Give the outbreak believable roots in real biology.
HAYLEY HENEGHAN · AI WORKFLOW CASE STUDY
I built quality control into multi-AI
production.
Alien Farming is a cozy space-farming game built from biology research, story and interconnected game systems. I used several AI assistants to develop it, then designed separate review and approval steps so a convincing draft couldn’t quietly become a bad game decision.
THE PROJECT
Four kinds of work had to agree on one game.
A detail about fungi could change the mystery. That could change a game mechanic, and the mechanic still had to feel gentle enough for a cozy game. With several AI assistants working at once, one unsupported idea could spread through every document.
Keep the same mystery consistent from planet to planet.
Make the connected systems easy to understand and forgiving to play.
Keep a record of what was approved as the game changed.
ONE SHARED PROJECT RECORDResearch → human decisions → story and systems → game design
THE PRODUCTION RISK
Each assistant could produce useful work quickly. The risk was drift: one confident answer could contradict earlier research or quietly invent a new rule.
PROJECT RESEARCH18 reportsCompleted and indexed before design work continued
DOCUMENTS COMPARED4 core filesThe game plan was checked against its supporting records
FIRST FULL REVIEW7 issues caughtIncluding missing links between decisions and the main game document
THE WORKFLOW I DESIGNED
Production moved fast.
Approval stayed separate.
FINAL AUTHORITYHUMAN
OWNERVision · decisions · scope
OWNERVision · decisions · scope
01CREATEOne assistant explored ideas and drafted
02INTEGRATEAnother checked the draft against the project record
03REVIEWA separate reviewer looked for conflicts and unsupported claims
04RELEASEOnly the reviewed version entered the official documents
→
→
→
WHERE QA FITS
The assistant that wrote something couldn’t approve it. A different role checked the work, and I made the final call.
QA IN PRACTICE
The QA review found a problem the drafting assistant missed.
The 241-line game plan summarized 83 recorded decisions but didn’t point back to any of them. It looked complete. But if one decision changed later, no one could reliably find every passage that needed to change with it.
A polished AI draft is only useful if a team can verify and update it.
WHAT A TEAM GETS
AI can move faster without making the work harder to trust.
AI handles exploration without treating its first answer as final.
A separate reviewer checks the work before it spreads.
The team can see where a claim came from and update it safely.
Tools can recommend. A human decides what becomes official.
HONEST BOUNDARY
This case study covers the workflow, strategy and documentation. The game is still in development and isn’t presented here as a finished release.