Introducing AI design system office hours
Sizing up AI against your design system? Bring it as-is and talk it through with people who’ve led design systems for companies like Atlassian, Reckon and Westpac. Your system doesn’t need to be perfect. If you’ve got a few gaps, or just a sense that something’s drifted, we’d love to chat.
We’ll help you see whether your system is set up to make AI’s decisions for it, where your Figma and codebase have pulled apart, and what a schema-led, single-source-of-truth approach could do for both humans and agents.
It’s genuinely free, with five spots open each month. Book a session.
From the blog
Is Your Design System Ready for AI?: Ben Derham traces how one team’s unwritten convention, always reserving space for a validation message, was invisible to AI and got skipped by default.
What AI use looks like in five enterprise design teams: Jordan Geizer watched how design teams at Stripe, Shopify, Ramp, Atlassian and Intercom prototype with AI, and found context and adoption mattered more than any single tool.
From Vibes to Durable Context: Max Khedri argues that coding agents fill ambiguity with plausible assumptions, and the fix is encoding decisions into tests, types and lint rules rather than documentation that goes stale.
Build vs buy: AI code review for teams that have opinions: Joe Prisk lays out when an off-the-shelf AI reviewer is the right call, and when a team’s size and conventions justify building a thin, model-agnostic review layer instead.
Curated links
How well do agents use your design system?: Sanity built an eval harness that scores how well coding agents build with their design system, and found a structured JSON schema over prose documentation nearly tripled build success rates, a fix they’ve open-sourced along with the tester.
Experiments with AI Code Review: Wealthfront’s engineering team details their AI reviewer, which pits one model arguing a flagged issue is real against another arguing it isn’t and has a senior model weigh the evidence, cutting false positives sharply while keeping human review as the blocking gate.
Tutorial: How to test a Figma design system with Claude Code: A free, open-source tool exports a Figma design system as a deterministic snapshot and checks it against per-component contracts, catching a single mistyped variant name in under a second when tested against GitHub’s Primer.
Thanks for reading! Until next time 👋
