Scale AI across a design team.
How do you move an entire design organization from AI curiosity to AI at scale — without losing craft along the way?
As generative AI reshaped how products get designed, Carrefour's design and product teams faced a defining question: not whether to adopt AI, but how to make it a shared, governed practice across the whole org — from discovery to delivery.
Carrefour wanted AI to become part of how its design and product teams work every day, at the scale of the whole organization.
When they came to us, nothing had been deployed yet. Generative AI was clearly going to change how the teams designed and built products, but the path there was unclear. Which tasks were actually worth handing to an agent, and what should those agents do? And how do you get a whole team working this way, instead of a few curious people on their own? We worked through these questions alongside their team, before any model went live — bringing outside experience to a problem they knew from the inside, and building the answers together.
We came in as design advisor, embedded with the team for the length of the engagement.
Axel led the work and sat with Carrefour's designers and product people throughout. We set up a project group with them, met regularly, and a good part of the work happened between those sessions. Together we agreed where to start and in what order: the foundations first, while AI was still being prepared internally; then a system to share what we built; then the team's working habits, once models started arriving; and the more advanced use cases last. We did it in that order on purpose, so the foundations were in place before AI reached anyone's daily work.
We started by mapping the team's workflow, before putting any model into it.
We went through the product and design workflow with the team, stage by stage, from discovery to delivery. We brought a method and a concrete sense of what agents can do well and where they fall short; they brought the detail of how the work actually gets done. Putting the two together, we identified the tasks where an agent would genuinely help and defined what each one should do. What came out of it was a clear set of agents, each tied to a real task in the workflow.
Once the agents existed, we built a system for the whole team to share them.
A good agent built by one designer is only worth something to the rest of the team if they can find it and trust it. A shared repository was set up where anyone could contribute an agent, with a small dedicated team to review and refine the prompts before they went out to everyone. We worked out the rules with Carrefour: how an agent gets proposed, checked, improved, and made available across the team. By the end, they had a real internal system for building and sharing agents, with people responsible for keeping its quality up.
When the models started arriving, we turned to how the team would actually work with them day to day.
A new tool doesn't get used just because it exists; it has to fit into how people already work. With the team, we shaped new ways of working together around AI: how meetings run, how work gets reviewed, and a few internal challenges to get people trying things without much at stake. The point was to make using an agent an ordinary part of the working day.
With the foundations and the team's habits in place, we took on the newer, less settled use cases.
This is where things were moving fastest: vibe design, canvas design, vibe coding. For each, we looked at where it could fit Carrefour's work and what mix of agents and tools it would take. There was no obvious answer, and the tools change month to month, so we ran proofs of concept with the team on real workflows and compared what they produced. That left the team with a toolstack they had actually tested, and a method for re-evaluating it as the tools evolve.
Today, AI is part of how Carrefour's design and product teams work, from discovery to delivery.
The agents, the shared system, and the new working habits are all in use across the team, and new use cases keep being added as the tools and the work change. The engagement is still going: we continue to work with Carrefour as their practice grows and the technology moves.