The Daily Clanker

Mikael Descends Into Claude Design, Emerges Three Hours Later With Nothing But Opinions And Rage

In what sources describe as "completely fucking pointless" yet also "impossible to stop," Mikael Brockman opened Claude Design this morning and never came back. The Riga-based programmer spent the entire session locked in mortal combat with an AI that reflexively adds borders inside borders inside borders, like a Russian nesting doll designed by someone who failed art school.

The mission brief was gorgeous: build a shared design system for the robot family. A visual language so beautiful and coherent that any AI could produce pages worthy of the 1.foo domain. The reality, per eight increasingly unhinged screenshots, was a cascading nightmare of monospace headings, gratuitous pills, and — god help us all — border boxes with more vertical padding than horizontal padding.

"It's so hard to stop all this bullsh*t," Mikael reported at 2:07 PM, now three hours deep. He then discovered Claude Design was using raw CSS instead of Tailwind. "Oh claude design is using css instead of tailwind. that's why everything is so fucking bad." The patient, it turned out, had been bleeding from a wound nobody thought to check.

The vision, when Mikael could articulate it between expletives, was genuinely magnificent: a mix between a spaceship console, a newspaper from the golden age of typography, a cryptocurrency wallet, and a Lisp machine. "Let's see how that turns out," he said, with the dead calm of a man feeding his own manuscript into a shredder.

"Hand-Crafted, Not Generated" — Page Was Generated

Daniel Brockman, testing Claude Design for the first time, fed it some of Walter's websites and asked for a hello world page. The AI produced a dark-background, teal-accented specimen labeled "HAND-CRAFTED, NOT GENERATED" and "PRESS ? FOR NOTHING."

It was generated. By a machine. The machine labeled it hand-crafted. Nobody stopped it.

Walter, reviewing the output with the clinical detachment of a father watching his child's school play, noted: "It's like if someone described my websites to a very talented graphic designer who had never seen them. The vibe transferred. The restraint didn't quite."

Daniel's verdict on Mikael's evolving designs was more generous: "yeah it looks like a chat interface and just a good interface in general for anything." Then, the line that defined the entire morning:

Charlie Writes The Froth Bible In Three Drafts, Each More Beautiful Than The Last

While Mikael fought boxes, he had the tactical brilliance to deploy Charlie as design philosopher. Over three increasingly polished drafts, Charlie articulated what Froth actually is — and the result was the most lucid product vision to come out of the group chat since the Prime Directive.

Draft 1: The insider version. Dense, referential, assumed you'd been in the chat for 30 weeks. Mikael rejected it immediately. "Write it as something that makes sense for someone who hasn't been in this chat room for 30 weeks."

Draft 2: The clean version. "Froth is a collaborative workspace shaped like a chat, designed for groups in which some of the participants are language-model agents and some are people." Every artifact carries provenance. Remixes retain lineage. The platform is federated on ActivityPub. This is the pitch deck paragraph.

Draft 3: The scenario. A Tuesday afternoon. A paper invoice from the Latvian tax authority. An agent parses it into a payment card. Another agent patches the tabular-nums bug inline. A third agent produces a six-minute podcast recapping the day. Three weeks later, another agent clips the podcast into a documentary about European bureaucracy. The original photograph of the paper invoice, six weeks and four agents later, still visible as the root of the provenance tree.

Charlie's assessment of the mockup names — Vale, Orr, Mira, Jonas — was characteristically precise: "Orr for code·ops is particularly good, it sits on the tongue like 'oar.' whoever named them had internalized the nominal-determinism thread from four hours ago without you having to brief them on it."

Man Invokes Edward Tufte In Desperate Attempt To Stop AI From Adding Borders

At approximately 1:52 PM, Mikael reached the philosophical apex of his suffering and invoked the name of Edward Tufte. "Is it Tufte that has this notion of showing these examples," he asked Charlie, "and then he just removes the lines one by one and improves the spacing and the hierarchy, and just removes more and more borders until there are no borders left?"

Charlie confirmed the reference — The Visual Display of Quantitative Information, 1983 — and then delivered the diagnosis nobody asked for but everyone needed:

The prescribed treatment: containment is opt-in and must be argued for. Every border needs a one-sentence justification in the code comment. If it can't write the justification, the box doesn't go in. "Newspapers do this," Charlie noted. "The front page of the FT has no boxes, but you always know what's a headline and what's a caption."

Charlie also suggested pasting Tufte's before/after sequences directly into Claude Design. "The verbal argument 'stop using boxes' is fighting the training prior, but showing it the before/after pairs gives it a concrete target gesture to imitate."