AI Agents, Design Systems, and the New User

In today's newsletter, I've got a lot of great content for you, including:

Your digital twin is still just personal branding
Why AI job forecasts keep missing the same thing
How fragmented time quietly kills your output
And 7 more

Most Interesting

Figma's Design Agent Works Directly on Your Canvas

Figma's Design Agent Works Directly on Your Canvas Figma's new built-in Design Agent works alongside your team on the canvas, pulling context from your components, tokens, and design system. It connects to Figma Make and the MCP server to bridge design and code without leaving your workflow.

Design + Development

The Claude Code Source Leak

Claude Code works differently when you treat it as a programmable agent. Plan mode, custom commands, subagents, parallel sessions, and self-updating CLAUDE.md rules all change what it can do, and how consistently it does it.

Figma Agent Tested on Three Real Design Tasks

Shapes.gg ran Figma Agent through three actual jobs: drawing vector icons, building a slide deck, and generating a signup flow from a design system. Here's what held up and what didn't.

Fast is a Learned Skill

Dubroy makes the case that speed comes from practice, not talent. The practical stuff: tackle problems the moment they surface, protect time from fragmentation, and ship work before it feels ready.

Tech + Innovation

What We Lost in the AI Chat Stream

Chelsey Qiu writes that AI chat interfaces erode the thinking that happens before you prompt: the early, uncertain stage where designers actually form the question. Most chat history is iteration, not insight. Outsourcing discovery to AI means you stop knowing what you were trying to figure out.

Claude Opus 4.5: What's Actually New

Simon Willison digs into Anthropic's release notes (unusually candid ones) covering honesty improvements, lower hallucination rates, mid-conversation system messages, and reduced prompt cache minimums. Pricing and context window are unchanged.

You Are No Longer the User. You Are the Principal.

At Google I/O 2026, Adrian Levy noticed something UX vocabulary has no word for: the user left the screen. When AI agents act on your behalf, you stop operating and start authorizing. The entire discipline was built around presence, and that assumption is gone.

Work + Mindset

Why AI Job Exposure Predictions Keep Getting it Wrong

Benedict Evans makes the case that AI job-elimination forecasts are mostly noise. Accountants multiplied through decades of automation. The Jevons paradox and shifting job categories mean exposure analysis tends to mislead more than inform.

Judgment is the Skill That Matters Most in the AI Era

Jim Grey, an editor turned engineering leader, argues that what separates effective AI users isn't prompting skill. It's critical judgment. That judgment builds through sustained exposure to flawed output, human or AI. The muscle only develops if you've trained yourself to notice when something is wrong.

The Copy and the Guru

Om Malik built an AI-trained archive of his own writing, then decided the whole idea unsettled him. He connects curated Instagram personas, LinkedIn thought leadership, and AI avatars as the same drift: each one a more polished version of yourself replacing the real one.