Articles

On October 29 Figma published how it trains its own editing model, plus post-launch numbers for its agent: 20% faster, 33% higher ratings, 60% fewer undos. We read what each figure is measured against, what the 62% match-or-beat rate leaves out, and what the post doesn't say.

Figma's November 5 release notes list four improvements to vector illustrations and icons made by the agent: strokes for line work, layers grouped by function, existing art as a style reference, and a structure ready for Figma Motion. Includes where Figma's own pages disagree.

Figma's October 30 release notes added a Google Drive connector for the Figma agent and Make, on all plans. What the release note and help page confirm about setup, permissions, write-tool defaults and file-size limits, and what they leave out.

Claude Design now works inside any Claude conversation, Claude Code and the mobile apps, and the standalone claude.ai/design site closes on December 14. Based on Anthropic's own help page and product page: where to start, plans, exports, and what to move first.

What Figma's November 2 release note and two help articles confirm about shaders going generally available, 4096px export and code layers, where the two documents disagree on browser requirements, and what changes after November 4.

Canva's Brush, announced October 29–30, is a model for editing existing designs by instruction. It is open to Pro subscribers first. What the reports confirm, what stays unknown because Canva's own page would not open, and what to do before trying it.

Figma's release notes say that since October 26, an outside AI agent can use Weave's video and image models (Seedance, Kling, Nano Banana) through the Figma MCP. What the official pages confirm, a rough per-run cost worked out from Weave's pricing page, and what is still unknown.

What Penpot's October 28 release note and user guide confirm about shared design tokens, per-side strokes and WebGL rendering, where the GitHub releases page disagrees, and what to back up before you try it.

When you edit a shader or generative plugin made by Figma's agent, earlier versions are now kept and can be restored. What the October 27 release note, Figma's help pages and developer docs confirm, what they leave out, and what to do before the next edit.

Google DeepMind released Genie 4 as a research preview. It does not output a clip; it keeps generating the scene as you move. What the official posts confirm about features, access and limits, and what is still unknown about price, resolution and usage.

Gemini 4 Argon is now in Figma Make's model picker. What Figma's release notes, help pages and Google's docs confirm, what they leave out (credit cost per run), and a simple way to log the same prompt on two models.

Runway added App Studio to every paid plan. Describe a task in plain language and it becomes an app teammates can run from a link. What Runway's own pages confirm about how it works, pricing and sharing, and what is still unknown.

On October 22 Canva released Craft, its first in-house image model, aimed at designs where layout and text matter. What press coverage says about its two-stage setup, the ranking claims, where you can use it, and what is still unknown, including price and commercial terms.

On October 22, Ideogram released Vector 1.0, its first vector model, which makes editable SVGs from text or from an existing image. What Ideogram's API docs and press coverage say, the reported prices, and what is still unknown about layers and live text.

On October 15, Webflow became a connected app in the Google Gemini app. Per the release, you can build pages, manage CMS content, audit a site and ask performance questions from chat. Free to install, and currently no Webflow AI credits. What the release and marketplace page say, and what is still unknown.

On October 15, Figma put the open-source CMS Payload inside Figma Make. Prompting an app now also sets up a database, file storage, user login and an admin panel. Checked against Figma's blog and release notes: what it does, hosting in free beta with pricing TBA, and what is still unknown.

Krea released its image editing model Krea 3 on October 15. It promises to keep a product's label and logo while changing the scene around it. Checked against Krea's blog and pricing page: speed, cost, how to prompt it, and what is still unverified.

Figma opened code layers in open beta on October 15. You can turn design frames, a GitHub repo, a website capture or a prompt into a clickable, code-backed prototype on the canvas. Checked against the release notes, help article and blog: what is confirmed and what is missing.

Lovable removed the step that showed three design directions before the first build, and design systems are available on all paid plans. Checked against its changelog and docs: what is confirmed, what is missing, and how to set your own rules once.

Figma's design agent has left beta and now consumes AI credits. Checked against Figma's FAQ and pricing page: what is confirmed, what is missing, and how to stretch a month's allowance.

ProtoPie has announced official MCP support. Studio MCP lets an AI read and edit the prototype you have open; Code MCP hands finished prototypes to coding agents. Checked against ProtoPie's own post and docs: what works and what isn't confirmed.

