AI was supposed to make work easier to manage. A new survey says it is making it harder to judge. According to TechCrunch's report on a Nielsen survey of 1,700 directors and senior managers across seven countries, commissioned by work-management company Monday.com, more than half of senior leaders say it is harder to evaluate employees' work after adopting AI, and nearly half spend more time reviewing and guiding their teams' output than before.
This is survey data on AI at work, which means it measures what managers say they experience, not output per hour. It is still a useful snapshot of the "verification tax" we have written about, now at the management layer. See our earlier analysis of AI verification tax and rework surveys and of shadow AI at work.
TL;DR
| Question | Answer |
|---|---|
| Who ran it? | Nielsen, commissioned by Monday.com |
| Sample | 1,700 directors and senior managers, seven countries |
| Evaluating work | More than 50% find it harder after AI adoption |
| Review time | Nearly half spend more time reviewing and guiding output |
| Changed evaluation | 78% changed how they evaluate employees |
| AI in decisions | 46% use AI to inform decisions; 39% of leaders use AI to draft performance reviews |
| Tools | 57% say more tools in circulation; about 51% say tools are more interdependent; only 29% call AI infrastructure fully optimized and centralized |
| Agents | 58% expect to oversee more AI agents than humans within a year; 46% worry about losing control; 35% fear being held responsible |
| Enthusiasm | 94% excited to improve work using AI |
Why work is harder to judge
TechCrunch explains the mechanism: much of employees' output now arrives polished in the manager's inbox, making it hard to tell how much came from the employee and how much from AI. A clean memo or code review no longer signals effort, understanding or skill. For managers who evaluate people by output quality, the signal has degraded.
Monday.com's head of market research, Michal Miller Levi, told TechCrunch that AI works well for managers individually, but that their role is also to drive the organization, and the struggle comes from AI moving faster than organizational processes. She likened the mood to the early internet: 94% of the managers polled are excited to improve work using AI.
Changing what gets rewarded
The survey says 78% of respondents changed how they evaluate employees, giving more weight to prompting, deciding which tasks to delegate to AI, and experimenting with new ways of working. That is a significant shift in what counts as good performance, and it raises a fairness question: how can employees be judged fairly when managers cannot easily see how much AI produced?
AI evaluating people
The survey found that 46% of managers use AI to help inform their decisions, and 39% of leaders use AI to draft performance reviews. TechCrunch notes the risk: LLMs are prediction machines that tend to repeat patterns in training data. It cites research from the Luxembourg Institute of Socio-Economic Research finding that, absent clear objective standards and when individuals are rated independently, LLMs tend to rate performance similarly to humans, and a paper from early 2025 finding AI models can be biased against Black male candidates. Levi told TechCrunch that companies have yet to work out how to handle such issues.
For managers, the practical read is that AI drafts of performance reviews should be treated as drafts: they can sound objective while encoding bias, and the human who signs a review is accountable for it.
Invisible labor and tool sprawl
The report suggests supervising AI is a new layer of "invisible labor": someone has to decide which tasks to hand to AI agents, check what they produce, and step in when something goes wrong.
The tool picture is messy:
- 57% of managers say their organizations now have more tools in circulation than before.
- About 51% say their tools have grown more interdependent and clustered since adopting AI.
- Only 29% describe their AI infrastructure as fully optimized and centralized.
- Levi said managers expected to consolidate tools in 2025 and instead tool counts grew, with more apps, integrations and shadow AI usage as employees adopt tools outside approved systems.
The agent horizon
The survey looks ahead too. 58% of managers expect to oversee more AI agents than human employees within the next year; 46% worry about losing control over those agents' actions; and 35% fear being held responsible for decisions made by AI. Levi said managers do not want agents to be fully autonomous, and want people consulted on crucial tasks and decisions.
That matches the control problems documented elsewhere. Anthropic's study of how coding-agent monitors fail (our summary) and the new federal guidance on rogue agent incidents (NIST, NSA and CISA's ten practices) both treat oversight as the hard part. Our guide to when to let an agent run covers approval design.
