Your Design Tools Just Changed Jobs. Here's What Yours Should Do Now

Your tools are doing different jobs than they did last year. AI is now inside the canvas, inside the code agent, inside the identity work. That is not a headline anymore. The real question is what to do about it before your next roadmap review.

Let me catch you up on four moves that landed close together, and what they mean for the team you run.

When the machine makes it, humans still pick the flavor

Figma's Brand Studio built the whole Config 2026 visual identity with AI in the loop, then spent the effort making it look hand-touched. They built tools in Figma Make to codify "lo-fi effects" they wanted: scribbly linework, blurred gradients, oval particles. They even made a dithering tool to give hundreds of speaker headshots one look without hours of tedious work.

The point they kept hitting: automation gave them speed, taste gave them a reason to trust the output. "We wanted to appreciate touches of imperfection," said Brand Designer Maria Chimishkyan. The MacPaw team building their AI assistant Eney said the same in plainer words. "Curiosity helps build taste over time, and taste becomes more valuable, not less, in the AI era."

So AI does the grind. Your people decide what good looks like. If your review only checks whether the work shipped fast, you are grading the wrong half.

The tool sprawl is real, and it is expensive

Count the tools showing up in these pieces. Figma Make. Claude Design. Chrome's guidance CLI. Then look at Lenny Rachitsky's Product Pass, now at 34 products including Cursor, Replit, Lovable, Factory, and Linear, priced at over $40,000 in value. The reason he keeps building it is worth reading twice.

He says he is "watching a gap widen between people who are having the most fun and success of their entire careers thanks to AI, and people who are feeling confused, stressed, and destabilized." His fix is a full free year per tool, not a trial, because "a typical free trial gives you only enough time to poke around."

That is the tradeoff for your team. Real skill takes durable habits, and habits take months. Buying ten seats of ten tools does not build them. Pick two or three, give people a full quarter, and see which ones earn a permanent spot.

Claude and Codrops show the ceiling is higher than a text box

Nick Babich showed Claude Design making animated 3D objects for the web, on top of the UI mocks and presentations it already does. The tool most people use for flat screens can now push into motion and dimension.

Compare that to the hand-built MERSI architecture site from studio FLOT NOIR. They kept Webflow as a CMS, then layered custom GSAP, Lenis, and Taxi.js on top: a cover that flips into each case study over 1.2 seconds, a grid that reorganizes itself on scroll. No UI frameworks, just vanilla JS and craft.

Both reach the same feel of a polished, physical, editorial site. One got there by prompt, one by hand. The gap between them is closing on the low end and staying wide on the high end. Know which end your product lives on before you decide who builds it.

The AI agent will happily ship you 2018

Here is the one that hits your codebase directly. AI coding agents "often solve modern frontend problems with legacy patterns," per LogRocket's walkthrough of Chrome's Modern Web Guidance. Ask an agent for a modal or tooltip and it reaches for extra JavaScript and another dependency, because the web platform ships faster than model training data.

The test was clean. Two copies of the same Next.js app, same prompt, same model. Without the guidance, the agent used a floating UI library for the tooltip, media queries for the cards, and synchronous filtering that could lag. With it, the agent used native <dialog>, CSS Container Queries, scheduler.yield(), and CSS Anchor Positioning.

The fix installs in one command and reads your Baseline browser target from a file like CLAUDE.md or AGENTS.md. It does not ban libraries. It stops them from being the automatic first answer.

The deep cut

The dependency creep from AI agents is invisible until it is a maintenance bill. Every extra library your agent adds means more bundle size, more failure points, more code your team has to reason about later. Nobody flags it in a demo because the app works. It only shows up six months out when performance is slow and nobody remembers why the stack is so heavy.

Make it visible now. Add a Baseline target to your agent's instruction file this week, then diff a real feature built with and without it. That is a 30-minute test that could save you a quarter of cleanup. It is the cheapest thing on this whole list and the one nobody will do unless you ask.

Three questions for your team

• Which two AI tools are we committing to for a full quarter, and who owns reporting back on whether they earned a permanent seat?
• What is our declared browser Baseline target, and is it written into the file our coding agent actually reads?
• On our next launch, where does taste have to win over speed, and how will we check that in review instead of just checking that it shipped?