Publishing note: I am on vacation next week. The AI UX Dispatch will be back on Wednesday, July 23.

AI and Product Development

The new role of design systems in the age of AI

I want to lead with this quote from Sam Anderson's excellent LinkedIn post Design Systems are Infrastructure:

"As AI tools become increasingly integrated into design and development workflows, the role of design systems will change dramatically. AI design tools are trained on data—patterns, languages, and principles. For companies that want to differentiate their products in an AI-driven world, design systems will become the training ground for their models."

Jack Anglesea expands on the idea of design systems as providing critical context for AI-based design and development:

"Systematic design isn’t just about building components or managing a design system doc. At its core, it’s about codifying design decisions, structure, hierarchy, interaction rules, voice, and tone in a way that’s repeatable, scalable, and interpretable by both humans and machines. ...Without a solid design system in place, AI outputs feel generic and inconsistent."

He shares his experience using Figma's MCP server to connect his design system to vibe-coding tools (Cursor, Subframe, and Figma Make):

"What’s most exciting is how this tool starts to feel like co-creation rather than generation. You’re not giving up control, you’re giving the AI smarter guardrails and getting more focused ideas in return. Instead of crafting every screen by hand, we’re building design logic and teaching the AI to execute within that framework."

Why systematic design matters even more in the age of AI
⚡ Quick Read (4 minutes)

Using AI

Why context engineering beats prompt engineering

Tobi Lutke on X: "I really like the term ‘context engineering’ over prompt engineering. It describes the core skill better: the art of providing all the context for the task to be plausibly solvable by the LLM."

Andrej Karpathy expands on this: "Context engineering is the delicate art and science of filling the context window with just the right information for the next step. ...Too little or of the wrong form and the LLM doesn't have the right context for optimal performance. Too much or too irrelevant and the LLM costs might go up and performance might come down. Doing this well is highly non-trivial."

The entire thread is worth reading to understand the broader concept of context engineering.

Andrej Karpathy: X thread on context engineering

Using AI

Deep research done right: A comprehensive guide

Are you curious to try deep research or improve your deep research skills?

Unlike standard LLM queries that rely on training data, deep research actively investigates in real-time, conducting web searches, reading full articles, and synthesizing findings into comprehensive reports. Performing competitive analysis, gathering background information on a target market or industry, and understanding design trends — these are all tasks that can be greatly aided by using deep research. However, the key is knowing how to achieve results that are well-founded and tailored to your purpose.

Torsten Walbaum has you covered with his "ultimate guide" to deep research, drawing on his hard-earned lessons. Particularly helpful are:

• Side-by-side examples showing the impact of good prompting

• Comparison of free alternatives to premium model subscriptions

• Prompting templates to get the best results

His guide is a long read, but you can bookmark it for future reference (particularly handy for the prompting suggestions).

Getting the most out of deep research
🔍 Long Read (21 minutes) | 💡 Bookmark as reference

AI in the Enterprise

Knowledge graphs can provide a missing link between AI and accurate answers

This past week, Nate Jones, on his excellent (but paywalled) Substack, conducted a deep dive into enterprise RAG implementations, including where they can fall short.

Here is how Nate characterizes RAG (Retrieval-Augmented Generation):

"RAG fundamentally reshapes what we thought possible from AI by handing these brilliant-but-flawed models a crucial upgrade: an external, dynamic memory. Imagine giving our hypothetical brilliant person access to an extensive, always-up-to-date digital library—now every answer can be checked, validated, and supported with actual data. It’s like turning that closed-book exam into an open-book test, enabling real-time, accurate, and trustworthy answers."

Large-scale RAG implementations on corporate data can become technically complex, with many “flavors” of RAG emerging to meet different needs. One powerful approach is Graph RAG, which incorporates structured relationships between data points to enable deeper reasoning and retrieval.

Graph RAG uses a knowledge graph to give the LLM structured, semantic context it doesn't inherently possess. Think of traditional RAG systems as reading your documents like scattered sticky notes—each fragment is disconnected from the others. Graph RAG is different: it builds a map of how ideas relate by using a knowledge graph. You can think of a knowledge graph as a sophisticated kind of information architecture—one that makes relationships between entities explicit and navigable.

Sections I and II of this article by Elastic Search Labs explains how Graph RAG works and the benefits of using this approach.

Graph RAG: Navigating graphs for Retrieval-Augmented Generation
☕ Medium Read (13 minutes)

AI and the Design Process

Automating information architecture with Claude Code

Noted information architect Jorge Arango has been experimenting with using AI to support the development of information architecture, using his content-rich personal site as a test bed. Most recently, he tried using Claude’s "Code mode”—an AI capability optimized for coding tasks—via a command-line interface (e.g., the Mac terminal). While Claude Code is designed primarily for programming, Jorge found it surprisingly effective for creating a taxonomy for his site.

This may all sound a bit nerdy, but information architecture is a critical element of UX, and it's important to understand how AI may affect this practice. Jorge does a great job of stepping through his process.

Using Claude Code for information architecture
☕ Medium Read (6 minutes)

That’s it for this week.

Thanks for reading, and see you on Wednesday, July 23 with more curated AI/UX news and insights. 👋

All the best, Heidi