Figma has introduced an AI design agent that can generate layouts, edit existing designs, apply components and variables, and automate repetitive design tasks through natural-language prompts.
The agent is built directly into the Figma canvas and can work with a team's existing design systems, tokens, components, and other contextual information.
Figma's announcement gives product teams a faster way to explore ideas and make large-scale changes without manually handling every adjustment.
As such, designers can generate multiple directions, apply changes across screens, summarize feedback, and continue refining work within the same canvas.
Figma adds an AI assistant to its collaborative canvas https://t.co/dqDE4Q8sK6
— TechCrunch (@TechCrunch) May 20, 2026
But that speed can often be a mixed blessing, especially when used at scale.
When AI can make hundreds of design decisions in a fraction of the time it would take a human designer, inconsistencies can spread just as quickly.
That's why for Alex Chaly, CTO at Shakuro, a leading product design and software development company, governance is an important part of adopting AI-assisted design.
"The biggest risk in AI-assisted design is the potential to create something that looks right to the untrained eye," he says.
"And if speed is valued over efficiency, design debt can scale just as quickly as design output."
AI Makes Design Systems More Important
The scenario Chaly outlines can be avoided by ensuring the AI tools used have access to a high-quality design system.
A design system has traditionally served as a shared source of truth for designers and developers.
And as AI takes on more of the work, it also becomes a set of constraints that tells the agent which decisions are acceptable.
This matters because AI can only reliably reproduce standards that have been defined clearly enough to follow.
As such, design systems should clarify the following:
- Which components should be used in specific situations
- Which typography, colors, spacing, and interaction patterns are approved
- How components should behave across different states
- Which accessibility requirements apply
- When an existing component should be reused rather than recreated
- How and when new patterns can be introduced
Extend Governance Beyond the Design System
A well-maintained design system gives AI clear boundaries, but those boundaries are only useful if teams also establish what happens when the agent produces something new.
In other words, teams need to determine which outputs can move directly into production, which ones require additional scrutiny, and who has authority to approve new patterns.
According to Shakuro, a practical governance model should address four areas:
1. Create risk-based review tiers
Not every AI-assisted change carries the same risk. That's why review requirements should reflect those differences rather than treat every AI-generated change equally.
Establish three review levels for AI-generated design changes and define what requires approval at each level.
- Low risk for copy changes, minor spacing adjustments, and simple visual variations
- Medium risk for changes to existing components, layouts, and interaction states
- High risk for new components, navigation, checkout flows, accessibility-sensitive interactions, and patterns intended for reuse
2. Add accessibility to the AI review checklist
Even the most stunning design fails when accessibility or usability is compromised.
"Accessibility can't be reduced to whether an interface passes a checklist after the design is finished," Chaly says.
"It has to influence the components, interaction patterns, and decisions that AI is allowed to reproduce in the first place."
Fortunately, addressing accessibility and usability problems early in the design process is considerably easier than after the issue has been implemented across multiple screens.
To do this, create a standard accessibility checklist and require designers to complete it before AI-generated work moves into development.
At the very least, teams should regularly check:
- Color contrast
- Keyboard navigation
- Focus states
- Form labels and error states
- Text readability
- Component states
3. Bring developers into the process earlier
Developers should have visibility into AI-generated patterns that introduce new components, interactions, or architectural requirements before those decisions become embedded in the product.
Require engineering review whenever an AI-generated design introduces a new component, interaction, or recurring pattern.
Keep in mind that this review should happen before the pattern is approved as production-ready, not after development has already begun.
Keep the System Ahead of the Agent
AI agents are valuable because they reduce the cost of experimentation and repetitive work. But speed also changes the scale at which mistakes can spread.
"We won't be able to stop AI from taking on more of the design process, but we can ensure that it operates within the standards, constraints, and review processes that protect quality," Chaly says.
"Finding that balance between speed and quality will be key."
For product teams, that means treating AI as an accelerator within a well-governed design process rather than allowing it to define that process on its own.
