capability the company had just committed to building.
The layer the line gets drawn on
The useful part is not 34% as a value. It is what 34% exposes about how cut lists get made.
Almost every reduction decision happens at the role layer. Someone pulls a job-title inventory out of the HR system, leadership draws lines through categories, a whole title goes red. It is fast, the savings are calculable, and legal can follow the paper. The cost is the assumption underneath: that a title is one indivisible thing you either keep or don’t.
No job is one thing. TalentNeuron’s other headline finding puts it plainly: no employee is 100% replaceable by AI and no job is fully automatable. A job is a bundle of tasks, and those tasks automate at wildly different rates. Some are gone this year. Some still need a person in a decade. They sit under the same title.
Erzsébet Malzenicky, global head of workforce strategy and transformation at Experian, went further: a task map cannot be a one-off exercise, because every quarter something automates, shifts, or disappears. The test of a work architecture, she said, is not how precise it is on day one but how easily it can be redesigned on day two hundred.
Name the source of the number
A pause is required here.
TalentNeuron sells task-level workforce analytics. “Role-level cutting is wrong, cut at the task level” is also its product pitch. The 34% has no third-party audit, the client is unnamed, and the methodology is not published.
So this is not an industry census. It is one vendor applying its own method to one of its own clients and reporting the result. We are writing it anyway, because the mechanism it points at is checkable and it lines up with what we have already reported. Treat it as a direction, not a percentage.
We covered the MyPerfectResume survey in May: 52% of 1,000 hiring managers already use AI to generate the productivity data that decides who gets cut, and 51% of those managers call the output fair (52% of hiring managers now use AI to generate the data that decides who gets cut). Put the two together and the picture closes. The lists are increasingly machine-generated, and the layer the machine works on is the role, not the task.
We also covered the finding that 87% of Americans want a human to sign off on AI-involved layoffs (87% of Americans want a human to approve AI layoffs). That read as a fairness demand at the time. TalentNeuron’s number hands it a cost argument instead: skip the review and the company pays for it too.
The roles growing are the ones that measure this
The second set of numbers runs the opposite direction from the first.
All seven companies are spending heavily on AI. Across them, combined demand for HR, strategic workforce planning, and people analytics rose 16% over two years. Broken out: strategic workforce planning up 33%, people analytics up 26%, learning and development up 42% — and L&D demand nearly doubled at Microsoft, Google, and Citi.
Separately, TalentNeuron counted 114,419 global job postings requiring core AI skills across 103 occupations. The total is not the interesting part. The 103 is. AI skill requirements have spilled well past the jobs that build software.
David Green of Insight222 supplied the mechanism: AI is compressing planning cycles from years to quarters, sometimes to weeks. Shorten the cycle and you run the keep-or-cut judgment far more often — which makes the people who run that judgment scarcer, not more expendable.
What this means for you
If you are on a list, or think you are about to be. That list was almost certainly drawn at the title layer. By this report’s own accounting, roughly a third of those lines are drawn wrong, and the failure mode is sweeping in roles whose judgment tasks the company still needs. Anyone who can decompose their own work into tasks and point at which ones are judgment-bearing and which connect directly to the stated strategy walks into the review round holding something. That is not spin. It is doing the task map the executive team skipped.
If you are choosing where to go next. The report is not subtle about direction: strategic workforce planning, people analytics, L&D. These functions are growing because somebody has to measure where automation actually stops. Measure it well and the cuts land right and the AI spend pays back. This is a job category the AI rollout is generating, not one it is sparing.
If you are not in HR. The 103 occupations is your signal. AI skills are becoming a standard line in job descriptions rather than a technical-role specialty. We saw the leading edge of this curve at the org-design layer in June (Talent leaders are putting AI agents on the org chart). It has now dropped into hiring requirements.
One closing note. The job losses are real, and we have tracked them for a year (117 days, 100,000 tech layoffs). What this report adds is that a meaningful share of those losses do not trace to the boundary of what AI can do. They trace to companies drawing the line at the wrong resolution.
Sources
- TalentNeuron Research Finds AI Is Driving Workforce Redesign, Not Solely Workforce Reduction — PR Newswire, September 1, 2026