A rubric for judging AI-assisted work
When polish is free, evaluation has to move to things AI cannot fake as easily. The table below is our suggestion, not part of the survey.
| Instead of judging... | Judge... | How to check |
|---|---|---|
| Writing quality | Whether the claims are right and sourced | Ask for the two or three sources behind the main claim and spot-check them |
| Speed of delivery | Whether the right problem was chosen | Ask what alternatives were considered and why they were rejected |
| Code that passes tests | Whether the person can explain and modify it | A short walkthrough or a small live change |
| Volume of output | Whether output was verified before sharing | Ask what was tested, by whom, and what failed |
| Prompting skill alone | Whether the person knows when not to use AI | Ask for an example of a task they did by hand and why |
None of these checks require detecting AI use. They shift attention from the artifact to the person's understanding and judgment, which is what a review is meant to assess anyway.
What employees can do
The survey is about managers, but the incentives cut both ways. Employees whose work is increasingly assessed on how well they use AI can protect themselves and their teams by keeping a short note of what they asked AI to do, what they checked, and what they changed. That record makes review faster and gives you evidence of your own contribution if a result is questioned. If your company has no policy on AI use in performance evaluation, ask for one. The survey suggests many do not yet have a worked-out answer, with Monday.com's own researcher saying companies have yet to work out how to handle bias in AI-informed judgments.
How to read this survey
| Strength | Weakness |
|---|---|
| Large sample of 1,700 senior managers across seven countries | Commissioned by a vendor selling work-management software, which benefits from a "governance gap" story |
| Nielsen is an established research firm | The full questionnaire and weighting were not published in TechCrunch's article |
| Specific percentages on tools, reviews and agents | Self-reported perceptions, not measured time or output |
| Matches other surveys on verification and rework | "Harder to evaluate" does not show total productivity fell |
Nothing in the article says AI made teams slower overall. It says managers feel more burden at the review stage, which is consistent with productivity gains that show up upstream and costs that show up downstream.
Why the "excited but overwhelmed" mix is not a contradiction
The survey's headline pairing, 94% excited and more than half struggling to evaluate, is easy to read as inconsistent. It is not. Individual productivity gains are visible and immediate: a manager drafts a plan in minutes. Organizational costs are diffuse and delayed: more tools to govern, more output to review, unclear authorship, and new accountability for agent actions. People report the first as excitement and the second as friction. That is why governance, not capability, is the bottleneck the article points to.
What to do about it
- Set a disclosure norm. Ask staff to note where AI helped on significant deliverables so reviewers know what to check. This is not about blame; it tells reviewers where the risk is.
- Judge outcomes and reasoning. Ask for the decision, evidence and trade-offs, not only a polished document. Oral walkthroughs expose understanding quickly.
- Keep humans accountable for reviews. Use AI to organize notes, not to assign ratings, and audit AI-assisted reviews for consistent wording across groups.
- Consolidate and approve tools. Publish an approved list, a request path for new tools, and a policy for data. Shadow AI is how data leaves the company.
- Design approval points for agents. Decide which actions require a human before they run, log agent actions, and assign an owner to each agent. A security layer such as AgentBeam, from the explainx.ai team, is built to stop agents before they take dangerous actions.
- Budget review time. If AI output increases, review capacity must too, or quality falls silently.
Related reading
- AI verification tax: rework surveys and how to cut it
- Shadow AI in the workplace: privacy risk guide
- Human in the loop: when to let an agent run
- UK employers cut entry-level jobs: Work Foundation survey
- Barclays 5,000 job cuts and the AI question
- NIST, NSA and CISA ten practices for rogue agents
- Source: TechCrunch: AI promises productivity, but for managers it's mostly a headache · Monday.com
Figures are from TechCrunch's October 21, 2026 report on the survey; the full dataset was not available to us.