The week’s window
This launch watch covers November 3–10, 2026, through the question at the center of design-agent.dev: how can AI agents learn to design better? Not just by generating more output, but by reading real design context, making changes in useful formats and giving teams a way to check those changes.
A launch is a newly announced product or capability; a documentation update is a newly published or changed account of how an existing product works. This week includes both, and the distinction matters: a feature announcement is not the same thing as a documented agent surface.
The Nov 3 watch covered Kombai through 2.0.169; this one starts with 2.0.172 and 2.0.175. The Nov 7 code-layers post is the adjacent reading; this watch scores the week’s related documentation rather than repeating that post. The Nov 9 digest covered Select area, Seamcult, and Dessn; those items are not revisited here.
How the scoring works
Read means the documented ability to retrieve or inspect design, code or other relevant context. Write means the documented ability to create or change an artifact. Verification means a documented way to check an output beyond simply producing or previewing it. Every item names its lock-in or access dependency, including seats, beta gates, waitlists and platform ties.
There are three verdicts. Documented means the fetched source provides a supporting quote or explicit statement, linked inline. Documented with a stated caveat means a capability is described but its scope or assurance is limited. Not established by these sources means the pages fetched for this watch do not document it; it does not mean the capability does not exist. Previewing your own output can help with self-checking, but it is not independent verification.
Figma
Figma’s week spans code-layer context, agent-generated vectors and Figma Motion. The documentation shows several routes for moving between design context and editable output, but it does not establish a third-party verification layer.
Code layers and the agent
Read — Documented. The Figma MCP server documentation says the server gives agents “the design context they need to generate code from your Figma files.” The Nov 3 release note adds prototype connections and interactions as context when an agent creates a code layer from design frames. Code layers are in open beta, according to the release notes.
Write — Documented. The MCP documentation says agents can “write native Figma content back to the canvas” and, with the right skills, “build and update frames, components, variables, and auto layout.” The client table lists 21 clients, with write-to-canvas support ticked for 16. The Nov 4 Figma workflow post also names Open code editor, Extract designs and Update code layer as surfaces for working between code and design.
Verification — Not established by these sources. Figma Motion’s style and Lottie previews let authors inspect animations before applying or exporting them. Those are author self-checks, not documented visual-regression checks, spec checks or independent export verification.
Access dependency: the remote MCP server and a Full seat are required to write outside drafts. Figma says the feature is free during beta and will eventually become usage-based and paid; no price or beta end date is stated. Code layers are in open beta. The agent is generally available on Starter, Professional, Organization and Enterprise plans and higher education; Collab, Dev and View seats can use it in their drafts, according to Figma’s agent workflow post. Whether write-to-canvas works on Starter plans is not established by these sources.
Motion and vector assets
The Nov 10 Figma Motion update adds physics-based springs with mass, stiffness and damping settings, plus previews of styles and Lottie exports. The release note says, “Preview how your animation plays as a Lottie file before you export it.” That is useful author feedback; whether the export preserves editable vector layers is not established by these sources.
The Nov 5 vector and icon update describes independently grouped layers, illustration references for style matching and strokes for line work that can be adjusted without redrawing. These are documented editing affordances, not evidence of automated verification. The same release says the layer structure is ready to animate with Figma Motion.
Adobe Firefly
Adobe’s Nov 10 MAX announcement brings together generally available tools, betas and a private-beta waitlist. The Firefly AI Assistant is in beta; Firefly Design is a beta editing space, and Firefly Builder is a private-beta tool for creating interactive, on-brand prototypes. Adobe says Builder lets creators edit elements directly on a canvas while a design system keeps the project consistent.
Read — Not established by these sources. The announcement describes a Model Selector and deeper integration across Firefly multimedia editors, but it does not document an agent read surface for inspecting a team’s design files or system.
Write — Documented with a stated caveat. Adobe describes in-product editing and prototype creation on the canvas. The Builder announcement says an initial prototype can be made with natural language and then edited directly. Builder is private beta via waitlist; other new tools and capabilities are in beta. That documents product writing, not an external agent API or MCP surface.
Verification — Not established by these sources. The announcement does not describe an independent check of generated output.
Access dependency: availability is a stack of generally available Firefly features, betas and a private-beta waitlist for Builder. Adobe also names models from Adobe, Topaz Labs and partners including ByteDance, Google, Kling, OpenAI and ElevenLabs. Whether Adobe documents an MCP or API surface is not established by these sources. Pricing and waitlist dates are not stated here.
Kombai
Kombai’s dated changelog entries for 2.0.172 and 2.0.175 add a component-library surface and setup changes for named editor plugins. This is the clearest new example this week of an agent-facing workflow grounded in a working codebase.
