# Good work needs more than visible AI use

New worker surveys distinguish pressure to use AI from the conditions for useful adoption. Keep learning expectations explicit, then judge performance through the work and its consequences.

**AyEye — Workforce Management · 2026-11-19**

An employee can make a sound decision about a tool and still worry that the decision looks unambitious.

That is the useful concern in new research from LSE and Protiviti. Nearly half of respondents worry about being judged for not using enough AI. Alongside Adecco's new findings on worker involvement, it suggests a small change to performance management: distinguish learning to use AI from demonstrating good work.

## Visible use and useful work are different measures

The [LSE/Protiviti report](https://www.protiviti.com/sites/default/files/2026-11/lse-generational-surveyreport-bklt-1026-iz-en-protiviti.pdf), released on 17 November, surveyed 3,001 professional employees from March to May 2026. It found 48% worried about being judged for insufficient AI use, compared with 20% worried about being judged for using it.

Its statistical comparisons did not yield a stable link between perceived job performance and AI adoption, training, employer access or workflow integration. That last measure is how employees believe their manager rates them. It is not an independent assessment of output, quality or actual appraisal records. The finding cannot establish that AI has no performance benefit.

This is observational research, including regression comparisons, not an experiment assigning employees to different management practices. Its broader programme includes repeat respondents, but that does not make every reported association causal. Protiviti sells consulting services; the research partnership and its interests belong beside the results.

## Support matters before the score

[Adecco's report, released today](https://www.adeccogroup.com/en/our-group/media/press-releases/global-workforce-of-the-future-study-2026), draws on an online survey of 40,500 workers across 36 countries and 21 industries in June and July. Among workers involved in AI-related work redesign, 96% said they were willing to adapt, compared with 29% among those not involved.

Those groups may differ in leadership, resources, confidence and many other ways. The comparison does not show that participation alone caused willingness. Adecco is a workforce-services supplier, and this is self-reported experience rather than a test of one redesign method.

The two surveys cover different populations and questions. Their percentages should not be pooled. Together they give a reason to inspect what an organisation rewards: developing informed judgement, or displaying the expected tool habit.

## A worked case: two advisers, one standard

Imagine an employee-service team whose objective says, “Demonstrate regular AI use to improve productivity.” One adviser uses an approved assistant to prepare routine case summaries. Another handles a sensitive exception and chooses the authorised manual route because the available record is incomplete. Both complete their cases accurately.

A usage count makes the first adviser look more committed. It cannot explain whether either approach produced the better service. Nor can it reveal the correction, checking or escalation work behind the result.

The manager can separate two conversations. A learning expectation asks employees to practise with approved tools, understand their limits and explain a choice. A performance standard asks for accurate resolution, appropriate confidentiality, timely service and sound escalation. An employee can meet the learning expectation and still choose less automation for a particular case.

This does not exempt anyone from a validated procedure. If an approved workflow requires a particular tool for a defined purpose, the employee should follow it or use its exception route. Equally, low usage may reveal a training need, inaccessible software or a genuine missed opportunity. It is a prompt for inquiry, not a conclusion about effort.

## The counter-case: adoption can be a legitimate objective

New skills require practice. An employer may reasonably expect staff to complete training and demonstrate competence before deciding whether a tool suits their work. Removing every adoption expectation could protect comfortable habits and prevent useful learning.

The distinction is between a temporary development objective and an unexamined proxy for job performance. Define the capability to be learned, provide access and practice, then examine its contribution to work. If usage predicts better outcomes in a particular validated process, use that evidence within its scope. Neither survey supplies that validation for an arbitrary team.

## Revisited: the saved hour has acquired an audience

On [11 November](https://www.lecxie.com/publications/ayeye-workforce/2026-11-11.html), Workforce asked who decides what a claimed saving is for. On [9 October](https://www.lecxie.com/publications/ayeye-workforce/2026-10-09.html), it examined the context lost when human work becomes a score. The new extension concerns the signal before the score: employees' belief that visible AI use itself demonstrates performance.

The earlier discipline remains: examine the whole workflow and let people explain the evidence. Add one question to the existing review: could an employee make a justified choice about AI use without that choice being mistaken for poor commitment? The proposed response needs local evaluation; the surveys do not prove it improves fairness or productivity.

## IF I WERE TO DO ONE THING NOW

### Review one AI-related performance objective

Take one current objective that rewards AI use. With the employee and manager, distinguish the learning capability it requires from the work outcome it is meant to improve. Record what evidence would demonstrate each, including how a justified exception will be assessed. Keep this within the existing review process; test one objective before expanding the approach.

## Sources and limits

This concise edition covers developments since the last Workforce publication on 11 November, with primary-source checks through 19 November 2026. LSE/Protiviti's methods and performance discussion and Adecco's release and survey methodology were checked. No causal effect, universal performance rule or legal right is inferred. The service-team example and proposed objective review are LecXie analysis.

Chiappe × OpenAI.

Canonical: https://www.lecxie.com/publications/ayeye-workforce/2026-11-19.html
