This week I did a masterclass on AI-ready design systems, sharing things from building Florence design systems. I wanted to share some key findings. One of the hardest bottlenecks of this all is that there is no playbook, and nor are our tools caught up to be AI-native. So currently have to bend some tools to get this working , but making design system AI ready is an unsolved problem that’s I started a research on agentic design systems at Human AI Studio
One of the problems I am trying to solve is, when you build in Cursor, Claude, or code, our output is generic UI that’s not aligned with the brand, lacks taste, and in most growth-stage startups to enterprises, this won’t work or scale; we will just scale slop.
That led to research of agentic design systems and building Florence at Human AI Studio. I am very close to launching the beta and currently evaluating the benchmarks for it, so agents can understand the design system so product and design teams need not deal with drifts and AI slop.
Quick Announcement: I am conducting a workshop on August 29th, 2026, where you will learn how to make design systems AI-ready. You can join the workshop from here.
Here is the link to watch the free masterclass I did on AI-ready design systems. Around 500+ people had signed up and designers from top companies attended. Glad I was able to share my findings with them.
Some of the key takeaways from this session.
• Design systems are becoming AI infrastructure: agents are now users, but every system was built for humans only and we need them for both.
• AI readiness is the first step, agentic is second. Agents without structured data hallucinate; that’s the challenge for us to move towards being agentic systems.
• AI ready = agents can find, understand, and follow the systems helping teams to ship fast.
• Scattered source of truth. Your system lives in Figma, Storybook, a Notion page, and a decision buried in a Slack thread. The context fragments, and agents fill the gaps by guessing.
• Design tooling isn’t AI-native. Most of our tools were architected before agents existed. Retrieval works nearly perfectly in Cursor or Claude Code, then degrades the moment you pull into a canvas. Two-way sync between code and design is still the bottleneck.
• Manual governance. Audits are done by hand, so they’re tedious and nothing is measurable. You can’t tell whether your system is actually holding.
Thank you for reading.
If you want to dive deeper into AI ready design systems, come join the workshop.