AI Code Governance and Design-Code Reconciliation
ReWeaver AI
ReWeaver AI builds tools for developers and designers to detect and reconcile differences between design systems and frontend code, including AI-generated UI. It checks designs and code against shared rules, surfacing issues such as hardcoded colors, missing accessibility patterns, and omitted loading or error states. Its workflow spans VS Code, Figma, and GitHub: users can fix, suppress, or defer findings, with decisions recorded in git. Reconciliation can update either code or the Figma design file.
Founded in 2025, ReWeaver is led by founder and CEO Jonathan Gordon, who moved from software development into software design after encountering usability problems. Its detection engine uses rules and abstract syntax trees rather than a language model. Deterministic rewrites handle supported fixes; when judgment is needed, a local model can draft a change for human review through a pull request. Its Production Drift Ratio weights drift frequency by severity and estimated repair time, normalized per component.
The founding beta is React-first and runs detection locally on users’ machines. Alongside the engine, ReWeaver publishes the AI Coding Survival Kit, offering prompts, rules, and skills for coding assistants. These guide generation rather than perform the engine’s verification. Its free DriftDetector lets users scan public GitHub repositories without installation or signup, returning a drift score, scan history, and file-and-line findings.
Explore the recordings
ReWeaver AI’s supplied archive contains one recording: Jonathan Gordon’s exploration of design-code roundtrips, generated-code drift, and reconciliation. The paths below point to different themes within that talk. Demonstrations, measurements, and product capabilities describe what Gordon presented in the recording; they do not establish ReWeaver AI’s current product status.
Start with the roundtrip problem
Watch The Design-Code Roundtrip That Isn't for Gordon’s definition of a true roundtrip: bidirectional movement between design and code without fidelity loss, with provenance connecting interface elements to their source. His reported tests of five tool setups illustrate where that promise broke down, including lost bindings and design changes that did not survive in code.
Follow the generated-code inspection path
For a practical route through the talk, focus on the innerHTML injection risk Gordon encountered while vibe coding, then the demonstration of scanning code and design together. He shows checks for quality, performance, design tokens, and accessibility, including a missing ARIA live region. These are recorded examples of why he advocates inspecting and steering generated code.
Explore drift and reconciliation
Return to Gordon’s iteration experiment and proposed workflow to examine how he frames accumulated drift as technical debt. He reports that pure model output began around 30% fidelity and degraded across a 12-iteration experiment; this is a speaker-reported result, not a general benchmark. His proposed response combines AI with deterministic checks, preserves provenance, surfaces issues it cannot safely fix, and retains human control. The presentation also introduces nine production-readiness guardrail dimensions.
1 talk
Newest first
1 speaker at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.