AI Pulse · AI Trends Pulse

Andrej Karpathy just gave Claude Opus 5 a $10 token budget and two hours, then asked it to procedurally render the opening of Lord of the Rings in 3D JavaScript. The model wrote 5,500 lines of working code. No human edits, no back and forth. Just prompt, wait, done.

This matters because it is not another toy eval. We have spent years watching models draw SVG pelicans on bicycles or solve contrived puzzles. Karpathy’s test shows a model reasoning through a real creative engineering problem, managing scope, writing modular code, and delivering something you could actually ship. The post on X hit the front page of Hacker News with 188 points and 150 comments, most of them from engineers who recognize the gap between this and the usual benchmark theater.

For anyone running a company, the takeaway is simple. If a model can autonomously build a complex 3D scene from scratch for ten dollars, it can probably handle a lot of the grunt work your team does manually. Prototyping tools, generating reports, building internal dashboards, writing integration scripts. The cost and time floor just dropped again.

The hard part is not the model. It is knowing what to ask for, how to structure the task, and how to route the output into your actual workflow. That is where something like the Omni Command Centre comes in, giving you a single place to test prompts, chain tasks, and connect AI output to the tools you already use. You do not need to become a prompt engineer. You need a system that lets you experiment fast and fail cheap.

Karpathy is not selling anything here. He is just showing what is possible right now, with a model anyone can access. The question is whether you are set up to use it.