Summary
Jonathan Gordon, founder of ReWeaver AI, delivers a conference talk at AI Engineer on what he calls ‘the design-code roundtrip that isn’t’ — the decades-old failure of tools to maintain fidelity when translating between design artifacts (Figma, Sketch) and production code in either direction. Drawing on 30+ years of experience at companies including Microsoft, Gordon argues that LLMs represent the first genuine opportunity to close this loop, because a single model can now understand both design intent and write code from it.
The talk includes the first public demo of ReWeaver AI, a tool that integrates simultaneously with a live web application and a Figma design canvas. ReWeaver’s ‘Show Drift’ feature scans both sides at once, surfacing issues across dimensions including design quality, code quality, performance, design tokens, and accessibility. Gordon demonstrates the tool catching and automatically fixing an accessibility violation — a missing ARIA live region that would prevent screen readers from announcing a dynamic element — which he notes LLMs generate by default because their training data underrepresents accessibility patterns.
Gordon also reflects candidly on lessons from intensive vibe-coding with Cursor, which placed him in the top 0.1% of usage globally. A key turning point was catching an innerHTML vulnerability that a coding LLM generated unprompted, which led him to realize that blind acceptance of LLM output is insufficient for production code — and which directly motivated building ReWeaver’s drift-detection and code-quality audit capabilities.
📺 Source: AI Engineer · Published September 10, 2026
🏷️ Format: Keynote Launch
