A 23-year-old Chinese developer runs 300 AI agents at the same time - and none of them can lie to him.
He opened the dashboard in front of everyone: 300 Kimi K3 agents were working in parallel, and each output was checked by Opus 5 for consistency with the source and facts. Target: 100 companies in the electric vehicle market. In the first round, 12 agents failed - with incorrect revenue data, invalid reference links, and blank fields. In the second round, only 3 failed.
Golem update — two days of refinements, from footsteps to rim lighting.
Opus 5, 13 prompts, ~3h of active agent time, 204M tokens, ~$234 API-equivalent. Mechanics — $91, four art rounds — $144. Lesson: art iteration is expensive not from thinking, but cache invalidation — every
Time to repost this but for opus 5
A 22-YEAR-OLD from china spins up 300 AI agents at once and none of them are allowed to lie
He opens the live dashboard: 300 Kimi K3 agents running in parallel while Opus 5 verifies every single output against its original source. Aimed at 100 EV-market companies. Pass 1 --
You have reached the end of the archive
All of Claude Opus 5