Wonderful! 5 artificial intelligence models were asked to design a durable bridge with a 3D printer.
🪙 Claude Opus 5.5: 59 kg 2️⃣ Meta Muse Spark 1.3: 12 kg 3️⃣ GPT-6 Astra: 8 kg ❌️ Grok 4.7: The bridge could not stand. ❌️ Some K3: The bridge could not be assembled.
5 top models got one task: design, engineer, and 3D print a bridge that holds the most weight possible — and the gap between the results was massive
The rules were identical for everyone: a 2-foot span, a max of 500 grams of filament, printing under 18 hours, with a fixed load
Most AI agents don't fail at the task.
They fail the user. A plan you never saw. A silent spinner for 3 minutes. A file changed without asking. An action you can't take back. Here's how to master agent UX with Opus 5.5. 5 moments to design on purpose: 1. Clarify. Vague goal?
Holy f*ck Jev is actually insane at picking models 😭
These guys run hundreds of Jevs on every single turn. each one reads a slice of your conversation, scores it against benchmarks for every frontier model, then they vote on which model takes the next step. watched it work
You have reached the end of the archive
All of Claude Opus 5