When AI changes work: conflict, meaning and the limits of a good business case
Change creates conflict because it changes meaning, not just tasks.
Leaders often introduce AI with a clear case: automate routine work, improve decisions, serve customers faster and free people for higher-value work.
That case can be sound. But employees hear something else as well. They ask: Will my expertise still matter? Will I have less autonomy? Will my role become smaller, more monitored or less secure? Who is accountable when AI gets it wrong?
Klarna’s AI customer-service assistant delivered impressive early results: faster resolution times, lower costs and the equivalent of hundreds of full-time roles replaced. But the company later acknowledged it had reduced human capacity too aggressively. Complex, ambiguous and emotionally charged cases still needed people.
Deutsche Telekom has taken a broader approach. It has trained more than 100,000 employees and says staff will set goals, review results and retain responsibility for decisions. That is a more credible starting point than simply promising efficiency.
Lufthansa is more direct. It links digitalisation and AI to efficiency gains, while planning around 4,000 job cuts by 2030. Employees are not being asked simply to adopt a tool. They are being asked to accept a change that may affect their employment.
In New Zealand, One NZ frames AI as a “superpower” while acknowledging that some areas will become leaner. In Australia, Bank of Queensland’s Capgemini partnership was presented as access to AI and business-processing capability, but affected staff experienced it as offshoring and job loss.
The lesson is not that AI change should be avoided. It is that leaders need to translate the business case into the realities of people’s work. Name what will improve and what will be lost. Be clear about accountability, retraining and who decides. Treat objections as information, not resistance.
Organisations often introduce AI with a clear logic. Routine work can be automated. Decisions can be supported by better information. Customers can be served faster. People can be freed from repetitive tasks to focus on judgement, relationships and improvement. That logic is not wrong. But it is incomplete.
AI change is not only a technology decision. It changes what people do, how their contribution is valued, who has authority and what counts as good work. That is why it can create conflict even when the business case is sound and leaders believe they have explained the change clearly.
The conflict is rarely simply about whether people accept AI. It is about meaning. What does AI mean for a person’s expertise, status, autonomy and future? What does “efficiency” mean to a leader trying to control cost, and to an employee whose role may become smaller, more monitored or less secure? Recent examples from New Zealand, Australia and Europe show that these questions cannot be resolved by better presentations alone.
The same words, different worlds
Hans-Georg Gadamer argued that people interpret the world from within a particular horizon, shaped by their history, experience and practical concerns. In an organisation, that horizon includes a person’s role, professional background, previous experience of change and relationship with those making the decision. So when a leader says AI will create “better use of capability”, people hear different things.
A manager may hear access to faster analysis and better customer service. A specialist may hear that the work which gives their role meaning can now be done by a machine. A frontline employee may hear closer monitoring. A finance director may hear a necessary response to cost pressure.
None of these responses is irrational. Each is a reasonable interpretation from where that person sits.
Ludwig Wittgenstein’s point was similar, though more linguistic. Meaning comes from use. Words such as “efficiency”, “augmentation”, “productivity” and “responsible AI” do not carry a single meaning. They mean what they mean in the practices, incentives and relationships surrounding them.
This is why repeating the message rarely resolves conflict. Leaders need to find out how the change is being interpreted, not simply repeat what they intended to say.
Klarna: efficiency measured too narrowly
Klarna, the Swedish financial technology company, provides one of the clearest recent examples of AI changing an operating model.
In 2024, Klarna deployed an OpenAI-based AI assistant across a large share of customer service. The company reported that the assistant handled 2.3 million conversations in its first month, about two-thirds of customer-service contacts, equivalent to the work of around 700 full-time agents. Average resolution time reportedly fell from 11 minutes to under two minutes.
The initial message was about speed, availability and cost. Klarna reported annual savings of about US$40 million from replacing around 700 full-time contractor roles.
But the company later acknowledged that it had cut human capacity too aggressively. AI handled straightforward requests well, but struggled with complex, ambiguous or emotionally charged cases. Klarna began rebuilding human support capacity and moving towards a more hybrid model.
The lesson is not that AI failed. It is that the initial measures of success were too narrow. Speed, volume and average satisfaction did not fully capture whether complex customer problems were being resolved well, or whether the organisation retained the human judgement those problems required.
