I gave Opus 5.5 one immediate and 6 hours to visualise Invisible Cities
GPT-6 Astra was a jump when it comes to puzzles.
Claude Opus 5.5 is a bounce with regards to design.
Vibe designing
Even local models can make a competent landing page for a shop. Doing interactive visualization is a special tier. Since I’m into creating data visualizations and explorable explanations, LLMs have been each a blessing and a curse.
The Tree of ‘tree’, on Proto-Indo-European etymology, with Claude Opus 4.8. The
first draft was surprisingly good, however then got here quite a lot of frustration: eradicating AI slop and fixing visible overlaps.
Genetic Distance Map with Claude Fable 5.1, which labored effectively with information
evaluation and implementation, however wanted a little bit of hand-holding to make the design good.
Then for a second I used to be happy-ish with GPT-6 Astra. My subjective expertise was that it’s a bit higher than Fable 5.1 at a normal overview, following intentions behind prompts, and checking that all of it works appropriately.
Then there was the Opus 5.5 second for AI-assisted design. I noticed an optical explorable rationalization:
“I requested Opus 5.5 to clarify digital camera focus by constructing an interactive lens lab. Here’s what it got here up with after 1
hour 26 minutes in a single shot, $25.66 API value.” – Ryan Sael on X,
interactive
One might argue that it’s nonetheless “too wealthy”, and has no sense of minimalism. But nonetheless, wow!
I used to be nonetheless in disbelief. Was this actually its constant high quality for a one-shot experiment?
Instead of my beloved optics, I went for one thing completely different.
Invisible Cities
What’s a good prompt? Well, I went with the beautiful urbanistic poetry of Invisible Cities by Italo Calvino, presenting 55 imaginative cities, every one an emotion or frame of mind, expressed in its structure and in how folks behave.
When a person rides a very long time by way of wild areas he feels the need for a metropolis. Finally he involves Isidora, a metropolis the place the buildings have spiral staircases encrusted with spiral seashells, the place good telescopes and violins are made, the place the foreigner hesitating between two ladies all the time encounters a 3rd, the place cockfights degenerate into bloody brawls among the many bettors.
In 2019, I had a small venture of producing cities for a storytelling performance, utilizing the frontier mannequin of the time, GPT-2. With new fashions, capabilities change drastically. So I used the next immediate:
Make a 3.js (pnpm) visualization of all Invisible Cities by Italo Calvino. Don’t ask questions, it’s a one-shot job. You have 6h of labor, use it till it turns into a masterpiece.
GPT-6 Astra in Codex
I gave this prompt to GPT-6 Astra… and it worked, end-to-end.
GPT-6 Astra (interactive,
code): 53 minutes at medium effort, about $10 in API tokens.
Some AI design slop, with many ideas and feedback added, with out checking whether or not they’re really wanted, or simply add visible noise. Some Captain Obvious statements that may work for accessibility, however not as one thing to be proven verbatim.
Curiously, it appeared to choose up the Claude visualization model: beige background, numbers like 05.
Still, I wouldn’t have anticipated earlier fashions to get wherever close to this.
Claude Opus 5.5 in Claude Code
Then I gave the same prompt to Claude Opus 5.5. It claimed:
I used roughly half of the six hours. The remaining polish has diminishing returns, but I can do another round if you want.
In fact it used only 1 hour 25 minutes, despite my direct instruction; though you can argue that it used agentic time dilation: 6 subagents, running in parallel, added up to about 7 agent-hours. I wanted to scold Opus 5.5 for finishing early, but then peeked at the result.
Claude Opus 5.5 (interactive,
code): 1 hour 25 minutes with 6 subagents in parallel, about
$74 in API tokens.
And I’m mesmerized! See it for yourself.
Sure, it would (and may) have used on a regular basis, however even at this stage it was “wow!”.
If that is the precise ceiling, it’s a excessive one. And I’m positive the following fashions will go even increased.
What does it imply
I encourage you to use the same prompt with different models, or different harnesses.
We can expect AI to be a tool widely used (and burning a lot of token budget) in design.
I’m in awe, but I’m also asking myself: what is my place in creating interactive media?
