A study published September 24, 2026 in Frontiers in Psychology matched self-reports from 642 hybrid and mostly remote workers across two surveys seven weeks apart, from 742 first-wave respondents. It found an association between self-reported reliance on AI delegation and concern about delegated work outside working hours. The observational design does not establish causation or objectively measured skill loss. [1]
An AI investment can change a job before anyone updates its description. Drafting may become reviewing. Preparing a recommendation may become checking one. An employee's responsibility for the result may remain even as the work required to produce it changes.
If HR joins only when the training invitation is ready, several important choices may already have been made: the expected workload, the staffing assumption, and what managers will count as good performance.
IBM's September 21 workforce study provides complementary context. Its announcement reports that 46% of organizations in its research do not involve the chief human resources officer when defining AI strategy. The research, conducted with Oxford Economics, surveyed 1,500 workforce leaders and 8,800 employees in two surveys between April and June 2026. [2]
These findings raise questions about work design. They do not establish that excluding HR causes an AI project to fail, or that AI has objectively reduced employees' abilities. Leaders should ask whether the people responsible for jobs and performance are helping design the change before the business commits to it.
The recommendations below apply those concerns to investment approval and work design. They are proposed management practices, not interventions tested by these studies.
Give HR a decision to help make
An invitation to an AI steering committee is a start. A defined responsibility makes that participation useful. Ask the HR leader to work with the business sponsor on the proposed changes to roles, workload, development, and performance expectations before the investment is approved. This applies IBM's reported work-redesign concerns to an investment meeting; its survey does not prove this meeting design improves results. [2]
The business sponsor should own the intended customer or operational outcome. Technology leaders should establish what the system can support. Finance should challenge the economics. HR should help determine whether the proposed organization can perform the work and whether management practices reinforce the behavior the project requires.
Culture Amp's September 17 benchmark provides another perspective: employees reported stronger encouragement to experiment with AI than clarity about its purpose. Participation was voluntary, so these responses should not be generalized to all employers. Culture Amp sells employee-experience software. [3]
For each affected role, explain the business purpose of the change and give a concrete example of acceptable work. Ask employees to describe what they understand, then use the gaps to improve the rollout plan.
For a smaller company, this responsibility may sit with an owner or people leader rather than a CHRO. The title matters less than having someone accountable for connecting the technology decision to the employee's actual job.
Record unresolved workforce assumptions alongside technical ones. If the proposal depends on employees recognizing subtle errors, identify who has that expertise and how the organization will know that the review is working.
Write the new job before setting the new target
Consider a hypothetical business services company introducing an AI assistant to prepare client proposals. Management expects account managers to handle more opportunities because initial drafting takes less time.
The account manager's new work includes checking the client's requirements, confirming delivery commitments, assessing whether the proposed scope is feasible, and resolving inconsistencies. A polished document still needs a person who understands what the company can deliver.
HR Dive's September 23 coverage highlights respondents' concerns about additional, easily overlooked work associated with AI, including verification and correction. That reporting draws on the same IBM study; it is not a separate measurement of the effect. [4]
Before increasing the target, ask account managers to describe the changed tasks during a limited pilot. Update the role expectations and workload estimate with what they learn. Make responsibility for checking commitments explicit, and identify which decisions require delivery or commercial approval. The proposal responds to verification work described in the IBM announcement and HR Dive coverage, rather than an independently tested staffing formula. [2][4]
A manager who wants careful review should allow time for it. Otherwise, the employee is being asked to meet two potentially conflicting expectations: process more proposals and investigate anything that looks uncertain.
Make the right to challenge usable
Write down what an employee should do when an AI recommendation appears wrong. The instruction should identify who can pause the work, who resolves the question, and how a customer deadline will be handled while the issue is open. This proposed operating rule responds to the accountability and challenge concerns reported by IBM; no causal benefit is established. [2]
In the proposal example, an account manager might discover that the assistant included a service the company cannot provide. The manager needs a workable route to correct the proposal and alert the delivery owner. A general instruction to use judgment leaves too much unspecified.
Define when review is expected and who takes over when an employee finishes work. In the proposal example, an unresolved commitment should have a named next owner. This is a proposed way to clarify responsibilities, not a tested remedy for the concerns described in the September research.
Review performance expectations at the same time. If a careful employee raises a valid concern, that action should count as sound work even when it delays completion. Supervisors need guidance on how to distinguish useful scrutiny from avoidable delay.
Do not turn the number of overrides into a target. Frequent disagreement could reflect a weak system, unclear instructions, or appropriate caution. Review representative cases and the reasons behind the decisions before drawing conclusions about the employee.
Protect how people learn the changed work
Campus Technology's September 21 account distinguishes acquiring new skills from maintaining existing ones as work changes. It also separates different survey measures of judgment and AI oversight that can look interchangeable in a headline. [5]
The practical implication is to examine how employees will learn the judgment their revised roles require. This is a proposed management response, not a demonstrated treatment for skill erosion.
For the proposal team, experienced colleagues could review selected cases with newer account managers. Discuss why a commitment is feasible, what evidence supports the recommendation, and what would change the decision. Use completed examples with appropriate handling of client information. IBM's featured practitioner perspectives and the distinction in Campus Technology provide context for this proposed learning arrangement; neither tests this hypothetical team. [2][5]
Schedule that learning as work. Assigning oversight to employees without giving them access to experienced colleagues may create responsibility without adequate preparation. Include supervisors in the development plan because they will need to evaluate reasoning as well as finished output.
Bring a workforce decision back to the investment review
Before expanding the pilot, ask the business sponsor and HR leader to return with a short account of how the jobs have changed. It should identify the responsibilities added or removed, the authority employees need, the time required for review and learning, and the performance expectations that need revision.
Pair that account with evidence from the work: accepted proposal quality, avoidable corrections, employee explanations of difficult cases, and the effect on customer commitments. Interpret the findings together rather than treating a faster first draft as proof that the new role is ready.
Then decide whether to expand, change the staffing or support arrangement, narrow the AI task, or continue the pilot. HR's contribution is to make the workforce assumptions visible while those choices are still open.
The September 24 research offers a useful starting point for revisiting these decisions. Bring the workforce leader into the proposal early enough to shape the work employees will actually be expected to do.
Questions for executives
- Who is responsible for the workforce assumptions in our next AI investment proposal?
- Can employees pause questionable work, obtain a decision, and still meet realistic performance expectations?
- What evidence will show that employees and supervisors are ready for their changed responsibilities?
Sources and further reading
- When AI works, employees worry: the vigilance cost of AI delegation in hybrid workYifan Zhang, Tianyou Li, and Chunming Chen, Frontiers in Psychology · Published 2026-09-24
- New IBM CHRO Study: AI Puts Critical Thinking at the Center of Workforce PrioritiesIBM · Published 2026-09-21
- Culture Amp’s 2026 AI at Work Benchmark: 85% Are Encouraged to Use AI. 42% Don’t Know Why.Culture Amp · Published 2026-09-17
- IBM report warns AI could erode human skills deemed vital by CHROsRyan Golden, HR Dive · Published 2026-09-23
- Report: AI May Be Eroding the Very Skills Employers Need MostDavid Ramel, Campus Technology · Published 2026-09-21