How to build a theme park
The author reports that simple prompts now produce functioning parks with guests, rides, paths, rollercoasters and themed worlds, including guest-following features.
news.ycombinator.comSuggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Build [theme park] with [themed worlds]. Require complete rollercoaster tracks with a drop, ride entrances connected to paths and scenery matching each theme. Have a separate agent grade the park, update rules and skill files from its findings, and iterate until the park works as a whole.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- thealexanderlee
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reports that simple prompts now produce functioning parks with guests, rides, paths, rollercoasters and themed worlds, including guest-following features.
- Human involvement
- The author designed the rules and evaluation rubric; the post does not describe a separate human review of each generated park.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Adapted design-system constraints into theme park rules.
- Generated parks with Magic Patterns and graded them with another agent.
- Updated rules and skill files before generating another attempt.
The author built an agent that generates Rollercoaster Tycoon-influenced theme parks from prompts. The approach borrowed ideas from their work on Magic Patterns’ Design System Agent, where generated websites should respect a company’s existing components, brand and conventions.
For parks, the author translated those constraints into practical rules. Rollercoasters need complete tracks, rides need entrances connected to paths, paths must connect areas, and themed worlds should use matching scenery. Simply giving the model appropriate components did not produce sufficiently coherent results.
The author therefore built an evaluation loop. Magic Patterns generated a park, another agent graded it against a rubric, and the generation agent updated its rules and skill files before trying again. The rubric checked valid rollercoaster tracks with at least one drop, ride access through paths, appropriate scenery within each world and whether the whole park worked together.
After many iterations, the author reports that a simple request can produce a functioning park with guests, rides, paths, rollercoasters and themed worlds. Users can follow guests, read their thoughts and watch them ride. Who it's for: builders experimenting with generated simulation environments where connectivity and thematic consistency need explicit evaluation.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · thealexanderlee
Frequently asked questions
What did the park rubric check?
It checked complete valid rollercoaster tracks with at least one drop, path access to rides, appropriately themed scenery and the coherence of the whole park.
How did the agent improve between attempts?
Another agent graded each generated park, then the generation agent updated its rules and skill files before trying again.
More details
Remaining steps
- Repeated evaluations until simple prompts produced functioning parks.
About this account
The generation agent updated rules and skill files after a separate agent graded parks against a rubric.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=49452037. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: thealexanderlee on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Magic Patterns · Grading agent · Skill files
Theme parks · Evaluation loops · Agent skills
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.