Ethan Mollick's latest post marks a significant shift from his previous "which AI should I use" guides. The big change is that using AI no longer just means chatting with a bot, and Mollick emphasizes the importance of understanding the three-layer framework:

Mollick concludes with practical advice:

I highly recommend you read his entire post, especially if you haven't yet explored AI beyond chatbots.

Per the above, if you're looking to expand your skills beyond the chatbot, Ruben Hassid's guide breaks down six distinct capabilities that Claude now offers and how to properly set those up. These include:

Hassid walks through installation steps and provides some starter prompts.

On a related note, Ethan Mollick maintains that Claude's Plugins hint at the near-term future of AI at work by allowing for domain-specific skill packs that non-technical experts can write, test, and scale across an organization:

"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."

Designer Nicolle Hazard had a familiar design system problem: the gap between what existed in code (shadcn/ui components backed by Tailwind) and what existed in her Figma environment was massive. Bridging this gap manually would take months of tedious, repetitive work—swapping variables, renaming tokens, checking consistency across dozens of components. Instead, she found a way to connect Claude Code directly to Figma through the browser, letting AI handle the mechanical labor of migrating and aligning an open-source Figma component library to her own system. The work took about a day. But the speed wasn't the only payoff. The process of explaining her design rules to an AI forced her to articulate decisions she'd been making intuitively—border colors, disabled states, hover patterns—and document them as she went, creating a living design reference as a byproduct.

There is a second part to her story: using Claude Code to "vibe design" in Figma:

"Once I had all the components built, published, and documented, I opened a brand new Figma file to start designing the actual interface. I gave Claude Code a fresh browser tab pointing to this new file and said: 'I need a preliminary version of this screen. Use the variables and text styles from the published library. Use the components we built.' It gave me a solid first pass. A real layout, using my real components, with my real tokens. Not a generic wireframe or a placeholder mockup, but an actual design artifact built from the system I'd just created."

In her post, Nicolle provides everything needed to get started on the approach she describes.

In a recent post, Margaret-Anne Storey discusses cognitive debt — a concept gaining currency.

Likely, you're familiar with technical debt: messy coding and design choices that accumulate over time, particularly when teams are moving fast. Cognitive debt is the human side of that problem: it's what happens when developers lose their understanding of why the software works the way it does. As AI agents generate more and more code, the speed of development comes at a cost if no one on the team can explain what the system actually does or how to safely change it. Storey argues that teams need deliberate practices: code reviews, documentation of intent, and regular knowledge sharing, to keep their shared understanding intact.

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

All the best, Heidi