Game AI Pro 3: Collected Wisdom of Game AI Professionals

Source metadata

Type: Edited practitioner anthology
Editor: Steve Rabin
Year: 2017
Publisher: CRC Press / Taylor & Francis
Scope: 42 chapters across general wisdom, architecture, movement/pathfinding, tactics/strategy, character behaviour, and odds and ends

Volume 3 leans harder into production tooling and robustness than the earlier books: logging visualisation, in-game scrubbing, stress testing, and extensible factory systems are treated as central AI work rather than as support tasks.
The architecture section deepens themes already visible in the 360 anthology: modular AI, pitfalls in behaviour-tree design, BT + state machine hybrids, lightweight FSMs, and stronger treatment of utility-based considerations.
The tactics and strategy material reinforces several pages already in the wiki: Paragon lane-space/front-line logic, MCTS pitfalls, autogenerated spatial queries, and hierarchical portfolio search.
The character-behaviour section is especially useful because it includes the now-canonical Dragon Age: Inquisition utility scoring architecture, ambient rule-based interactions, and 1000-NPC performance lessons.
The volume also broadens the notion of AI work through recommendation systems, tag-based content selection, and debugging infrastructure.

The series is increasingly clear that production AI quality depends on tooling, not just on algorithms.
Utility AI becomes less abstract here and more obviously a design-facing scoring discipline.
Several chapters stress the same production lesson: make the system inspectable, replayable, and debuggable, or iteration speed will collapse.

Directly strengthens:

utility-ai — Dragon Age scoring architecture; choosing effective considerations
ai-architecture-patterns — modular AI, factory reuse, BT pitfalls, debug tooling
ai-state-machine-pattern — reusable lightweight FSM framing
game-ai-agent-design — BT/FSM hybrid thinking and broader architecture toolbox
pathfinding-algorithms and steering-behaviours — goal bounding, RVO/ORCA, locomotion, complex vehicle steering
mcts and strategic-ai-rts — MCTS pitfalls and hierarchical portfolio search
npc-performance-at-scale — Project Highrise lessons echoed by later chapters

Also relevant to future extraction:

Which debugging/tooling patterns from this volume should become explicit production pages rather than staying buried inside AI architecture summaries?
Should recommendation systems live under AI, analytics, or service-design in this vault?
How much of the modular AI/factory material belongs in the main AI route for students versus in an advanced architecture route?