Workers Want AI to Manage Tasks—but Not Deliver Their Reviews
September 12, 2026 at 01:09 PM
|Author: QUASA Editorial Team|5 min read|2
A UK workplace survey drew fresh attention on September 8 after finding a sharp divide between expected efficiency and preferred human oversight. According to <a href="https://www.techradar.com/pro/a-third-of-workers-believe-having-an-ai-boss-would-make-them-more-productive” rel=”nofollow noopener” target=”_blank”>TechRadar’s September 8 report, 36.7% of 600 full-time UK employees thought an AI manager would improve their productivity, while 74.2% would rather receive performance feedback from a human manager.
The findings suggest a defensible boundary for workplace AI: software may assist with routine coordination, but consequential feedback should remain under accountable human control. That boundary is partly an editorial interpretation, however, because the poll measured expectations and preferences—not actual productivity or support for automating each individual management task.
The survey records expectations, not measured output
The Careerminds UK research release, dated August 19, says Pollfish surveyed 600 employed UK adults in May 2026. It found that 12.5% anticipated a significant productivity increase under an AI manager and 24.2% a slight increase, while 37.8% expected no change and 25.5% expected productivity to decline.
No respondents were placed under an AI manager, and the research did not compare their completed work, quality or output with a control group. The headline productivity figure therefore describes what employees think might happen. It does not demonstrate that an AI manager makes people more productive.
The term “AI manager” also leaves room for different interpretations. A respondent might have imagined software supporting a human supervisor, a system issuing instructions independently or something between those models. Without a tested implementation, the results cannot establish which design employees expected or which one could produce the anticipated gains.
The workable boundary separates coordination from judgement
Softonic’s September 8 account identified task allocation, scheduling and performance tracking as areas where software-led management could help, although it also incorrectly described 38.5% as the share of managers expecting no gain. The underlying survey did not ask respondents to approve those three duties separately, so they should be treated as proposed uses rather than direct poll results.
- Coordination suitable for AI assistance: preparing schedules, distributing routine assignments, monitoring deadlines and assembling status records.
- Decisions requiring human ownership: interpreting performance evidence, delivering reviews, deciding promotions and handling development conversations.
This matrix does not make routine coordination consequence-free. Task allocation can affect workloads and access to valuable assignments, while tracking systems can favour activity that is easy to count. Human managers would still need to review exceptions, examine the quality of the indicators and take responsibility when automated recommendations affect an employee.
Nor does retaining a person for the final conversation automatically amount to meaningful oversight. If software has already selected the evidence, assigned its importance and determined the outcome, a manager who merely communicates that conclusion has little authority. Human ownership requires the ability to question, change and explain a decision.
Managers are less optimistic than directors—but 38.5% still expect gains
The primary breakdown shows that 64% of directors and more senior respondents expected a productivity increase, compared with 42.6% of senior employees and 38.5% of managers. The manager figure represents respondents expecting a gain, not respondents expecting no improvement.
Directors were therefore 25.5 percentage points more likely than managers to anticipate higher productivity. That establishes a difference in attitudes by seniority, but it does not explain its cause. The possibility that directors focus more on efficiency while managers focus on employees’ daily needs remains an interpretation rather than a causal finding from the poll.
The discrepancy in the secondary coverage is important because it changes the meaning of the result. Managers were less optimistic than directors, but the available detailed figures do not support the stronger claim that 38.5% of managers expected no benefit. The primary table does not publish a combined no-gain percentage for managers.
Performance feedback is the clearest human limit
The feedback question produced a more direct result than the broad productivity question. Respondents chose between receiving performance feedback from an AI system or from their line manager: 74.2% selected the manager, 14.2% selected AI and 11.7% expressed no preference.
A performance discussion can involve disputed evidence, circumstances missing from activity records and work that cannot be reduced to a simple metric. It also gives an employee an identifiable decision-maker who can answer questions and be challenged. Those features make feedback materially different from generating a rota or compiling a progress report.
The survey did not ask separate questions about promotions, pay, disciplinary action or development conversations. It would therefore go beyond the evidence to claim that UK employees rejected every use of AI in consequential employment decisions. What the results establish is narrower: nearly three-quarters preferred a person for the performance-feedback duty they were actually asked about.
For now, the evidence supports a provisional division of responsibility rather than a verdict on AI management. Automated tools can be considered for coordination under human supervision, while evaluative conversations have the clearest mandate for accountable human control. Evidence from real workplace deployments would be needed to determine whether AI-supported management raises output and whether employees can genuinely contest the decisions it shapes.
- U.S. Startups Can Drop BOI Filing—but Not Their Local Compliance Map
- Remote Workers Can Qualify for FMLA—Their Home Is Not the Worksite
- Unstructured Data Management for Your Business: Nothing But Benefits
