The debate about artificial intelligence and employment is usually framed as a choice between people and machines. That is understandable, but incomplete. In most organisations the immediate question is not which jobs will disappear, but how the tasks that make up each job will change, which new responsibilities will appear and which capabilities people will need to develop.

The Future of Jobs Report 2025 estimates that by 2030 around 39% of current skills will be transformed or lose relevance. In addition, 63% of employers identify the skills gap as one of the main obstacles to business transformation. Technology advances quickly; the organisational capacity to absorb it does not always keep pace.

Automating tasks is not the same as removing professions

A job is made of different activities. Some are repetitive and predictable. Others require reading context, negotiating interests, managing emotions, taking responsibility or deciding with incomplete information. Artificial intelligence affects each of them differently: it can draft a first answer, classify a request, summarise a meeting or detect a pattern, but the final value may still depend on the judgement of a person who understands the situation and takes responsibility for acting.

That is why useful analysis does not start by asking which jobs AI can replace. It starts by breaking work down:

  • Which tasks can be fully automated.
  • Which can be accelerated with assistance.
  • Which need supervision.
  • Which must remain a human responsibility.
  • Which new tasks will appear as a result of the change.

Productivity is not the only criterion

Available research shows gains in writing, programming, consulting and customer-service tasks, although effects vary widely by activity and professional profile. But an organisation should not judge AI on speed alone. It must also measure output quality, error rates, customer experience, the team’s cognitive load, learning capacity, professional autonomy and the level of supervision required.

Automating a task at the cost of more downstream checking may deliver no real benefit. Likewise, cutting response time while damaging customer trust means optimising one metric and worsening the overall result.

The risk of automating without redesigning

When a new tool is bolted onto an existing process, people often end up working for the system: reviewing its answers, correcting its errors, entering duplicate data or handling exceptions the automation never considered. To avoid this, implementation must redesign the whole process:

  • Define the outcome you want.
  • Remove tasks that no longer add value.
  • Decide which part the system can take on.
  • Set control points and escalation mechanisms.
  • Assign responsibilities.
  • Evaluate the experience of the people affected.

The best automation is not the one that touches the most tasks, but the one that removes friction without introducing disproportionate risk.

AI literacy is already a corporate responsibility

People cannot properly supervise a technology they do not understand. They need to know its capabilities, but also its limits: plausible errors, bias, exposure of confidential information, over-reliance or use of incorrect sources. In the European Union, Article 4 of the AI Act, applicable since 2 February 2025, requires organisations using AI systems to ensure a sufficient level of AI literacy among their staff.

This does not mean everyone must become a technical specialist. It means training must match each person’s context and responsibility: a sales team needs to know what information may be entered into a system; a manager, when a recommendation needs review; a developer, the security and traceability requirements; the board, which risks it is accepting.

Leading change with people

Rollouts fail when they are presented as closed technical decisions and the team is simply asked to adopt them. The people who run a process know exceptions, risks and needs that never appear in a diagram; involving them in the design improves the solution and reduces resistance. A responsible process includes participation from the start, clear communication about goals and limits, training tied to real situations, channels to report errors, periodic impact reviews and protection of professional autonomy.

Trust is not earned by promising that artificial intelligence will change nothing. It is earned by explaining what will change, why, and with what safeguards.

Artificial intelligence can take on part of the execution. Judgement, responsibility, creativity and trust remain deeply human.