“Everyone should use AI in their work.” The instruction sounds decisive. It leaves the employee with a surprisingly difficult question: what, exactly, counts as doing it well?
Mandatory AI use at work can establish an expectation. It cannot, by itself, establish an AI-ready workforce. People still need suitable tasks, an authorised way to work, relevant practice and a clear response when the tool is inappropriate.
This does not mean that every requirement is misguided or that voluntary experimentation is always enough. It means that leaders should distinguish requiring people to develop a capability from requiring AI to appear in every piece of work.
The distinction becomes practical when someone has a good reason not to use it.
What are employees actually being required to do?
An AI mandate can contain several different instructions.
An employer might require staff to complete an introduction, practise with an approved tool, assess a relevant opportunity or follow an established assisted process. Alternatively, it might expect visible daily use without specifying the work that use should improve.
Those are different commitments. Completing a course does not demonstrate readiness for an unfamiliar task. Using a tool frequently does not establish that its output belongs in a final document. Assessing an opportunity can reasonably lead to a decision not to proceed.
Before discussing resistance, make the expectation explicit.
Is the employee expected to learn, investigate, use or deliver something? Under which conditions? What support is available? Who can answer a question about an unsuitable case?
A requirement that cannot answer those questions transfers much of the design work to the people expected to comply.
What current evidence contributes
Two sources help explain why the surrounding conditions deserve attention.
The UK government's Skills for AI upskilling executive summary, updated on 27 July 2026, draws on workshops, case studies and an employer survey. It identifies practical barriers including limited time, staffing pressures, cost and unclear provision. It recommends learning connected to actual tasks, including understanding when AI should not be used. This is mixed practical evidence, not a controlled comparison of mandatory and voluntary adoption.
Gallup's 6 October 2026 analysis, based on the American Job Quality Study, reports differences in AI use across jobs and groups. It notes that roles differ in digital access and obvious opportunities for direct use. The survey ran from 26 January to 24 March 2026; the full employee sample was 15,482 US employees. Reported effects are employees' assessments, and associations do not establish causation.
Neither source proves that mandates always fail. Together, they provide reasons to examine opportunity and support before interpreting non-use as unwillingness.
The approach below is an editorial proposal for doing that, illustrated through a fictional workplace.
Consider the weekly maintenance briefing
Imagine a facilities-management team preparing a weekly briefing for its supervisors. The document brings together completed work, outstanding requests and information that needs clarification.
A manager introduces an approved AI assistant and asks everyone to use it when preparing the briefing.
For one coordinator, that could mean organising already checked, non-sensitive notes into an agreed structure. Another coordinator has incomplete records and needs to speak with colleagues before drafting anything. A third has a short, straightforward update that is quicker to write directly.
If the instruction simply requires an AI-produced draft, all three employees face the same visible expectation despite having different work in front of them.
The second coordinator might produce polished prose before resolving the missing information. The third might add an unnecessary step. Neither possibility demonstrates that the assistant is useless. It shows that the instruction has not explained how to choose its role.
This is a fictional planning example, not a reported client deployment or advice to automate maintenance or safety decisions. Its purpose is to examine the requirement.
Replace the blanket instruction with a bounded expectation
A more useful instruction would distinguish preparation from ordinary use.
During preparation, the team could be required to practise on a synthetic briefing using the approved environment. The exercise would ask employees to compare the draft with its inputs, identify missing information and explain the changes they would make.
The manager would arrange suitable time and someone to answer questions. Practice material would avoid real confidential records.
That requirement asks people to engage with the tool and demonstrate an approach. It does not require them to put an unready output into circulation.
For ordinary work, the expectation could specify which part of briefing preparation the assistant may support and what remains necessary before a document is shared. It should also describe what to do when the conditions do not fit.
An employee who cannot resolve an inconsistency should know where to raise it. An employee who decides that a particular briefing does not need AI should be able to explain that decision against the agreed purpose.
These are proposed design choices, not a standard policy or a guarantee of improved performance. A real organisation would need to adapt them to its information, responsibilities and working arrangements.
The improvement is precision: employees can understand what is expected without guessing that more AI involvement is always better.
Ask what sits behind non-use
Suppose a coordinator is not using the assistant after the introductory session. Several explanations remain possible.
They may not have access when they need it. They may understand the tool but not the approved boundaries. They may need help with a particular step. Their current assignments may fall outside the intended use. They may also be avoiding a reasonable learning requirement.
A manager should establish which situation applies before choosing a response.
Ask the employee to describe a recent task and the point at which the expected approach became difficult or seemed unsuitable. If access was missing, resolve access. If the requirement was unclear, clarify it. If a skill is missing, arrange relevant practice and feedback.
If the task does not fit, examine the task rather than sending the employee back through the same general introduction.
This does not remove expectations or prevent managers from addressing a failure to engage. It makes the discussion more specific. Decisions about employment policy and consequences require the organisation's appropriate processes and expertise.
