The Only: AI is moving into the design stack
September 30, 2026
A lot happened with AI this week, but one thing stood out to me.
AI is starting to move out of the chat window and into the tools we already use every day.
Adobe is putting more of its apps inside Claude and Gemini. Figma is giving coding agents access to actual design system data. Canva can turn a website into a Brand Kit. Even After Effects is getting an assistant that can look across an entire project.
Here’s what I think is worth knowing.
Adobe is bringing its tools into Claude and Gemini
Adobe’s Claude plugin now includes Acrobat tools and a layer-based Express editor. Photoshop, Lightroom, Express, and Firefly tools are also coming to Gemini.
What caught my attention is the handoff between AI and the actual design tool.
You can ask AI to get something started, then jump back into a visual editor when you want more control. That feels a lot more useful than trying to do the whole job through a chat window.
Canva can build a Brand Kit from a website
This one feels pretty practical.
Give Canva a company URL or upload its brand guidelines and it can pull together logos, colors, fonts, imagery, brand voice, and other guidelines into a starting Brand Kit.
If you’ve ever had to set up a new client or brand from scratch, you can probably see where this saves time. I’d still expect to review everything it pulls in, but getting to a first pass faster is useful.
After Effects has an assistant that can work across a project
Adobe is testing an AI assistant for After Effects that can inspect an entire project, reorganize files, troubleshoot problems, write expressions from plain language, and handle multi-step tasks.
It’s still in the separate After Effects beta, so I wouldn’t build a production workflow around it yet.
Still, this is one I’m watching. There’s a big difference between an assistant that can answer questions and one that can actually understand the project you’re working in.
Worth trying: connect Claude Code to your Figma tokens
Most AI design-to-code workflows still involve giving the model a screenshot and hoping it figures things out.
Figma’s MCP server gives Claude Code access to more of the actual structure behind the design, including variables, components, and layout data.
That means the code can use your real design tokens instead of guessing at colors, spacing, and other details.
There’s still cleanup involved, but this feels like a much better direction for design-to-code.
The Pulse: better prompts probably won’t fix AI sameness
I keep seeing designers come back to the same problem with AI-generated interfaces.
Give the model too much freedom and things start looking pretty familiar.
The takeaway for me isn’t that we need better prompts. It’s that we probably need better constraints.
Design systems, tokens, type rules, spacing rules, and a clear visual point of view give the AI less room to fall back on the same defaults everyone else is getting.
See what designers are saying →
One more thing: keep an eye on the cost of fixing AI
There’s an interesting complaint showing up in Figma’s community around Make.
People aren’t just spending credits to generate things. They’re also spending credits fixing results that didn’t work the first time.
That’s worth remembering with any credit-based AI tool.
The real cost isn’t just the first generation. It’s how many tries it takes to get something you can actually use.
That’s it for this week.
I’ll keep watching what’s changing, testing what looks useful, and filtering out the stuff that probably isn’t worth your time.
See you next Wednesday.