AI reskilling needs paid practice and a voice in the rollout

Training is easier to promise than to make usable. Time, worker participation and a credible route into changed roles determine whether it helps.

An employee can be told to learn an AI tool while facing a full workload, limited access and no clear explanation of the role they are preparing for. That is a training announcement, not a workable transition. The useful question is what the person can practise, with whose support, during which paid hours and toward which realistic opportunity.

## What the research adds The June 2026 brief synthesises earlier evidence; its latest included observations date from late 2024, and parts of its training evidence draw on 2023 employer and worker surveys. Its publication date is not the date of a new workplace experiment.

The OECD's 5 June 2026 AI and skills brief identifies skills as a barrier to adoption and describes better reported outcomes associated with training. These associations do not prove that any course will cause an improvement. The ILO's 10 July 2025 social-dialogue study examines cases across five continents in which worker representatives influence AI decisions. The OECD's 2023 Employment Outlook supplies further analysis of dialogue and collective bargaining. These sources support examining how a rollout is organized, rather than treating training as a stand-alone purchase.

## Begin with a real task and an accessible baseline A program should identify the work changing and the knowledge already present. Experienced employees may understand failure modes, customers and local exceptions that a new interface hides. Before assessing their technical skills, ask them to explain those decisions. Otherwise the employer risks defining competence as familiarity with a tool instead of the ability to deliver a reliable service.

Check access as carefully as curriculum. Can shift workers attend? Is the material available in a language they can use? Are accessible interfaces provided? Do contract or temporary staff performing the same work receive support? Publishing a video link does not answer these questions. Attendance data should show who could participate, not merely who finished.

## Protect practice time Paid practice is an editorial recommendation here, not a statement that the same legal entitlement exists in every country. Its operational purpose is straightforward: a person needs time to make mistakes safely, receive feedback and repeat the task. If training is added on top of unchanged targets, the organization may inadvertently reward shortcuts or unpaid effort.

Separate practice from performance scoring. Use owned or synthetic examples and let participants question the system without fearing that every experiment will count against them. A training task should include a misleading suggestion and a service failure, not only an easy success. The person should learn when to stop, escalate and preserve a customer's information.

## An illustrative transition agreement Consider a fictional back-office team introducing an assistant for document classification. Staff representatives and managers agree on limited practice sessions during paid work time, a trial period without automated ranking and a named reviewer for disputed classifications. They compare accuracy and correction effort before changing production targets.

The agreement also states how workers can move into exception handling and how unsuccessful transfers will be supported. This is an illustrative arrangement, not a claimed universal policy or a case copied from the ILO report. Its value is that people can understand the bargain before the technology changes their daily responsibilities.

## Give consultation something to influence A meeting after procurement is complete may offer little room to affect design. Ask workers about task selection, monitoring, escalation and the pace of rollout while choices remain open. Record proposals, management responses and unresolved concerns. Where representatives exist, give them enough explanation and access to appropriate expertise to discuss the actual system.

Consultation should not require employees to surrender private data merely to prove a concern. They can report a recurring error, excessive correction work or an inaccessible interface through safe channels. Managers should investigate those reports and communicate what changed. Local labor, privacy and consultation duties need qualified jurisdiction-specific review.

## Evaluate the route, not just the certificate Track protected practice hours, access across shifts, skill demonstrations, supported transfers and workload after training. Ask whether a participant can spot an unsafe answer and recover from failure. A completion badge is useful administration, but it cannot establish readiness for a consequential decision.

There are limits: not every changed role has an immediate alternative, and training cannot compensate for an employer's decision to remove work entirely. A credible program acknowledges that constraint, provides transparent transition information and offers support beyond a course. Reskilling becomes meaningful when it is connected to time, voice and an honest account of the work that remains.