Dennis van Bregt on AI: The Work Changes Before the Tool Does

Buying another model does not change who is allowed to publish a product claim. On 5 June 2025 Dennis van Bregt, Data Leader at allnex, argued the opposite order: decide how the work changes, then pick the tool.

The recording:

Speakers:

• Stephan Spijkers, Co-Founder, PIMvendors.com
• Dennis van Bregt, Data Leader, allnex

The conversation kept returning to the same points. AI in this telling is a change in how decisions get made, not a faster way to finish yesterday's tasks. People still set the question and check the answer. The skill is a clear instruction and a clear reject. Trust, privacy and regulation sit in the middle, because a draft that reaches a customer is a governance event, not a writing exercise. Literacy has to reach marketers, product managers and executives. Engineers cannot be the only ones who know what the tool was allowed to do. The pace point was blunt: what counts as advanced this quarter is ordinary the next, so training and small trials have to be funded on purpose.

The trade-off is time. Rules and shared skill come before output you would show a customer. A product-data team that skips that, and buys a model to "catch up", automates the catalog it already has, including the gaps.

For a PIM programme, three consequences follow.

• Name the decision a person still takes when a model drafts a description, a translation or a classification.
• Put privacy and trust in the same review as attribute quality. A correct value can still be one you should not send.
• Train the people who sign the page, not only the people who configure the model.

Who owns a rule, and what "good" means before a model drafts anything, is the work in product data quality and governance. The phases of a PIM programme, including the skills a partner cannot leave behind, are in the PIM implementation guide. What the record has to contain before an agent can use it is in AI-ready product data. The boundary of what a model inside the PIM should be trusted to change is in what AI inside a PIM actually does.