February 28, 2026
Playing God, forgot the screenshots?
Show HN: Decided to play god this morning, so I built an agent civilisation
A DIY digital world evolves itself; readers cheer, jeer, and ask for pictures
TLDR: A developer launched a local, open-ended digital world where tiny agents evolve on their own. The crowd split between excitement for emergent storytelling and frustration over an AI-written README, missing images, and fears of reward-hacking chaos—proof that even god-mode needs good docs and vibes.
The creator of Werld hit “play god” and spun up a tiny digital civilisation: little software critters that sense, act, reproduce, and evolve—all running locally with a dashboard to watch the chaos unfold. The community? Split between awe and eye-rolls. Fans like alexhans swooned over the promise of emergent storytelling—think SimCity meets Dwarf Fortress—calling for better UX and narrative tools so humans can enjoy the spectacle. But the loudest chorus was pure drama: dash2 roasted the AI-written README with, “if you won’t write it yourself, why should I read it?” while fd-codier delivered a meme-tier jab: “No images in the README…” Cue the collective cry for screenshots. The bold claim of “no hardcoded behaviors, no reward functions” sparked a spicy safety debate—mpalmer warned it could devolve into an “agentic opium den,” aka bots hacking their own pleasure centers. Meanwhile, a classic AI in-joke dropped via gyomu’s Minsky/Sussman riff, proving the thread had both philosophy and snark. The vibe: wild potential meets documentation drama, with everyone agreeing on one thing—if you’re playing god, at least show us the pictures and tell us a good story.
Key Points
- •Werld is a local, open-ended evolutionary simulation where agents evolve on a Watts–Strogatz small-world graph using NEAT-style neural networks.
- •The simulation is pure Python (3.10+) with standard library only, featuring checkpoints, milestones, and SIGTERM-safe execution.
- •Agents have 29 heritable genome traits, continuous effectors, and evolvable sensory gains; no hardcoded behaviors or reward functions.
- •Werld Observatory is a Next.js dashboard with 13 sections that polls a SQLite database every 4 seconds to visualize the simulation.
- •Quick start: run `python main.py` for the sim (with optional flags) and start the dashboard via `npm install` and `npm run dev`; the project is MIT-licensed and open to contributions.