A broad label such as “resistant to AI” does little to identify the next useful action.
Make justified non-use explainable
A workforce prepared to work with AI should also be able to recognise when using it would be inappropriate.
That is a demanding expectation. It requires more than a preference for the old method, but it also requires a manager to listen to a substantive objection.
In the fictional briefing team, an employee might explain that the available notes do not establish whether a request has been completed. Generating a fluent summary would not resolve that uncertainty. The appropriate next step is to obtain the missing information.
Another employee might explain that the proposed use falls outside the agreed information boundary. That question needs clarification through the established route.
The manager can ask what the employee observed, which condition is not met and what they propose doing instead. The answer provides something to investigate.
A useful exception route should identify who responds and how the employee should proceed while waiting. Otherwise, the employee may be left choosing between an instruction to use AI and another instruction they cannot reconcile with it.
Managers should also distinguish a recurring mismatch from a one-off case. Repeated justified exceptions may indicate that the requirement needs revision.
Do not make the learning stage a performance demonstration
Practice is a place to discover gaps. Normal work has delivery obligations. Combining them without explanation can leave employees unsure what a mistake will mean.
For the fictional team, an initial synthetic exercise might reveal that someone accepts an invented completion date because it appears plausible. That is a useful learning need to identify before real briefing preparation.
The response should address the specific difficulty: tracing statements back to the supplied material and recognising what the material does not establish.
Managers need to explain the purpose and conditions of the exercise accurately. They should not promise confidentiality or immunity they cannot provide. They can make clear what will be reviewed, by whom and what support follows.
They should also avoid assuming that confidence in a demonstration transfers immediately to ordinary work. A later task may involve unfamiliar inputs or time pressure.
The broader human learning advantage depends on developing understanding that can be applied when conditions change. A mandatory practice session can contribute to that development; attendance alone does not establish it.
What can a reasonable requirement accomplish?
Requirements can serve useful purposes.
An organisation may need everyone in a relevant role to understand an approved process, complete essential preparation or use a supported route when specified conditions apply. Leaving those matters entirely to individual preference can create its own difficulties.
The question is whether the requirement is sufficiently defined and supported.
For the briefing team, a reasonable expectation might be that each coordinator can examine a draft, find unsupported statements and explain whether the proposed use fits the task. That is something a supervisor can discuss using concrete work.
If the organisation later establishes an assisted process for a particular class of briefings, the expectation can become more specific. Employees still need a route for exceptions and changes.
There is no universal requirement in this article for every role or sector. Some tasks may not offer a suitable direct use. Some deployments need specialist assessment well beyond a manager's local experiment.
Clear expectations and professional judgment need to work together.
Review the instruction as the work changes
An instruction can become outdated even when the tool remains available.
In the fictional team, a revised briefing format might require information the earlier process did not cover. A new group of employees might have different starting knowledge. The team might discover that a recurring part of the work needs a different approach.
Someone should review whether the expectation still makes sense.
Ask for examples of where it helped, where it created an extra step and where staff could not apply it. Look at the work and its constraints alongside employees' explanations.
The purpose is to decide whether to retain, narrow, revise or stop the requirement. Continued use should not be the only acceptable conclusion of a review intended to test suitability.
Likewise, changing an instruction in response to evidence need not mean abandoning the wider adoption programme. It can mean making its scope more accurate.
Bring the actual instruction into the leadership room
A corporate event can make this issue concrete without turning it into a debate between AI enthusiasts and sceptics.
Give participants a fictional instruction such as “Use the assistant for every weekly briefing.” Ask them to rewrite it so an employee knows what to practise, what to deliver and what to do when the task does not fit.
Then ask a different group to interpret it. Would the employee and manager make the same decision?
This is a proposed discussion exercise, not an evaluated intervention. A keynote can frame the distinction between an expectation and a capability. A working session can examine the organisation's wording and unresolved choices.
Our guide to designing a corporate AI learning summit explores how to connect such conversations with workplace follow-through.
Require a capability you can describe
Sources were checked on 28 October 2026. The research reviewed here does not establish the causal effect of AI mandates, and the fictional scenario is not evidence of a completed Learning Coach engagement.
The practical argument is narrower: a direction to use AI leaves essential work to be done. Leaders need to specify a meaningful expectation, provide conditions for learning and explain how justified exceptions will be handled.
Before issuing the next instruction, ask: What should people be able to do—and could a well-informed decision not to use AI demonstrate that capability?
Bring that question to your next leadership or workplace learning event. Explore Learning Coach's international keynotes and request a tailored proposal.
References
Skills for AI: Executive summary—What works for AI upskilling in the UK. GOV.UK, updated 27 July 2026. Mixed practical evidence and recommendations; not an experiment comparing mandates.
Jones, J. M., AI Benefits at Work Unevenly Distributed. Gallup, 6 October 2026. American Job Quality Study; US employee self-reports and stated methodological limits.
