Building agents that hold their secrets
In today's newsletter, I've got a lot of great content for you, including:
- Career advice in the age of AI
- How a public hack challenge stress-tested Claude Opus
- A free book that turns psychology research into design decisions
- And 7 more
Most Interesting
39 Principles for Designing Human AI Interaction
UX Collective lays out 39 principles for AI interface design, covering appropriate reliance, user control, trust, and transparency.
Design + Development
Three Figma Motion Designers on What Makes Motion Work
Three Figma motion designers break down easing, timing, and transitions, and why grounding motion in film, nature, and physics produces work that holds up longer than whatever's trending.
Pangram Pangram's latest serif, built for branding, editorial, and digital interfaces.
Wouter de Bres wrote a free online book connecting psychology research to product design practice. Each chapter covers a specific finding: your own cognitive biases, user behavior patterns, cultural context, and where the ethics get complicated when psychology shapes interface decisions.
Tech + Innovation
What Happened After 2,000 People Tried to Hack My AI Assistant
Fernando Irarrázaval ran a public challenge: 2,000+ people tried to trick an AI assistant into leaking a secrets file. The secret held. He breaks down the attack methods, what failed operationally, and what he learned about prompt injection resistance with Claude Opus.
Four months, 100,000 lines of TypeScript, and one open-source Slack agent called Junior. The author explains why he built a generalized agent instead of leaning on vendor tools, and how Junior handles GitHub issues, code tracing, and visual QA without leaving Slack.
Google Reader Was Building the Wrong Future
Justin Duke of Buttondown argues Google Reader's design pushed toward a centralized, platform-owned web, and that its failure still shapes how we think about RSS, newsletters, and who actually controls an audience.
Work + Mindset
The Organizational Cost of Low Taste
Aurélie Radom's argument: organizational dysfunction is a taste problem. When shared quality standards erode, decisions slow, meetings multiply, and the people most sensitive to quality leave.
Career advice in the age of AI
Phil Chen reflects on six lessons from designing at his own startup and at companies of wildly different sizes: how collaboration shifts, ambiguity gets harder to read, and team culture quietly shapes what good design even means.
A set of rules Vercel requires from their Design Engineers. An aspirational list for anyone building on the web today.
Resources & Market Signals
Edition #120
Design Systems Meet AI, Process Evolves
Edition #144
