GPT-6 Astra beats WoW starting zone in 40 minutes without seeing a single frame
量子位 · wechat · 2026-10-04
The open-source project agent-wow hooked GPT-6 Astra into a World of Warcraft private server (AzerothCore, client 3.3.5a) without ever showing it a frame of the game. Instead, Astra subscribed to 28 types of raw server packets (SMSG messages), built a world model in memory, and cleared the orc starting zone's full quest chain in 40 minutes with zero deaths.
Highlights: It followed the quest chain in order, sold junk loot, upgraded gear, learned skills before entering the final cave, and batched two cave quests into one trip — all veteran-player strategies it inferred on its own, with no prompt hints. It also exploited a collision-detection mismatch between the navmesh and terrain to clip through walls.
How it works: The framework defines no game logic — it only exposes a module system and lets the AI build its own capabilities. Perception comes from a Python script polling packets; quest intel is mined directly from AzerothCore's SQL database (like reading the Wowhead database instead of the website); pathfinding is a C++ program Astra itself generated, loading the mmaps navmesh via Detour and outputting JSON waypoints. The developer started the entire build with a single Codex prompt at high reasoning effort.
Next steps: testing whether one agent can level 1→80 solo, and whether 25 agents can raid Icecrown Citadel on heroic difficulty.
More from coding & agent
- What's the Actual Difference Between AI Models and AI Agents? — HarperSCarroll · 2026-11-20
- Victor Taelin's Bend2 talk from SF Lean Meetup explains the rationale behind the language design — pbaylies · 2026-11-20
- Drex Router cuts small-game build cost to $1.26 vs $13.44 on Claude Opus alone — testingcatalog · 2026-11-20
- User teaches AI companion to ignore "drop it" said to his dog, then fixes the over-pause — manosaie · 2026-11-20
- NVIDIA Dynamo cuts RL rollout bottlenecks with KV-aware routing and delta weight updates — NVIDIAAI · 2026-11-20
- LangChain Academy Answers: Can a Better Model Fix a Broken Agent? — LangChain · 2026-11-20