An Engineer Turns Complex Topics Into Playable Simulations With AI Agents
Instead of asking a chatbot to explain a subject, developer Laurentiu Raducu has coding agents build a verified knowledge base, then a low-poly simulation of it. His first result, ChipTycoon, follows sand through chip fabrication.
What happened
On 9 August 2026, engineer Laurentiu Raducu published a learning workflow that reached the top of Hacker News. His premise: explanations written by a model are often too shallow and emoji-heavy to stick, so he stopped asking for explanations and started asking for simulations.
The four-step flow
Working in plan mode with a coding agent such as Claude Code or OpenCode, he first has the model assemble a foundational knowledge base on a topic. Second, he asks it to review that base for accuracy. Third, the agent builds a low-poly, Rollercoaster Tycoon-style animation of the subject, with firm UX rules: readable on large and small screens, with controls to stop the flow at any point. Finally, the project is pushed to a new repository and published with GitHub Pages.
Why it matters
The idea came while Raducu was studying bottlenecks in data-center buildout and realised how little he knew about chip manufacturing. His first build, ChipTycoon, follows a cart from sand collection to a finished chip delivered to a data center, so every production step becomes something you can watch. He argues this beats scattered articles or bulleted lists, and suggests adding quizzes and puzzles for retention. The caveat: accuracy still rests on the same model reviewing its own knowledge base.