Uploaded September 2026 | Updated November 2026, 1 day ago
How do you maintain system performance when AI agents ship 70% more PRs and your user base scales to 900 million weekly active users?
In this talk, Martin Spier (Head of ChatGPT Performance Team at OpenAI, ex-Netflix) breaks down the compound impact of agentic development on system architecture. As developer workflows shift toward multi-threaded AI execution, traditional performance engineering must evolve. Martin reveals how OpenAI scales non-GPU infrastructure, restructures latency budgets, and deploys autonomous performance agents to fix regressions before production breaks.
⏱️ Video Timestamps (For Navigation)
00:00 - The Dual Acceleration: Rapid Growth & Agentic Coding
02:15 - Scaling ChatGPT to 900M Weekly Active Users
04:30 - The Hidden Engineering Cost of AI-Generated PRs
07:10 - Why Non-GPU Infrastructure (CPU & Memory) Bottlenecks AI
09:50 - Measuring What Perception Matters: Deconstructing Latency
12:40 - Rethinking Performance Engineering: The Always-On Agentic Loop
15:35 - Building Autonomous Safeguards, Benchmarks, and Observability
18:10 - The Future Role of Engineering Leaders in an AI-First Stack
🔗 Transcript available on InfoQ: bit.ly/3U7tHnJ
#SoftwareArchitecture #OpenAI #PerformanceEngineering #AgenticAI #DevOps
How do you maintain system performance when AI agents ship 70% more PRs and your user base scales to 900 million weekly active users?
In this talk, Martin Spier (Head of ChatGPT Performance Team at OpenAI, ex-Netflix) breaks down the compound impact of agentic development on system architecture. As developer workflows shift toward multi-threaded AI execution, traditional performance engineering must evolve. Martin reveals how OpenAI scales non-GPU infrastructure, restructures latency budgets, and deploys autonomous performance agents to fix regressions before production breaks.
⏱️ Video Timestamps (For Navigation)
00:00 - The Dual Acceleration: Rapid Growth & Agentic Coding
02:15 - Scaling ChatGPT to 900M Weekly Active Users
04:30 - The Hidden Engineering Cost of AI-Generated PRs
07:10 - Why Non-GPU Infrastructure (CPU & Memory) Bottlenecks AI
09:50 - Measuring What Perception Matters: Deconstructing Latency
12:40 - Rethinking Performance Engineering: The Always-On Agentic Loop
15:35 - Building Autonomous Safeguards, Benchmarks, and Observability
18:10 - The Future Role of Engineering Leaders in an AI-First Stack
🔗 Transcript available on InfoQ: bit.ly/3U7tHnJ
#SoftwareArchitecture #OpenAI #PerformanceEngineering #AgenticAI #DevOps










