Forty PRs, Three Weeks, Four Designers

Forty pull requests. Three weeks. Four designers.

That's what my team shipped to Adobe's production codebase this past month, and I got to talk about it at Adobe AI Day last week. The number isn't the story. The story is what it took to get there — and what it means for how design teams operate going forward.

From prototypes to production

We started the year like most design teams experimenting with AI: using Cursor to vibe-code prototypes. It worked well enough for early exploration. But there was a consistent gap between what we built in a throwaway sandbox and what actually shipped. Implementation quality suffered. Speed suffered. We were generating more specification artifacts and fewer finished things.

So with support from product, engineering, and design leadership, we got access to the production codebase. All of it. The real thing.

What 40 PRs actually look like

Some were copy fixes. Some were icon swaps. A lot of them were Figma designs translated directly into code using the Figma MCP — mapping our custom components more precisely than any handoff document ever could. We implemented interaction patterns that made the experience more intuitive, changes that had been on the list for months but kept getting deprioritized.

One of our OKRs this year is a quality framework to document product gaps. What started as a backlog of JIRA tickets became something different once designers had direct access: "let me just go fix it." In three weeks, the product looks and feels more refined than it did.

The difference isn't AI. It's proximity.

Code-aware designers are a different resource

Richie Taylor built an entirely new agentic onboarding interface — a dynamic walkthrough that adapts to where the user actually is in their setup flow. Kim Pimmel built the next feature on our roadmap, taking something that had been theoretical for months and validating it in production.

These aren't just prototypes. They're in the product. Speed like this doesn't come from AI tools alone. It comes from designers who understand what clean code looks like, who take implementation quality personally, and who have enough codebase context to make good tradeoffs. Vibe-coding into production is a disaster. Code-aware designers in production is a force multiplier.

The distinction matters. The teams who figure this out will stop designing specs in figma, and instead will write design specs for AI to build.