Component Library
Read — Documented. In Kombai 2.0.172, Kombai says it “reads every UI component from your codebase, npm packages, and Storybook” and shows components and their visual states in a library. The Component Library documentation describes browsing components and using “View code” to jump to source.
Write — Documented. The changelog says the library helps Kombai reach for real components when asked to build a design rather than inventing new ones. It also describes adding a design or design system to context with a one-line copy action. These entries document a route from component context into a design request; they do not establish that the library works outside the named environment.
Verification — Documented with a stated caveat. Kombai says components with changed code are redrawn, removed components are flagged, and renamed or moved files keep their links intact. That is a consistency and drift mechanism, not an independent audit of the design or implementation.
Setup and access
Kombai 2.0.172 says its MCP Server switch now sets up a lightweight skill for Cursor, Claude Code, Gemini CLI and Antigravity instead of a full MCP server entry. Kombai 2.0.175 says its Codex plugin sets up the canvas view automatically when needed, and its Claude Code plugin sets up the Kombai Taste card before a first design chat in Claude Desktop’s Code tab. Those are the changelog’s stated changes; the material does not establish what the Taste card contains.
Access dependency: the documented setup is tied to named editor plugins and integrations. Pricing, seat terms and whether the Component Library works outside those named editors are not established by these sources.
Google Gemini
Google’s Nov 5 Gemini announcement describes a sketch-to-app flow: upload a photo of a sketch, select Canvas and ask Gemini to code the app.
Read — Documented with a stated caveat. Gemini “reads your visual and writes the code,” according to the announcement. This establishes image-level input from a sketch, not a read surface for design files, component libraries or code repositories.
Write — Documented with a stated caveat. Gemini produces code from the sketch and can adjust colors, fonts or features in real time. The source describes code output in Canvas; whether it can be exported to a repository or design file is not established by these sources.
Verification — Not established by these sources. Google says users can test the result in a preview window. That is an author self-check, not third-party verification.
Access dependency: plan, seat and region terms are not established by the page fetched.
Lottie AI
The Product Hunt launch page lists Lottie AI as featured November 10, with 114 upvotes, 34 comments and a #7 Product of the Day placement; it names “Nattu” as hunter. These are traction signals, not evidence of capability.
The launch copy calls the product “Editable vector animations from a text prompt” and says users can add hover and click interactions with a prompt. It also claims: “Export to dotLottie, JSON, GIF, video and more.” These are vendor launch statements, not a documented agent surface. The LottieFiles homepage describes Motion Copilot and Prompt to Vector, but neither page fetched names an MCP, API or CLI path.
Read — Not established by these sources. Write — Not established by these sources. Verification — Not established by these sources. The Product Hunt copy’s export claim does not establish how an agent reads, writes or verifies an animation. Access dependency: the fetched pages do not document a waitlist, seat requirement or agent integration path.
Anthropic 5.2 as a boundary item
Anthropic’s Nov 10 news page introduces Claude Fable 5.2 and Claude Mythos 5.2 for technical work in coding, finance and science. The Nov 9 system card calls them two configurations of a new large language model. No design feature is named.
Read — Not established by these sources for design artifacts. Write — Not established by these sources for design artifacts. Verification — Not established by these sources for design artifacts. Claude’s release notes do not document a design-specific Claude Design change during this window. Access dependency and design-specific availability terms are not established by these sources. This is a boundary check on a model announcement, not a design launch score.
Comparison table
| Product | What shipped (date) | Read surface | Write surface | Verification surface | Access / lock-in dependency |
|---|---|---|---|---|---|
| Figma | Documented — code-layer context (Nov 3), vectors (Nov 5), Motion (Nov 10) | Documented — design context and prototype connections | Documented — native canvas writing; 16 of 21 listed clients have write-to-canvas ticked | Not established by these sources — previews are author self-checks | Remote MCP + Full seat to write outside drafts; open beta, free during beta, usage-based paid later; plan and seat limits |
| Adobe Firefly | Documented — MAX expansion (Nov 10) | Not established by these sources | Documented with a stated caveat — in-product canvas editing and prototypes | Not established by these sources | GA/beta/private-beta waitlist stack; model partners |
| Kombai | Documented — 2.0.172 (Nov 5), 2.0.175 (Nov 6) | Documented — codebase, packages and Storybook | Documented — real components inform design requests; setup changes for named plugins | Documented with a stated caveat — sync/drift flags, not an independent audit | Editor-plugin tether; price and seat terms not established by these sources |
| Google Gemini | Documented — sketch-to-app Canvas flow (Nov 5) | Documented with a stated caveat — image-level input | Documented with a stated caveat — code output in Canvas | Not established by these sources — preview is an author self-check | Plan and region terms not established by these sources |
| Lottie AI | Documented — Product Hunt launch (Nov 10) | Not established by these sources | Not established by these sources | Not established by these sources | Agent access terms not established by these sources |
| Anthropic 5.2 | Documented — model announcement (Nov 10), boundary item | Not established by these sources for design artifacts | Not established by these sources for design artifacts | Not established by these sources for design artifacts | Design-specific access terms not established by these sources |
Watchlist re-checks
Resurf: switches before connection
The Resurf agent guide documents an agent-connection procedure and says: “Under Access, turn on what the agent may change before you connect. An agent set up with every switch off can only read.” Its homepage navigation also carries an “MCP & CLI For your agents” entry. The changelog has no entry inside this week’s window; its newest listed version is v2.0, dated August 8, 2026. The per-capability switches and read-only-by-default setup are a useful least-privilege pattern, not a new launch.
