That was the focus of our latest Ateliers Live session, "Agreeing as a Team on How to Use AI." Ateliers is a series of small, workshop based events for design leaders across Europe, focused on leadership, strategy and how teams work with AI. For this online session we were joined by Brian Lin, Senior Product Design Manager at Discord, and Oscar Garcia, Head of Design and author. Here are the key takeaways for design leaders, product teams and anyone shaping how their organisation adopts AI.

Who is accountable when AI agents ship design work?

We opened with a question many teams are already facing: when a designer's agent ships something, who is accountable, and how do you make that ownership explicit without slowing experimentation down?

Brian described how his Core Product team at Discord handles it. They have guardrails in place so that no generative AI output gets shipped directly. Every department can use AI to generate things and make ideas visual, but his team decides what actually gets built. Engineering works the same way. Developers are free to vibe code, but they are the ones who decide what gets built and how.

The result, Brian explained, is that experimentation stays open while quality stays protected. Anyone can explore, but ownership of what reaches users is clear.

On quality itself, Brian added an honest caveat. Quality is something we want as design teams, but not every company holds it as a value. Eventually, teams have to align with the values of the business first. The risk of slop is real, though. When everything gets produced simply because it can be, designers and product people can quickly feel overwhelmed. That came back later in the conversation.

Choosing AI tools for designers when regulation limits them

We asked Oscar what designers should do when regulation restricts which AI tools they can use. His answer was pragmatic: learn to work with what you have. If you truly need something specific, make the case for it.

Before arguing for a tool, though, Oscar urged teams to ask the harder questions first. What are we trying to do here? What is the goal? A tool request without a clear goal is easy to dismiss.

He pointed to design systems as the place where AI is currently having the most impact. New tools seem to solve more and more, but we used to do much of this work manually and it was fine. Not every new capability is necessary.

Oscar has seen designers successfully make the case for adopting specific technologies. His reminder was that a tool is only a tool. It should never become a blocker for good work.

When to standardise AI use and when to experiment

When should a team standardise, and when should it leave room to experiment? Both speakers came back to communication. Teams need to talk openly to agree on the moments and environments that are suited to play, and the ones that are not.

That led to a harder question from the audience. What happens when not everyone holds the same level of accountability? Brand, engineering and design may each treat ownership very differently.

Brian shared that at Discord, it took several meetings in a single month to have the conversations needed to set those boundaries. Alignment like this doesn't happen by accident. It takes deliberate time.

Oscar offered an image that stuck with many of us: lighting a fire in a fire pit in your garden. The area for the fire is controlled. You always know where the fire can reach and what it can and can't do outside of it. Experimentation with AI works the same way when the boundaries are clear to everyone.

Keeping design in the loop when PMs prototype with AI

Several audience questions circled the same tension. How do you deal with PMs building prototypes and taking them straight to developers? How do you advocate for design when AI tools make keeping a designer in the loop feel slow or optional? And how do we make sure the promise of faster shipping doesn't stop us from thinking critically about what to build?

Brian returned to his answer from the first question. The organisation has to be clear about what gets shipped to production and who owns quality. Drawing on the themes of his kickoff session at Agentic Ateliers, he stressed the importance of principles, guidelines and good communication within the team.

In the end, he said, each team has to define ownership and responsibility for itself. For the user experience, that responsibility ultimately falls on designers. For marketing or other areas, it may sit somewhere else.

Oscar added the context layer. Companies need a shared system and knowledge base, so that people don't have to make these decisions on their own every time. For good collaboration, he named a few key steps:

• Find where context gets lost between steps.

• Decide what needs to travel with the work.

• Decide, for each step, what runs on its own, what needs a human check and what stays fully human.

His closing point put things in perspective. We are still doing the same work we always did. The speed at which AI pushes us to work makes these problems feel new, but they were there before.

Agentic Ateliers 2027: AI workshops for design teams in Berlin

If these questions sound familiar, join Brian and Oscar at Agentic Ateliers on May 13 and 14, 2027 in Berlin. It's two days of hands on workshops for design leaders and their teams, limited to 70 participants.

Together with other design leaders, we'll work through the real challenges of using AI as a team, beyond the technology: ownership, business value, workflows, design systems and more.

Learn more at ateliers.design.