Inside the Architectures Powering Modern AI Systems: QCon San Francisco 2025

"Senior software engineers are being asked to design and verify AI architectures without established patterns, modernize systems under delivery pressure, and make long-term technical decisions while core technologies are still evolving. With AI adoption accelerating faster than best practices can emerge, teams need validated patterns from production systems, not experimental frameworks. The gap isn't conceptual; it's implementation: How do you actually build reliable AI infrastructure at scale? How do you de-risk agentic systems? What does a production RAG pipeline look like at 100 million users?"

"GenAI infrastructure at 100M-user scale: Maggie Hu, group product manager for the Core AI Platform, and Merrin Kurian, distinguished engineer for AI Foundations at Intuit, break down their production stack: vector stores, prompt management, RAG pipelines. Agentic system architecture: Adam Wolff, engineer on Claude Code at Anthropic, shares the architectural tradeoffs from building the first agentically accelerated software project. Speed over complexity is the theme."

""Senior engineers don't need another talk on 'what is GenAI'", said Hien Luu, QCon program committee member. "They're asking: 'How do I build production-grade infrastructure for this? How do I validate my architecture for an agentic system?' These sessions answer those questions from people who've already built it"."

Senior software engineers face pressure to design and verify AI architectures without established patterns, modernize systems under delivery timelines, and make long-term decisions while core technologies evolve. AI adoption is outpacing the emergence of best practices, so teams require validated production patterns instead of experimental frameworks. The core challenge is implementation: building reliable AI infrastructure at scale, de-risking agentic systems, and creating production RAG pipelines for very large user bases. QCon San Francisco 2025 gathers practitioners who solved these problems to share architectural decisions, implementation constraints, and lessons from teams at major companies.

Read at InfoQ