For employees, the conflict was understandable. If leaders describe AI as a way to replace the equivalent of hundreds of roles, people will hear a message about their future, whatever else is said about innovation or customer experience.
Deutsche Telekom: adoption without avoiding the harder questions
Deutsche Telekom offers a different European example. It has pursued an ambition to become an “AI-native telco”, embedding AI across network management, customer care and internal work. More than 100,000 employees have been trained in AI use, and the company provides tools including ChatGPT Enterprise, Microsoft Copilot and its internal AskT assistant.
The company’s stated approach is significant. It says employees will set goals, review results and retain responsibility for decisions. In its 2025 reporting, Deutsche Telekom said 77 per cent of employees regularly used AI in their daily work, and it committed to socially responsible staff restructuring and ongoing training.
This is a more credible framing than simply announcing efficiency gains. It acknowledges that AI changes workflows, while also stating that people remain responsible for decisions.
But the tensions remain. Industry reporting on Deutsche Telekom’s OpenAI partnership notes unresolved questions about hallucination risk, privacy, customer-care containment rates and the impact on frontline employment. It also notes that scaling automation across customer care and network operations will require serious engagement with organised labour and large-scale retraining.
Deutsche Telekom is useful because it shows both sides of the change. Broad training and employee access can build capability and reduce fear. But AI at scale still raises questions about accountability, employment and the division of work between people and machines. Those questions cannot be answered by adoption statistics alone.
Lufthansa: when efficiency and employment collide
Lufthansa Group shows the conflict in its most direct form. In its 2025 annual reporting, the company said that digitalisation and increased use of artificial intelligence would deliver efficiency gains across many areas and processes. It also announced plans to cut around 4,000 jobs worldwide by 2030, mainly in Germany, with the changes to be implemented in consultation with union partners.
Here, the connection between AI and job reduction is explicit. Lufthansa is not claiming that AI will simply augment everyone’s work. It is saying that technology, efficiency and structural adjustment will reduce the number of roles.
That does not make the decision illegitimate. Airlines face cost pressure, competitive markets and the need to invest in new capability. But it does change the nature of the conversation.
Employees are being asked to accept a change that may reduce their employment prospects. In that context, reassurances about “augmentation” or “new opportunities” will carry limited weight unless they are backed by concrete commitments on redeployment, retraining, consultation and protection for affected workers.
Lufthansa’s approach at least recognises the need for negotiation with unions. That matters. AI change becomes more workable when employees have a genuine mechanism to influence how it is implemented, not merely an opportunity to be informed after the decision has been made.
One NZ: capability-building with an honest tension
One NZ provides a New Zealand example. By late 2025, the company was scaling AI across customer service, network operations, marketing and software development. It has used AI in contact centres to generate and summarise call notes, helping staff understand and resolve customer queries more quickly.
One NZ has framed AI as a “superpower” and encouraged staff to ask whether AI could do parts of their work better. It has also been more direct than many organisations: leaders have said that some jobs will evolve, new roles will emerge and some areas will become leaner.
The company has backed that message with an internal AI School, responsible-use training and mandatory AI guardrails. It has also encouraged people to move from repetitive work towards higher-value, creative and strategic work.
This is not a conflict-free story. People may still worry about job security, workload or the loss of familiar work. But One NZ’s approach is more credible because it does not pretend that nothing will change. It names the possibility of leaner areas while offering training and a narrative about how people might move into different work.
Australia: capability partnership or displacement?
In Australia, Bank of Queensland’s 2025 partnership with Capgemini shows how quickly a capability narrative can be experienced as displacement.
The bank announced a strategic partnership with Capgemini covering agentic AI, information technology and business processing services. From the organisation’s perspective, this was a way to access specialist capability and modernise operations. From the perspective of affected staff, it meant the loss of local contact-centre roles and the transfer of work to a Capgemini-run centre in India.
The union challenged the move, and customer commentary raised concerns about offshore service quality. That does not prove the partnership will fail. But it shows how a decision framed as strategic capability can be experienced as a decision about whose work, relationships and employment will continue.
