Elevator pitch: The handoff between design and engineering is where most Flutter products lose fidelity, time, and trust. We'll show how a VGV product designer and engineer used Claude end-to-end on a real client engagement by merging two native design systems into one AI-readable Figma system, then generating theme-integrated, golden-tested Flutter widgets that match design intent pixel-for-pixel. Description (~250 words): Most "AI in Flutter" talks are told from one side of the wall: an engineer using a model to write code faster. The interesting workflow lives on both sides of that wall, and across it.In this session, a VGV lead product designer and engineering lead walk through a real client engagement where Claude became a shared collaborator across the full design-to-code pipeline. • Refining the design system for AI readability: Before a line of code was written, we used Claude to interrogate, restructure, and improve an existing Figma system. Think token naming, component conventions, variant structure, documentation. Not just for human clarity, but for machine readability. This included establishing AI readability best practices that made downstream code generation dramatically more reliable. • Claude Code inside Figma: Using Claude Code via the Figma Desktop Bridge MCP, allowed us to accelerate the design system refinement itself. By enabling this, Claude had the ability to directly edit inside Figma in real time: bulk renaming, correcting inconsistencies, writing documentation, creating variables. Fast, precise, and collaborative. • Generating the specs: Turning Figma artifacts into structured, machine-readable specifications engineering can actually consume. • Implementing in Flutter: A Figma-to-Flutter agent skill that reads design tokens directly via the Figma MCP server, generates theme-integrated widgets, and validates pixel fidelity with LLM driven visual comparison, and Widgetbook use cases for human validation. • VGV good practices: Feature-First Clean Architecture, Widgetbook, Very Good CLI scaffolding, and the guardrails that let AI accelerate without eroding quality.You'll leave with a concrete blueprint for a cross-functional AI workflow (not a demo reel), honest lessons from what works and what doesn't yet, and a clear sense of why pairing a product designer and engineer on AI tooling produces something neither gets to alone. Built on a live client engagement, presented by the people who actually shipped it. What attendees will learn • A repeatable pattern for using AI in the design system definition phase, before engineering starts. • How to structure Figma so an agent can consume it reliably, including token, component, and variant conventions. • The mechanics of a Figma-to-Flutter skill: MCP integration, code generation, theme integration, golden-test validation. • Where the human stays in the loop on both the design and engineering sides, and why that matters. • How this fits into Feature-First Clean Architecture and the wider VGV agentic engineering practice.