This year, AI moved from a side panel to the centre of design tools. At Config 2026, Figma upgraded its agent with reusable custom skills, connections to external tools and live web data, and made agent conversations visible to the whole team by default. Other tools have moved in the same direction, adding agents that can generate layouts, edit existing work and connect to the rest of a team’s toolchain. For studios and product teams, the question is no longer whether to use these agents, but how to use them well. Here is what we have learned.
Where agents genuinely help
Agents are strongest at work that is repetitive and well defined. Producing variations of a layout, applying an existing design system to a new screen, renaming and tidying layers, generating realistic placeholder content, or resizing designs for different formats all fall into this category.
These tasks share one quality: there is a clear standard to check the result against. When the output can be reviewed quickly, the time saved is real.
Where they struggle
Agents are weaker at work that depends on context they cannot see. The conversation with a customer last week, the technical constraint from engineering, the reason a decision was made three months ago, the political sensitivity around a particular feature. None of that lives in the design file.
They can also produce work that looks finished but is subtly wrong: inconsistent spacing, inaccessible colour combinations, or patterns that do not match the product’s logic. Polish can hide problems, so review matters even more.
Use AI to widen the options. Keep the decision with the people accountable for it.
Turn your standards into skills
Reusable skills are one of the most practical developments. A skill is a set of instructions an agent follows every time it performs a task. A well-written skill turns a team’s standards, such as how to name layers, which components to use or how to write button labels, into something the agent applies consistently.
Writing skills has a useful side effect. It forces teams to make their standards explicit. Many teams discover that rules they thought were shared actually lived only in one person’s head.
Keep the work visible
Making agent conversations visible to the whole team is an important change. It means anyone can see what was asked, what the agent did and why the result looks the way it does.
That visibility supports review and learning. It also prevents a common problem with AI-assisted work: decisions that nobody can explain later because they were made in a private chat.
Lessons for the products we design
The same principles apply to AI features in the products we build for clients. People increasingly expect to see why an AI suggestion appeared and to be able to change or undo it. Transparency is becoming part of good UX.
Designing those moments, such as explanations, confidence indicators and easy corrections, is now a core part of product design, not an afterthought.
A workflow that works
Our pattern is simple. A designer frames the problem and sets direction. Agents help explore variations and handle repetitive production. The designer reviews, chooses and refines. Engineering reviews anything that will be built.
At each step it is clear who is responsible. The agent accelerates the work, but the decisions and their reasoning remain with people.
Protecting quality
Agent output should be checked against the same standards as any other work: design-system consistency, accessibility, content quality and fit with the product’s logic.
It helps to keep a short list of common agent mistakes your team has seen, such as inventing new colours or detaching components, and check for them routinely.
Keep the judgement human
An agent can produce ten directions in a minute. Choosing the right one still depends on understanding the business, the audience and the trade-offs. We use agents to widen the set of options and to speed up routine work, and we keep decisions with the people who are accountable for them.
A practical starting point
- Start with repetitive tasks that have a clear standard.
- Write your team’s standards down as reusable skills.
- Keep agent work visible and reviewable by the team.
- Review polished AI output as carefully as rough drafts.