Officeworks provides a similar example. In 2026, the company was reported to be shifting hundreds of white-collar roles to India and the Philippines, including replacing its Sydney-based customer-service centre with a call centre in the Philippines. The company described the move as creating a “globally connected team” and keeping costs low for customers.
Affected staff heard something different. They saw local roles being displaced and work being advertised in another country while their own futures were uncertain. One employee reported that local staff saw their jobs advertised in Manila before the consultation process had finished.
“Globally connected team” is intended to signal scale and collaboration. For people whose roles are affected, it can signal that their work is transferable, their local knowledge is less valuable and their employment is contingent.
Why resistance is information
Edgar Schein argued that organisational culture rests on deeper assumptions: what counts as good work, who should decide and what the organisation is really for. AI disturbs those assumptions.
Chris Argyris made a related point about defensive routines. Organisations often address visible symptoms while avoiding the underlying assumptions that create the problem. A team may complain about an AI tool when the deeper issue is that the tool changes who has authority, whose judgement is needed and how performance is assessed.
Jeffrey Ford, Laurie Ford and Angelo D’Amelio have argued that resistance should be understood as communication. People who object may be identifying implementation risks, unintended consequences or a gap between the stated intention and what they experience.
This does not mean every objection should determine the outcome. Leaders still need to make decisions. But objections should be understood before they are dismissed. In AI change, resistance may point to real issues: unclear accountability, poor data quality, unsafe use in high-stakes decisions, loss of customer context, excessive monitoring or the erosion of skills the organisation will still need.
Change involves loss
Daniel Kahneman and Amos Tversky’s work on loss aversion helps explain why AI change can provoke strong reactions. People tend to experience losses more strongly than equivalent gains.
The losses are not always material. They can be professional and symbolic.
Charles Taylor’s work on recognition is useful here. People need their contributions, commitments and identities to be recognised. When AI appears to devalue what someone has built, they may respond strongly, not because they oppose improvement, but because they feel their professional worth is being questioned.
An experienced customer-service worker may have deep knowledge of difficult cases, local context and customer history. If AI takes over the routine contacts and the organisation reduces human capacity too quickly, that worker may feel their expertise has been treated as disposable.
The issue is not simply whether they keep a job. It is whether the organisation recognises the judgement, relationships and contextual knowledge that made their work valuable in the first place.
From announcement to dialogue
Jürgen Habermas distinguished communication that transmits information from communication that seeks mutual understanding. In the first, the aim is compliance. In the second, the speaker accepts that others may question, interpret and influence the outcome.
Many AI programmes claim to consult but operate as announcement. People are told what will happen, invited to ask questions and then expected to accept the decision.
That can produce compliance. It rarely produces commitment.
A more credible approach is to be specific about what is open to influence:
- Which tasks will AI perform, and which will remain human?
- How will accuracy, privacy and accountability be managed?
- What happens when AI gets it wrong?
- Which roles will change, which will reduce and which will grow?
- What retraining, redeployment or transition support will be available?
- How will employees and their representatives influence implementation?
These questions do not need to be answered with complete certainty. But they need to be acknowledged honestly. If leaders ask for feedback but make no visible changes, people learn that voice is decorative.
What leaders can do
A few practices make a difference.
- Separate adoption from transformation. Datacom’s 2026 research found that 91 per cent of New Zealand organisations use AI in some form, but only 4 per cent say AI is transforming core operations. Many organisations are adding tools without redesigning the work around them.
- Be clear about human accountability. Deutsche Telekom’s stated principle that employees set goals, review results and retain responsibility for decisions is a useful starting point. The test is whether that principle is visible in actual workflows, not only in strategy documents.
- Measure what matters. Klarna’s experience shows that speed, volume and average satisfaction can hide problems in complex or emotionally sensitive work. AI change needs measures that include quality, escalation, customer outcomes, employee capability and the retention of critical human judgement.
- Name the employment consequences. Lufthansa’s approach is direct: it links AI and digitalisation to efficiency gains and job reductions, while committing to consultation with unions. Leaders may not like that conversation, but avoiding it does not make the underlying issue disappear.
- Protect the ability to challenge. Amy Edmondson’s work on psychological safety shows that people will only raise concerns and admit uncertainty when doing so is safe. If questioning AI leads to being labelled resistant, the organisation loses access to useful information.