You Wouldn't Let a New Developer Sling Code at Production Without Guardrails with Patrick Wyatt

To view slides from this talk, along with Patrick’s other Game Developers Conference talks and interviews, visit https://www.codeofhonor.com. Game industry veteran Patrick Wyatt—who, among other accomplishments, was a founder of ArenaNet, VP of Blizzard, and lead developer of _Warcraft_ _and_ _Guild_ _Wars_ —shares hard-won lessons about AI coding disasters and the guardrails he's developed to prevent them. Drawing from 30 years of building online games and recent experiences publishing iOS apps with AI-written code, Patrick reveals how he's transformed from an AI skeptic to someone running multiple AI agents in parallel to dramatically accelerate development. Learn why treating AI like a "PhD-level junior developer" requires the same guardrails you'd use for any new team member. (You’re not trying to prevent accidents from happening. You’re just trying to “reduce the scope of damage.”) Along the way, Patrick demonstrates practical techniques that actually work: using Git hooks to prevent AI from bypassing tests (yes, they'll try), running multiple agents simultaneously with Git worktrees, implementing deterministic simulation for instant bug reproduction, sandboxing AI to protect your AWS credentials and SSH keys, and more. Watch now to discover why test-driven development is perfect for AI agents, how to reduce context window waste, and why AI will sometimes comment out failing tests rather than fix them. Stick around until the end for Patrick's vision for the future: specialized AI agents with restricted capabilities working together, and personalized software where users' own AI agents create custom plug-ins tailored to their needs. Follow O'Reilly on: LinkedIn: https://www.linkedin.com/company/oreilly/ Facebook: http://facebook.com/OReilly Instagram: https://www.instagram.com/oreillymedia BlueSky: https://bsky.app/profile/oreilly.bsky.social