On 6 October 2026, DNB announced plans to reduce its Technology & Services workforce by about 400 full-time equivalents as part of a wider restructuring. The announcement raises a question that reaches beyond banking: when AI changes the work, how should leaders decide what happens to the people doing it?

A role can change substantially without a system learning to perform every part of it. The important management decision is how the remaining responsibilities are organised, resourced and supervised.

What happened

DNB’s announcement says agentic AI is reducing manual work in customer-data controls, know-your-customer processes, technology development and coding. The bank expects downsizing during the fourth quarter of 2026, with the full cost effects reflected from the second quarter of 2027.

This is a company statement about its plans and reported experience. It does not establish a one-for-one exchange between employees and AI agents, or demonstrate that the affected roles have been automated in full. The broader organisational change matters to interpreting the announcement.

Why it matters

A staffing decision needs a more complete account of the work than a demonstration of faster output. Preparing a report, checking its evidence, resolving an exception and accepting responsibility for the result are separate activities. Removing effort from one does not remove the others.

Consider a hypothetical team whose system now prepares routine case summaries. Managers could use the recovered capacity to address a backlog, improve service, develop staff or reduce expenditure. Those are choices with different consequences. None follows automatically from the model’s ability to summarise.

Before changing staffing, I would ask the operational owner to show which tasks have actually diminished, which remain and which have been added. Include checking, maintaining reference information, resolving unusual cases and helping customers when the automated process is inadequate.

Then test the proposed arrangement under pressure. Can the smaller or differently organised team handle an absent specialist, a surge in difficult cases or a system interruption? A financial estimate needs to be assessed alongside that operating capacity. Otherwise an apparent saving may leave another team carrying work that was missing from the calculation.

The bigger shift

The useful unit of analysis is the task, but the consequences reach the whole organisation. Work is connected through handovers, expertise and relationships. A decision to automate one activity can change what colleagues need from each other, even when their own tasks remain manual.

This makes role design a leadership responsibility. A technology team can explain what a system does; it cannot alone decide how the organisation distributes its gains or supports people through a change.

The distinction developed in Why AI Strategy Beats AI Tools applies here: business priorities should guide the use of technology. A lower staffing cost and a stronger service are different objectives. Leaders should state which they are pursuing and what evidence would show that the intended benefit is being achieved.

My take

AI changes the work. Leaders decide who carries the cost.

I would put an accountability map beside every proposed AI-related staffing plan. For each responsibility, name the person who will hold it after the change, the evidence they need and the support available when the system fails. The CEO’s Guide to AI Governance provides the wider framework for assigning that ownership.

Employees deserve a clear account of what is changing and what remains undecided. Reassurance about new opportunities means little without allocated learning time, a credible route into revised work and honest discussion of the decisions already made.

The next question for an executive team is practical: can we explain how the proposed organisation will deliver the work, including its difficult exceptions, before we approve the cost reduction? Answering that requires evidence about the process and a deliberate leadership choice. It cannot be delegated to an AI performance benchmark.