Publication note: The AI UX Dispatch will be on holiday break for the next two weeks, returning to publication on January 7.

Frontier Models

GPT-5.2: A shift toward longer, delegated work

OpenAI released GPT-5.2 on December 11, positioning it as a step forward for professional work inside ChatGPT and the API, especially generating artifacts like spreadsheets and presentations. The headline metric OpenAI is emphasizing is GDPval, a new benchmark of knowledge-work tasks across 44 occupations; GPT-5.2 Thinking reportedly beats or ties human professionals 70.9% of the time. Ethan Mollick called the new GDPval score “a very big deal” because it’s closer to an economically relevant measure than other AI benchmarks.

The release comes amid competitive pressure from Google's Gemini 3, which has recently gained market share. What makes GPT-5.2 notable is its improved ability to maintain coherence across large files. With this capability, the workflow is shifting: rather than prompting for quick answers, we're moving toward defining and delegating meaningful chunks of work. This requires a new skill set, what Nate Jones calls "delegation craft."

How do the three leading frontier models compare at this point? Nate Jones weighs in: "Gemini 3 excels at rapid, accurate summarization of massive documents and has strong analysis capabilities—but the ergonomics make it harder to get context in. Opus wins on beauty, writerly taste, and handling ambiguous or messy human situations. GPT-5.2 wins on systematic analytical coherence across large datasets where you need everything to add up and be thoroughly squared off. The right choice depends entirely on what kind of work you’re actually doing."

Ars Technica | OpenAI releases GPT-5.2 after “code red” Google threat alert
⚡ Quick Read (4 minutes)

Nate Jones | ChatGPT 5.2 complete teardown ($)
Watch Time: 15 minutes

AI Product Development

Cursor brings visual design tools to its AI coding platform

Cursor, the AI coding platform that has quickly become a go-to tool for software developers, just launched its Visual Editor, which allows users to manipulate a web interface visually. Think drag-and-drop layout changes, live color pickers, and typography controls—changes that Cursor hands off to an AI agent to apply directly to the codebase. Users can also click on any element and describe changes in plain language ("make this button bigger," "swap these sections"), and Cursor implements the changes in code.

What sets Cursor's approach apart from other visual AI tools is its focus on professional development workflows rather than rapid prototyping. Cursor's Visual Editor is built for working in large, existing codebases. It's less about spinning up something new quickly and more about bridging the gap between design and development in production software.

Cursor | A visual editor for the Cursor Browser
⚡ Quick Read (2 minutes)

Wired | Cursor launches an AI coding tool for designers ($)
⚡ Quick Read (5 minutes)

AI Product Development

A designer's case for prototyping in Claude Code

In a recent video, Michael Riddering (Ridd) walks through creating what he calls his favorite prototype ever—using Claude Code instead of traditional design tools. His approach: clone a section of his production codebase, strip it down to just the component he wants to explore, and build a "playground" where he can rapidly iterate on the design.

His key insight is knowing when AI coding tools make sense. He finds that Figma can hit its limits with text-heavy interactions, and standalone prototyping tools like Lovable or V0 require recreating existing production UI. Claude Code lets him work directly with real code while exploring variations.

This short episode is worth watching for the practical advice, including developing a shared language with Claude, using annotated screenshots liberally, and relying on voice input for richer context. The result? A simplified branch his engineer is now building on directly. It's a glimpse at how the designer-to-developer workflow might evolve when prototypes begin with production code.

Ridd | My favorite prototype I’ve ever made
Watch Time: 15 minutes

AI Use Cases

Anthropic's new use case library

Anthropic has published a use cases gallery showcasing dozens of practical examples of what you can actually do with Claude. The examples span everyday tasks (travel planning, recipe cards, weekly prep) to professional applications (contract redlining, sales reports, campaign analysis, customer personas).

Each use case demonstrates a complete workflow with real outputs. It's a practical starting point for anyone trying to figure out where AI might actually fit into their work.

Anthropic | Get inspired by what you can do with Claude
💡 Bookmark for Reference

A little year-end fun

I loaded all my newsletter issues into NotebookLM and asked it to create a "six months in review" video. The result is a surprisingly delightful recap of where we've been—and a fitting way to close out 2025.

📽️ The Great Acceleration

That’s it for this week.

Happy holidays, and see you on January 7 with more curated AI/UX news and insights. 👋

All the best, Heidi