Figma Workflow Lab: staying in the flow with Figma's agent
Most conversations about AI in design focus on generation — producing drafts, exploring variations, building mockups faster. A Figma Workflow Lab post from October 2026 takes a different angle: what happens when a feature is nearly done and the real work is maintenance, review, and handoff?
The article traces how Brett McMillin, a designer advocate at Figma, added temporary host permissions to an existing permissions flow in the Offtrail app. The walkthrough covers three stages, each with a clear division between what the agent handles and what the designer decides.
The three-stage workflow
Inspect and repair. Before touching the design, McMillin runs a /repair-library-usage skill on the frames involved. The agent scans for deprecated components and variables, proposes fixes where a clear equivalent exists, and annotates cases it leaves alone because the right call is not obvious. The designer reviews the list and accepts only the clear matches, keeping ambiguous decisions in their own hands.
Review feedback. Rather than reading comment threads by hand, McMillin asks the agent to read all design critique threads for the feature, group recurring themes, and flag contradictions and open questions. A /compliance-review skill separately checks the flow against the compliance partners’ written guidance and annotates the affected frames with the specific concerns. The output is a short brief: what to act on, what still needs a decision, and where the two sides of a comment thread disagree.
Spec the system change. The feature needs a change to the shared item-row component — it has no way to show when temporary access expires. Rather than building a custom one-off component, McMillin uses a /document-component-enhancement skill to draft a spec for the design systems team. The spec sits in the same Figma file, giving reviewers direct access to the design and the reasoning behind the enhancement request.
What the workflow demonstrates
The practical value here is not speed, though the workflow is faster. It is that all decisions about product behavior, policy exceptions, and design system direction stay with a person. The agent’s role is focused inspection and repetitive repair — tasks where its speed helps and where its mistakes are easy to catch.
The article also makes a case for keeping everything on one canvas. Comments, annotations, specs, and the designs they refer to are all in the same file, so any reviewer can trace a finding back to the work that produced it.
Who this is relevant for
This walkthrough targets designers already working with the Figma agent in a design-systems context. The specific skills it uses — /repair-library-usage, /compliance-review, /document-component-enhancement — are described as Figma agent skills that can be configured by teams, so some setup is required. For teams still evaluating where to start with AI in their design process, the three-stage structure is worth reading as a model for how to split work between human judgment and agent execution.