Simon Fin portfolio
Contact

Case study

AI Game Agents

An AI-assisted game-development experiment where a small team of role-based agents helps build a 2D trader simulation prototype.

Workflow Back to projects
Trader Agent prototype loopNPC trading, expedition and inventory flow.
StatusAI experiment / personal prototype
PrototypeTrader Agent, a small 2D shop/trader simulation
StackPhaser, Codex, Linear, Telegram bots, free assets
RoleProject owner, workflow designer, AI-assisted developer

Positioning

This is not an art case. It is a workflow case.

AI Game Agents uses a small web-game prototype as a test field for agentic development. The prototype is a simple trader simulation: NPCs come in, buy and sell items, can be sent on expeditions and bring new goods back into a basic economy.

The visual side uses free assets honestly; the value is the workflow: directing agents, structuring tasks, testing roles and checking whether an AI-assisted production process can create working output.

What it proves

  • Curiosity and hands-on use of modern AI development tools.
  • Ability to design workflows, not just prompts.
  • Understanding of PM, coder and QA roles in a small production loop.
  • Fast prototyping mindset around gameplay concepts.

Agent workflow

Role-based agents connected through Linear and Telegram.

The workflow is documented without exposing private tokens, chats or internal credentials, focusing on agent roles, task flow and QA signals.

DirectionSemyon sets concept and priorities
PM agentBreaks work into tasks
Coder agentImplements game changes
QA agentChecks outcomes and reports issues
Source of truthLinear plus Telegram handoffs

Prototype concept

Simple game loop, useful workflow test.

The game itself stays intentionally small so the case remains about process and automation.

1

Trading loop

NPCs enter the shop, buy items, sell items and change the state of a small economy.

2

Expeditions

The player sends NPCs out to bring back new items and inventory opportunities.

3

Agent iteration

The agent team receives tasks, implements changes and reports back through the workflow.

Media checklist

Proof media queue.

Use redacted, clean visuals. Do not publish private project data, bot tokens, raw chats or credentials.

NPC trading loop GIFBuy, sell, inventory, expedition return.
Short prototype video20-40 sec walkthrough of the current build.
Agent role diagramPM, coder, QA, Linear, Telegram, Codex.
Redacted Linear screenshotTasks, handshakes, status, no private data.
Telegram bot screenshotClean communication proof, no sensitive data.
Concept to playable prototype flowIdea, tasks, agents, implementation, QA.

Honest note

Free assets, workflow focus.

The visual assets in this project are not presented as original art. The case exists to show agentic development experiments, automation thinking, task structure, QA flow and rapid game prototyping.

Do show

  • Workflow diagrams and redacted task screenshots.
  • Playable prototype loops and before-after iterations.
  • What the agent workflow made easier or harder.