# Figma Config 2026: Weave Is Here. Your Token Layer Still Isn't

> Figma's Config 2026 keynote leaned hard on Weave and Code Layers. A quieter industry note says most design systems still lack the token metadata AI needs.

Source: https://waalee.ai/blog/figma-config-2026-weave-token-metadata-gap
Published: 2026-11-19 · 4 min read · By The WAALEE Team
Topics: Figma Config 2026, Figma Weave AI, design tokens for AI agents, Figma Code Layers, AI design system gap, extract design tokens from website

Figma's Config 2026 keynote was, by every account, a big show for AI inside the design canvas. Weave lets you chain more than twenty prebuilt AI workflows directly into a Figma file, anyone can eventually publish their own, and it's free while in beta. Code Layers push code-native design further into that same canvas. Most of the coverage was about speed and taste. One newsletter recap buried the line that actually matters if you're trying to get an AI coding tool to output something that looks like your brand: most public design systems still don't carry the semantic metadata AI agents need to use design tokens reliably. That gap didn't close at Config. It just got a more impressive stage.

## What actually shipped at Config 2026

Weave is a multi-model pipeline, not a new foundation model. Figma doesn't build its own; the workflows chain external models, reportedly including Anthropic's, into one inspectable sequence instead of a single black-box prompt. More than twenty prebuilt Weave tools landed inside Figma Design at launch, and Figma says any user will eventually be able to publish their own. Code Layers is the complementary push, bringing actual code closer into the design canvas so the line between designing and building blurs further. Figma's own framing leaned on human judgment over bigger models, betting that workflow integration beats raw automation.

## The line everyone skipped past

Here's the part that matters if you're shipping with v0, Lovable, Cursor, or Claude. A TLDR Design recap of Config 2026 noted that most public design systems lack the semantic metadata AI agents need to use design tokens reliably. Translation: even teams with a real Figma file usually haven't exposed their colors, type scale, and spacing as named, structured tokens an agent can actually read and reuse. The values exist. They're just not written down in a form anything but a human eye can parse.

- Weave and Code Layers both assume you're already designing inside Figma, with a file worth building from.
- Neither tool manufactures a token layer out of a brand that only exists as a shipped, rendered website with no matching Figma file.
- If your "design system" is a two-year-old redesign and a Notion page, Config's new AI tools have nothing structured to point at.

## Who already has the token layer these tools assume

This is where a tool like [Zeroheight](https://zeroheight.com/) earns its keep, for the teams that qualify. Zeroheight turns an existing, maintained Figma library into living documentation and distributable design tokens, so a real token layer exists for anything downstream to consume. The catch is in that sentence: existing, maintained, library. If nobody kept the Figma file in sync with what actually shipped, which describes most teams past year two, there's nothing current for Zeroheight, Weave, or any AI coding agent to read.

## Where WAALEE fits instead

WAALEE skips the Figma file requirement entirely. It reads your actual live, rendered website, the one real users see today, and extracts your real colors, type, and spacing as proper W3C design tokens, whether or not a current design file exists anywhere. That's the token layer Config's new AI tools, and every AI coding tool downstream of them, are quietly assuming you already built.

Three answers to "where does the token layer actually come from":

- [Figma Weave + Code Layers](https://www.figma.com/blog/config-2026-recap/): New multi-model AI workflows and code-native layers inside the Figma canvas. Faster design and handoff, but assumes you're already working from a current, maintained Figma file.
- [Zeroheight](https://zeroheight.com/): Turns an existing Figma library into documentation and distributable tokens. Works well, but only for teams whose Figma file still matches what's actually live.
- [WAALEE](https://waalee.ai/): Extracts real W3C design tokens straight from your live site's rendered output. No Figma file, current or otherwise, required.

None of this makes Weave or Code Layers a bad bet. They're a real step toward faster, more inspectable AI workflows for teams who already have the underlying system wired up. But "wired up" is doing a lot of work in that sentence. If your actual brand lives in production and not in a file anyone's touched since the last redesign, the AI tooling arms race at Config 2026 didn't build you a token layer. You still have to go get one before any of these tools are worth pointing at your project.
