FULL STORY

Indie Dev Ships Games Fast with AI Agent Astra

Indie developer Dimillian used the Astra agent to build two playable games, refining them through rapid feedback loops where AI plays and iterates alongside him.

2026-09-05 ~ 2026-09-06 · 3 episodes · 11 posts

Episode 1 · Developer builds explorable universe game with Astra (2026-09-05, 4 posts)

Developer Dimillian used OpenAI's Astra model to build Void Explorer, an online space game with real Earth-sized planets and star systems, praising its procedural 3D generation—though the demo only got 20k views.

Episode 2 · Dev Uses Astra Agent's Fast Feedback Loop to Polish AI-Generated Game (2026-09-06, 2 posts)

Dimillian reports spending 90% of his time collaborating with Google's Astra agent, using a play-and-fix feedback loop to iterate on fast travel, combat and first-person controls.

Episode 3 · Dev Builds Playable Daggerfall-Inspired Game Codexfall End-to-End with Astra (2026-09-06, 5 posts)

Developer Dimillian demonstrated building Codexfall, a Daggerfall-inspired procedurally generated game, end to end with the AI coding tool Astra, as a test of how well an AI agent can bring a full game world to life. The game is playable in the browser and serves as a concrete benchmark for AI agents on large-scale projects.

Confirmed

  • Workflow: Dimillian had Astra list all systems the game needed, turn the list into a roadmap document for tracking, then build the first vertical slice containing all core systems in ultra mode.
  • Content: Codexfall is a large procedural world with towns, dungeons, combat, talkable NPCs, plus magic and guild systems — a first-person dungeon crawler themed around steel, magic and ruin exploration.
  • End-to-end generation: the trailer, including its soundtrack, was made entirely by Astra; entering any seed on the new-game screen generates a brand-new world.
  • Iteration: Dimillian iterated from actually playing the game, polishing fast travel, combat feel and the first-person perspective.

Why it matters

  • The project shows AI agents covering not just code but system planning (roadmap docs), music, promo video, and seed-based procedural content generation — a fairly complete stress test of end-to-end agent development.