Figma has put markdown documents on the canvas with Specs, now in open beta on paid plans. Checked against Figma's own post and release notes: what works, what isn't confirmed, and how it differs from DESIGN.md.

Google's image model Nano Banana 2.1 is out. 1K and 2K images cost half as much, but 4K doesn't, and text got better while the model card lists weak spots. Checked against Google's pricing page and model card, with charts and numbers.

Muse Wonder, released by Meta on October 12, now ranks first for websites and UI on Design Arena, where people pick AI designs by eye. How the voting works, and what a No. 1 does and doesn't tell you, with diagrams and numbers.

More clients and bosses are handing designers drafts made in ChatGPT or Claude and asking only for a finish. Rates go down while the work stays. A diagram of what's actually left, and an order you can use for quotes and conversations.

Heathrow Airport teamed up with Adobe to measure how AI answers mention its brand. With one in four travelers reportedly turning to AI first, here's what designers working on brands should look at, with a diagram and tables.

macOS 27 removed the switch that turned off Apple Intelligence entirely. Users are building outside tools to strip AI out. What changed, what to check on a designer's Mac, and the question left for anyone designing AI features.

Fanta's Halloween campaign, made with an AI studio, is getting slammed online. Yet consumer testing came out the other way. Who did what, why reactions split, and what designers working on brands should take from it.

A developer is building seven free, open-source apps with AI to replace Adobe's. The Photoshop replacement opens and saves PSDs but is still in early alpha. Here's where it stands and what designers should look at.

ChatGPT now answers with buttons and charts, and the Figma agent is out of beta. Two features from the same week, side by side, with diagrams on who sets the rules for the screen.

Six AI and design stories from September 30 to October 6, on one page. The whole store, the rules, what to edit, the order of shots: what was left to luck is turning into something people decide. With numbers on how much designers use AI.

Kling 4.0 takes 30 seconds, 10 keyframes and 15 references in one generation. It moves from giving just a first and last frame to laying out the flow of a scene yourself. Here's what changes and what isn't out yet, in a diagram and a table.

On October 5, OpenAI announced image ads that appear in ChatGPT's image generation screen. Here's the eight months from text ads to image ads, and what it means for designers, both the new work and the things to watch out for.

MIT researchers released InstructMesh, a tool for fixing AI-generated 3D models before printing. About 80% of generated models had structural flaws, and novices fixed about 90% with the tool. Here's how the fixing works and what to check before you print.

Ideogram 4.5 on September 30, FLUX 3 Image on October 2. This week's two image models led with "change one spot only" rather than prettier pictures. Here's box-first layout and price by size, in diagrams and numbers.

As AI makes more of our drafts and code, where you write down your rules starts to matter. I wrote this site's rules in Google's DESIGN.md format and ran the official checker on it. Here's what it caught.

Shopify's Canvas, launched October 1, puts every page of a store on one surface. You click to edit, or tell Sidekick what you want. Here's what changed and what doesn't work yet, with diagrams and a table.

OpenAI's DevDay brought more than twenty announcements. Here are only the ones that touch design work, laid out with diagrams and numbers.

Anthropic's Sonnet 5.5, released September 28, put design sense front and center. Same price, faster work. Plus what happened when someone had it build a one-page site, in numbers.

While companies shift to cheap open models, Alibaba's new image model Qwen-Image-2.1 shut the door on commercial use. The difference between being able to download a model and being allowed to use it for work.

On September 22, Opus 5.5, GPT-6 Sol and Luna all launched on the same day and the price sheets changed dramatically. What that means, in numbers, for the credits and subscriptions designers pay for.

Two frontier models released two days apart. Official specs, third-party evaluations and early reactions from designers, side by side with sources.

Exploration and mass production go to AI; direction and final checks stay with people. A boundary I only saw after months of use.

Whether it's a client or a prompt box, you end up explaining style in words. A way of breaking style into six layers.

Twenty-plus boards, and still I start every project by searching from scratch. How to collect references you can actually pull back out.

When you want to read, when you want to show your work, when you need an answer fast. Korea's design communities, sorted by what they're good for.

Copyright, terms of service, lookalikes, contracts. What to check in practice, in order. Written from Korea, with the US for comparison.