Linear published a detailed behind-the-scenes account this week of its latest visual interface refresh — and the more interesting story isn't the redesign itself, but how a two-person team pulled it off.
Charlie Aufmann and Maxime Heckel had been in the codebase for just two months when they took on the project. The brief was familiar: years of accumulated SaaS entropy had left Linear's interface cluttered — header bars with too much going on, icons doing work that labels should be doing, borders scattered across screens without clear purpose. The refresh aimed to restore visual hierarchy and structural clarity to a product that had quietly drifted from its own design principles.
Working fast in an unfamiliar codebase meant relying on AI coding agents from the start. The pair used Claude Code, Cursor, Codex, and Linear's own internal agent for the kind of exploratory work that typically stalls new contributors: finding where components lived, tracing how they were used across the product, and tracking down which engineers and designers had historical context on specific areas.
The clearest example of what agents could do came when the team needed to experiment with a color palette shift — moving Linear's default cool, blue-tinged grays toward something warmer. The normal process, mocking in Figma then opening a PR and waiting for a preview build, was too slow for the volume of iteration they needed. So they built a custom color picker directly inside Linear's existing dev toolbar using Claude Code. It exposed fine-grained controls over hue, chroma, and lightness at the design token level, and let them package and share palette experiments as JSON files that could be imported straight into Figma. What would have taken weeks of back-and-forth collapsed into hours.
Agents also made it practical to explore competing design directions simultaneously. When the team hit a fork — two plausible paths for a component or layout decision — they could prototype both quickly rather than committing engineering time to a hunch. Feature flags let them roll changes out incrementally, keeping the rest of the team unblocked while the design system shifted underneath them.
The result is one of the more concrete public examples yet of AI coding agents accelerating the design and development process itself, not just automating feature output. Two people, new to the codebase, executing a large-scale UI overhaul on a production SaaS platform in a matter of months — that's the productivity argument the agent tool vendors have been making. Linear has now put a real example behind it.