
Nobody told her the job had changed.
Her title was the same. Her grade was the same. Her objectives looked almost identical to the ones in last year’s review.
But the work had changed.
She was a recruitment coordinator. She used to read every CV that came in. Now, an AI tool shortlisted candidates before she saw them.
Her role was no longer primarily about reading applications. It was about checking the tool’s judgement.
That is a different job.
Nobody had updated the job description. Nobody had clarified which decisions belonged to her, which belonged to the tool and which needed to be shared.
So she did what many people do when responsibility changes without being discussed.
She created her own safety net.
At lunchtime, she quietly re-read the rejected applications. Not because she distrusted the technology in principle, but because she knew who would be asked if a strong candidate had been filtered out.
Her manager saw the slower turnaround and called it resistance.
It was not resistance.
It was an unmarked boundary between human judgement and machine judgement.
Most organisations still talk about AI adoption as if it means introducing a new system.
Choose the tool. Train the users. Measure usage. Move on.
But the deeper change is not the arrival of the tool. It is the redistribution of responsibility around it.
A recruitment coordinator using AI may now be expected to:
That is not the same role with a faster inbox.
It is a role with a new decision boundary. For a practical framework, see how to define AI-alone, human-alone and shared work.
The problem is that organisations often change the workflow without changing the surrounding system. The job description remains static. The performance objectives remain static. The training remains focused on how to use the tool, rather than when to trust it.
The employee is left to work out the rest.
When nobody defines the new boundaries, people compensate in private.
They double-check outputs.
They keep parallel records.
They ask a colleague to review the tool’s recommendation.
They repeat work the system was supposed to remove.
They avoid using the tool for decisions that feel too risky.
They use it officially, then rely on their own process underneath.
This hidden work is easy to misread.
A manager sees slower turnaround, inconsistent tool usage, repeated checking, low confidence and reluctance to automate.
The employee experiences unclear accountability, fear of missing something important, no agreed escalation route, uncertainty about what good judgement looks like and responsibility without authority.
The result is often described as a people problem.
But the real issue is usually a design problem. This is also why senior staff may work around new AI systems when ownership has not been clarified.
The organisation has changed the workflow without designing the role around it.
The useful question is not, “Are people using the AI tool?”
It is:
“What decisions are people now making because the AI tool is part of the workflow?”
That question reveals the work that the implementation plan may have missed.
For each role affected by AI, leaders need to understand:
Without these answers, the organisation has automation, but not accountability.
A practical way to redesign a role is to divide its responsibilities into three categories.
These are tasks the system can complete without routine human review, such as sorting applications by clearly defined criteria, summarising documents, identifying missing information, producing a first draft or classifying routine requests.
The important question is not whether AI can perform the task. It is whether the consequences of an error are low enough for the task to run without human intervention.
Some decisions should remain with a person because they require accountability, context or ethical judgement.
Examples include making a final hiring recommendation, deciding whether an exception is justified, handling a sensitive employee situation, interpreting competing organisational priorities or taking responsibility for a decision that affects someone’s future.
Human-alone does not mean technology is irrelevant. The tool may provide information, but the authority remains with the person.
This is where many AI-enabled roles actually sit.
The tool may identify a pattern, recommend an action or produce an initial assessment. The human then checks, interprets, challenges or improves the result.
Shared work needs the clearest boundaries because it is where responsibility can become blurred.
The person needs to know what they are expected to check, what they are allowed to override, what evidence they should use, when they must escalate, how much review is enough and who owns the outcome.
Without this clarity, people either over-check everything or trust the tool too much.
Neither response is sustainable.
Traditional job descriptions are built around stable responsibilities. They describe what someone is expected to do, who they report to and what experience they need.
AI-enabled work requires something more dynamic. It also needs to describe which tasks are supported by AI, which decisions require human judgement, what level of oversight is expected, where accountability sits, what happens when the system is wrong and how the role is expected to develop as the tool changes.
This does not necessarily mean creating a separate AI policy for every job.
It means making the changed work visible.
If a role now includes reviewing automated recommendations, that should be reflected in its objectives. If the employee is accountable for the quality of an AI-assisted decision, they need the authority and training to challenge the system.
Otherwise, the organisation has created responsibility without giving the person a clear mandate.
When these boundaries remain unclear, the costs appear in places leaders may not immediately connect to AI adoption.
They show up as delays that look like inefficiency, workarounds that look like non-compliance, low adoption that looks like resistance, stress that looks like a capability gap and inconsistent decisions that look like poor performance.
The organisation may then respond by buying another tool, running another training session or setting a higher adoption target.
But more pressure rarely solves unclear accountability.
People do not need to be told to trust the system. They need to know what trust means in their role.
Start with one role that has visibly changed.
Do not begin with a broad statement about the future of work. Begin with the actual workflow.
Map the role as it is performed today, not as it appears on paper.
Then identify what has been removed, what has been added, what has become faster, what has become more difficult, what now requires checking, what decisions have moved closer to the employee, what decisions have moved closer to the system and what nobody has formally claimed.
Finally, assign an owner to each part of the work.
Not just an owner for the tool.
An owner for the decision.
That is the distinction many AI programmes miss. For another example of how unclear ownership creates hidden work, read where shadow AI accountability breaks down.
The recruitment coordinator’s job changed before anyone held a meeting about it.
That is how AI-enabled work is arriving in many organisations. Not through a formal restructure, but through small changes in workflow that accumulate until the role is no longer what it was.
The employee notices first.
They feel the new risk. They create the workaround. They carry the uncertainty.
By the time the organisation notices, the new way of working may already be embedded.
The answer is not to eliminate human judgement or to ask people to trust AI blindly.
The answer is to make the division of work explicit.
Clarify what AI can do alone.
Clarify what people must own.
Clarify where the two need to work together.
Then give each part an accountable owner.
When the line is drawn, the lunchtime re-reads can stop. The employee knows what is hers to check, what she can let go and what needs to be escalated.
That is not resistance.
That is responsible AI adoption.
Elvorta helps leadership teams map which parts of a role are AI-alone, human-alone and shared, then assign clear ownership for each part.
If roles in your organisation have changed this year but the paperwork has not, request a 30-minute clarity call.
You may not have an adoption problem.
You may have a role-design problem.