Your Employer Wants You to Use AI. What Should Be Clear First?

“Could you use AI to get this done faster?”

It sounds like a straightforward request. You have a report to prepare, a deadline approaching and a tool that might help.

But several questions sit behind it. Which tool can you use? Can you upload the source material? Who checks the result? And if you find that doing the task yourself would be more reliable, is that an acceptable decision?

These are the conversations we need around AI at work.

I think professionals should have opportunities to understand and test useful technology. But an expectation to work differently should come with an explanation of what good work now looks like. Otherwise, employees are left to work out the rules while managers wait for the promised improvement.

“Use AI” can mean several different things

Consider a hypothetical business analyst preparing requirements for a customer service system.

Their manager suggests using AI to produce the first draft. The analyst has workshop notes, customer examples and internal process documents. They could ask a tool to organise the notes, identify unanswered questions or draft user stories.

Those are potentially useful activities. They also involve different decisions about information, accuracy and responsibility.

Organising fictional sample notes is different from uploading a document containing customer details. Suggesting questions is different from declaring that a requirement has been agreed. Drafting a user story is different from confirming that a development team can deliver it.

Before discussing whether employees should use AI, we need to be precise about the work involved.

A useful instruction describes the task, the approved approach and the expected checks. “Use AI wherever possible” leaves too much to interpretation.

Is AI use mandatory for employees?

There are two questions here: whether your organisation expects you to use a particular tool, and whether that expectation is enforceable in your circumstances.

An employer may introduce AI into a workflow or make training part of a role. That does not, by itself, settle every employee’s obligations. The answer can depend on local law, employment terms, workplace agreements, the nature of the instruction and individual circumstances.

This article cannot determine whether a particular instruction is lawful or whether someone can refuse it.

For example, the UK House of Commons Library’s briefing on AI and employment law explains how existing areas of law can constrain workplace AI use. A new technology does not sit outside the wider employment relationship. Readers elsewhere need the equivalent guidance for their jurisdiction.

The practical starting point is to clarify the expectation. Is the employer asking staff to learn a tool, try it on suitable tasks or use it in a defined process? Those are different requirements.

If an instruction creates a concern, raise the specific issue: access, accuracy, confidentiality, accessibility or a conflict with another policy. Seek clarification from your manager, HR team or employee representative. A dispute about an instruction may require local advice.

Employees should not have to invent the boundaries

Return to our analyst. Their immediate concern might be simple: can the workshop notes go into the tool?

A general instruction to be careful is insufficient. They need to know which product and account are approved, what information is permitted and whether any material requires additional authorisation.

Australia’s privacy regulator, the Office of the Australian Information Commissioner, recommends due diligence when selecting commercial AI products. As a best-practice measure, it advises organisations against entering personal—and especially sensitive—information into publicly available generative AI tools. That recommendation arises in an Australian regulatory context; employees elsewhere must also follow their applicable rules and organisational requirements.

The wider lesson is useful wherever you work: access to a tool is not evidence that every document may be uploaded to it.

Nor should employees assume that removing names makes material safe. Internal plans, commercial terms and identifiable circumstances may still require protection.

When permission is unclear, ask before uploading. An employer should provide a workable route for getting that answer.

What managers need to put in place

I would expect a team introducing AI at work to be able to answer six questions:

• Which tools and accounts are approved?
• Which tasks are permitted, restricted or excluded?
• What information may employees enter?
• What checks are required before using or sharing an output?
• Who makes the final decision or approves the work?
• Where should staff report errors, concerns or unintended disclosures?

These are the beginnings of AI governance: decisions about authority, boundaries and oversight. Governance becomes useful when people can apply it during an ordinary working day.

Employees also need training suited to their tasks. A demonstration of how to generate text does not necessarily prepare an analyst to validate generated SQL, or a manager to assess an AI-produced performance summary.

Managers should make room for discussion before changing routines. Acas guidance on consultation explains the value of talking and listening to employees about changes affecting them, while distinguishing good-practice consultation from circumstances where consultation is legally required. Its legal context is British, but listening to the people doing the work is a sound management practice elsewhere too.

A colleague questioning a proposed use may have identified a problem worth investigating.

Employees still need to understand what they submit

Suppose the analyst receives ten well-written user stories from an AI tool. Each follows the expected format and includes acceptance criteria.

That is a draft, not evidence that the requirements are correct.

The analyst still needs to compare the stories with the source material, confirm assumptions with stakeholders and identify anything the tool has added without support. A plausible requirement can create work that nobody requested.

I would be uncomfortable submitting a requirement I could not explain simply because its wording looked professional.

This is where individual responsibility matters. Be honest about what you checked and what remains uncertain. Follow the organisation’s requirements for recording or disclosing AI assistance. Do not present an inferred statement as an approved decision.

Managers have responsibilities too. They need to provide time, guidance and access to the people who can resolve uncertainties. “The employee should have checked” is a weak response if the workflow made proper checking impractical.

What AI at work means for an analyst’s career

For business analysts, data analysts and other technology professionals, AI offers useful opportunities to experiment.

An analyst might use an approved tool to suggest interview questions, critique a draft, explain unfamiliar code or generate test scenarios from a specification. The value depends on the task and the quality of the review.

My concern is that a polished output can make it easy to skip the learning underneath it. If a tool writes a query, can you explain its joins and filters? If it drafts acceptance criteria, can you recognise the missing exception?

Professional growth requires practice at those judgments.

The International Labour Organization’s 2026 briefing on AI exposure makes an important distinction: estimates of what AI could do are not predictions of what will happen to jobs. Actual outcomes depend on adoption choices, economic conditions and how work changes.

For an individual professional, I would focus on learning the tools alongside the subject knowledge needed to question them. Keep practising the difficult parts of your role.

Faster needs a fair test

Once AI is introduced, there may be pressure to expect more output. Before changing targets, examine the complete task.

Include preparation, review, corrections and exception handling. Consider whether employees have comparable access, training and suitable equipment. An approach tested by one experienced colleague may require different support for a new starter.

Our hypothetical analyst might produce a first draft sooner but spend longer checking unsupported assumptions. Another task might deliver a clear saving. Both findings deserve an honest review.

The test should be whether the process produces dependable work with a worthwhile use of resources.

If you are asked to use AI at work tomorrow, start the conversation with three questions: What are we trying to improve? What are the boundaries? How will we know it helped?

Those questions give employees and managers something better than an instruction to experiment and hope.

If your team needs to turn an AI idea into a clear workflow, contact me to discuss the requirements, responsibilities and measures of success.