The Live Medium
Design judgment gets sharper when it can touch the finished thing.
For years we put a lot of design judgment upstream in Figma files, then hoped the product would survive translation. That made sense when production was expensive. It makes less sense when the next version can appear in minutes and quietly break the last thing you liked.
Fulya Lisa Neubert, Senior Product Designer at Slack, gets right to the material problem: keyboard-driven search can’t be fully judged in a static prototype. You need to press Tab, watch focus move, see content reflow, and feel whether the behavior is right under your hands. Her shift into the live browser reads as design moving closer to the medium where the experience becomes real.
That same pressure shows up in the Figma workflow examples. Emma Webster, writing in the Figma blog, describes teams testing interactions in code before choosing a direction, then bringing the work back onto the canvas with design-system context intact. A working model becomes the first serious product question: does the idea survive data, logic, motion, permissions, and system constraints? A flat mockup can still be useful, but it can also make teams overconfident about behavior they have never actually felt.
This is why Claude Design interests me more as a process story than as a tool demo. Dan Carey, product manager at Anthropic, describes a team that skipped PRDs and used prototypes to decide what was worth building. They shipped incredibly fast, then removed fine-grained controls after usage showed that vocal power users were not the whole market. Working prototypes exposed the wrong assumption before the team spent months polishing it.
The same principle applies when the product stops being a single screen. Alejandro Gonzalez, VP of Engineering at Mozilla.ai, argues that agent-native software may move the source of truth from the visible artifact to the structured object underneath: the commitments, constraints, permissions, states, and records that different surfaces render for different audiences. A deck, dashboard, memo, checklist, and agent-readable schema can all point to the same product object. Designers still care about pixels, but the pixels are no longer the only place design decisions live.
Ferrari’s Luce is a useful reminder that moving into the medium means more than code. The car’s problem, at least in the criticism I covered, is that the exterior asks the badge to carry too much identity. The interior may be considered, the EV strategy may make sense, and LoveFrom may have done admirable surface work. But a Ferrari has to prove Ferrari-ness through stance, proportion, face, performance, sound replacement, and ritual. Brand truth has to be verified in the object, not attached at the end.
That is where I’d put the bar for designers right now. We don’t need to romanticize code, and we definitely don’t need every designer pretending to be a frontend engineer. But we do need to stop judging dynamic products from artifacts that can’t express dynamics. If the customer will experience the behavior in a browser, a car, an agent, a dashboard, or a checkout flow, design judgment has to get close enough to inspect the thing directly.
The old handoff let us describe intent. The new tools let us test it sooner. I want the second one.
What I’m Consuming
Quality in the Age of Slop. Sinclair Target uses Zen and the Art of Motorcycle Maintenance to work through why AI-generated code can feel wrong even when the practical argument for using it is strong. The essay is long, self-aware, and more emotionally honest than most AI craft debates. Its real subject is whether caring about the work still matters when output gets cheap. (Sinclair Target)
The Unsolved Mystery of Lorem Ipsum. Emily Zhang’s Rabbit Hole follows Lorem ipsum from Aldus PageMaker to Cicero, Letraset sheets, Richard McClintock’s correction, and a very specific page in a 1914 Rackham translation. The joy here is watching a design-history footnote turn into actual detective work. Placeholder text should not be this interesting. Somehow it is. (Emily Zhang / Rabbit Hole)
The most important Design System in 2026 that designers missed was built by a developer. Pawel Klasa makes the case that shadcn/ui has become the default design system of AI-generated React interfaces. Designers missed it because it lives in GitHub, starter templates, Vercel’s v0, and now MCP workflows rather than in Figma libraries. That should sting a little. (Pawel Klasa / Medium)
Code-as-Content Era. máuhan compares vibe-coded micro-apps to memes, SoundCloud drops, and internet-native cultural objects. The piece is a little breathless in places, but the behavior it describes feels real: software is becoming cheap enough to publish as expression rather than only as durable product. (máuhan / dead.online)
Crawling a billion web pages in just over 24 hours. Andrew K. Chan built a 12-node crawler that fetched 1.005 billion HTML pages in about 25.5 hours for roughly $462. The fun part is the operational detail: parsing became the bottleneck, SSL handshakes ate CPU, and massive in-memory frontiers created their own failure modes. It’s a rare systems writeup where the scale sounds absurd and the tradeoffs stay concrete. (Andrew K. Chan)
Clanker: A Word For The Machine. Armin Ronacher draws a useful boundary around AI language: the machine isn’t a coworker, a person, or a moral actor. His point is not anti-AI. It’s about keeping responsibility with the humans and organizations deploying these systems, especially when anthropomorphic language makes blame conveniently blurry. (Armin Ronacher)