My agent army writes Code Connect | Dev Mindset

By Figma

Categories: Design, Product

Summary

In 2026, AI agents can automate documentation maintenance at scale—Figma's Jake Edmondson shows how to deploy an agent army to generate and validate Code Connect docs automatically, reducing manual context window bloat by 60-80% and catching library changes instantly.

Key Takeaways

1. Use agents with MCP tools (like Figma's Git Context) to automatically generate Code Connect documentation from component libraries, eliminating manual doc writing for large systems.
2. Implement agent-driven validation workflows: agents can monitor code repositories and design systems for changes, detecting inconsistencies in real-time and publishing corrections automatically.
3. Optimize LLM context windows by providing 'potent context'—precise, component-linked documentation instead of verbose design specs. Code Connect reduces context bloat significantly, preventing lossy summarization.
4. Distribute agent capabilities as plugins and skills (similar to Figma's approach) so teams don't rebuild core functionality—the Code Connect API skill shipped with the platform enables agents immediately.
5. Create a translation layer between design and code systems that agents understand—Code Connect acts as a bridge, allowing AI to generate accurate implementations from design intent with minimal ambiguity.

Related topics

• Design
• Product
• MCP Tools for Design Systems
• AI Agent Documentation Automation
• LLM Context Window Optimization
• Code Connect API
• Design-to-Code Agent Workflows

Transcript Excerpt

I've been coding for a long time. I enjoy writing code. I spend a lot of time doing it. But it's 2026, people. We don't need to do that anymore. I can clone myself. I can take my knowledge and multiply it. Just watch me. I created an army of agents to create Code Connect on my behalf. So, in the first video, we learned that documentation of our system is really critical, especially when we're providing design context to agents. We're sitting here with a component library that's really big. So, we need to write all that documentation. And we actually might have one or more code libraries that it represents. That's a whole lot of Code Connect. I love writing Code Connect. And I'm not saying that because I work at Figma. I don't want to do it. I'm going to try to do it. We can train our agent…

More from Figma

• 13 Designs that Inspired MIT Professor Zach Lieberman I Source Material
• 9 Designs That Inspired Meshtimes, One of Tech's Most Creative Minds | Source Material
• Figma Motion basics: Path trim
• What inspires Thom Bettridge, Editor in Chief of i-D Magazine | Source Material
• Using agents in Figma Design (and coming soon to FigJam and Slides)