Publication note: The AI UX Dispatch will be on Spring Break for the next couple of weeks—I’ll see you again on Wednesday, March 25.

As AI coding agents get faster and more capable, there's a growing tension in the design world: if agents can just build things from prompts, do designers still need a canvas? The team behind Paper, a new desktop design tool, makes the case that the answer is a strong yes, but the canvas itself needs to change. They argue that chat-based prompting creates a "slot machine" workflow (prompt, wait, output, repeat) that's limiting for the kind of spatial thinking design actually requires: holding multiple directions in view, comparing options side by side, and maintaining coherence across a product.

What makes Paper's approach interesting is that it's built on actual web standards (HTML, CSS, DOM), so designs aren't abstract pictures of UI — they're the real thing. That means agents can read and write to the canvas, and designers can pull live code onto it or push decisions back to the codebase. It's a bi-directional workflow where the canvas becomes a shared space for humans and agents to collaborate, rather than a dead-end handoff.

Jenny Wen, head of design for Claude at Anthropic (and former director of design at Figma), joins Lenny Rachitsky for an honest look at how the design role is being reshaped by the sheer speed at which engineers can now build with AI tools.

I recommend the full episode, but on X, Lenny summarizes the episode, focusing on ten key takeaways:

In last week's issue of this newsletter from Ethan Mollick on Claude Skills:

“If you know a field well, you can probably edit or write your own skill which, since it can be run many times by many people, can have a large impact. And they are written in plain language that a non-technical expert can compose, test & scale. Will be valuable for organizations.'"

If you'd like to get started with Claude Skills, Carmen Rincon at yummy-labs.com recently published two free custom Skills for Claude: one for UX design and one for UI design.

Installing a skill just takes a couple of minutes: download the skills at the link below, go to Claude's Customize tab, and add each SKILL.md file. Note: You will need a paid Claude plan with code execution enabled.

A viral essay from Citrini Research called "The 2028 Global Intelligence Crisis" rattled Wall Street this past week, triggering stock drops of 8%+ at companies like DoorDash, American Express, and Blackstone. The Hard Fork podcast responded by bringing in economist Anton Korinek (University of Virginia, Anthropic Economic Advisory Council), who's been studying AI's economic effects for over a decade. He maintains that the hard economic data isn't there yet—and the gap between frontier AI capabilities and actual workplace deployment remains enormous. In fact, that gap may even be growing as capabilities accelerate faster than organizations can adopt them.

Particularly notable in this interview was Korinek's candor about the uncertainty and the factors surrounding it. He doesn't dismiss the transformative potential of AI (he expects AI capabilities to keep climbing with no clear ceiling), but he's clear that we're still in the realm of expectations rather than evidence. He believes the future is somewhere between "nothing is changing" and "everything is collapsing," and advises staying informed without panicking.

Thanks for reading, and see you on Wednesday, March 25, with more curated AI/UX news and insights. 👋

All the Best, Heidi