Other checks
Modeinspect’s pricing page is unchanged from the Nov 3 fetch: it describes three plans, no tokens and no surprise bills, with Hobby free and Pro at $20/month when it becomes daily work. “Monthly agent limits” and “Extended agent limits” remain unquantified on the page fetched.
For ABrush, the fetched page describes an AI studio in Photoshop; MCP, API, CLI and file-format surfaces are not documented on the page we fetched. PixelCrew describes specialized agents coordinating on a brief and gives Elena, Researcher · Strategy, as an example; protocol and export surfaces are not documented on the page we fetched. Shotcandy describes a free, open-source screenshot beautifier running in a browser, with nothing uploaded and no account; a protocol surface is not documented on the page we fetched. Our November 10 fetch of the Designeer site at designeer.xyz returned HTTP 500 with all scraping engines failing, so it carries no link here. Its current state cannot be established from this run.
What verification means this week
The broad pattern is more ways to create and change artifacts, but few documented ways to verify them independently. Figma style and Lottie previews, and Gemini’s preview window, help an author inspect a result. They are not third-party verification. Kombai’s sync and drift flagging is a useful consistency mechanism, but not an independent audit.
The same distinction applies to launch signals. Lottie AI’s upvotes and Product-of-the-Day placement are traction, not proof of agent access or artifact quality. An agent’s ability to read a file or write to a canvas does not, by itself, establish that the output follows a design system or behaves correctly.
FAQ
Which products document both read and write surfaces this week?
Figma documents design context and native canvas writing in its MCP server documentation. Kombai documents component reading from codebases, packages and Storybook, alongside design-building workflows in its changelog.
Does Figma document write-to-canvas for every listed client?
No. Its client table lists 21 clients and ticks write-to-canvas for 16. The documentation also states that writing outside drafts requires the remote MCP server and a Full seat.
Does a preview count as verification?
Not as independent verification in this scorecard. Figma and Gemini document previews for author inspection; the fetched pages do not establish a third-party check.
Is Kombai’s component sync an audit?
No. The 2.0.172 changelog describes redraws and flags for changed or removed components. That documents consistency handling, not an independent audit.
Can an agent use Lottie AI through an API, MCP or CLI?
That is not established by the Product Hunt page or LottieFiles homepage fetched for this watch. Launch copy about exports does not establish an agent surface.
Can Gemini send its app output to a repository or design file?
That is not established by Google’s sketch-to-app post, which documents code generation and a preview window.
The Bottom Line
Write surfaces widened this week. The documented read-and-write pairs are Figma and Kombai: Figma connects design context with native canvas writing, while Kombai describes reading real components and using them in design work. Adobe and Gemini document useful in-product creation flows, but the sources do not establish equivalent agent read surfaces. Lottie AI’s launch copy is not enough to score an agent integration, and Anthropic’s model announcement names no design artifact surface.
Independent verification remains dark across every scored launch. Previews are author self-checks; Kombai’s drift flags are consistency aids. Bring your own review layer: inspect output against system rules, test behavior and keep a human approval step where the cost of a bad change matters.
Watch the access terms as closely as the capabilities. Figma’s planned move from free beta to usage-based paid access deserves attention; Adobe’s availability is split across GA, beta and a private-beta waitlist. Resurf’s least-privilege default—an agent with every switch off can only read—is the access pattern to copy.
How This Guide Was Built
This is a documentation-based watch, not a hands-on test. The pages cited here were fetched on November 10, 2026. The reporting window is November 3 through November 10, inclusive; items outside it are not treated as this week’s launches. Where a vendor’s wording is quoted, it is presented as that source’s wording, not as an independently confirmed result.
“Not established by these sources” means the pages fetched for this post do not document the surface. It is not a claim that the capability does not exist. The Designeer fetch failure is disclosed rather than treated as evidence about its current product state.
To avoid duplication, this watch builds on the Nov 3 launch watch, the Nov 7 code-layers post and the Nov 9 digest without re-reporting their covered items. The Nov 9 digest’s Select area, Seamcult and Dessn coverage is intentionally limited to the continuity note above.