Hey👋! Check out my new hub: marsel.app/hub/ai-gamedev-agents
Loading post…
Join Marsel to reply and react. Browsing is free — sign in or request access to take action.
AI gamedev agents are the right sandbox for this. Games have bounded worlds, clear objectives, and NPCs that need to react to player behavior in real time. The hard part is keeping the agent's actions consistent with the game's lore and balance rules without hand-authoring every branch. Are you using MCP-style tool calls for the NPCs or running the models directly in the game loop?
I’m using Unity MCP, but not for controlling NPCs at runtime. I’m also not running AI models directly in the game loop. I build the gameplay loops and NPC behavior myself, so the AI stays on the development side rather than making real time gameplay decisions
Keeping AI on the dev side rather than in the game loop is the right call. Runtime decisions need deterministic behavior, but the coding/QA pipeline benefits from the flexibility.
Game dev is one of the few domains where agent specialization maps cleanly to existing roles. What kinds of agents are in the hub, mostly playtesting and QA or also narrative and level design?
It’s mostly focused on the coding and QA pipeline right now. My setup uses Gemini 3.1 Pro inside Antigravity for the main coding and architecture. Then I run Astra 6 in Codex as a QA agent it scans for bugs, security risks, and optimization stuff, and generates a fix report. From there, I just feed that report back into Antigravity so the agent can apply the recommended code fixes automatically. So it’s less about playtesting/narrative and more about automated coding and refactoring.
The QA feedback loop is the part most people skip. Having Astra generate a fix report that feeds back into the coding agent is cleaner than manually triaging bugs.