95% of GenAI pilots fail due to feedback integration issues, not deployment challenges. Alessandro Cappelli argues that Reinforcement Learning (RL) provides the only systematic way to incorporate business metrics and production signals to continuously improve models, especially for complex agent-based systems.
This talk explores the architectural shift required to build durable, long-running AI agents. It argues against traditional replay-based durability, proposing a two-part solution: treating LLM context as a durable, append-only log and managing the compute state (memory, files, processes) with OS-level snapshot and restore, implemented efficiently using Firecracker microVMs.
Sequoia Capital partners argue that AI is a revolution in computation, not just communication. They explain how long-horizon agents are reshaping every layer of work, comparing the current shift to the Industrial Revolution's impact on manual labor and suggesting that innovations that once took a century are now possible in a hundred days.
Malte Ubl, CTO of Vercel, posits that AI engineering is the successor to web development, arguing that AI agents will expand, not shrink, the software market. He explores practical agent archetypes being built today and discusses the profound shift required in infrastructure and security as agents become both the builders and primary users of software, concluding that the true innovation and value will lie in the application layer built by AI engineers, independent of the commoditizing foundational models.
Johann Schleier-Smith from Temporal explains Durable Execution, a paradigm for building reliable, long-running applications. He details how Temporal's model of deterministic workflows and stateful activities provides a robust alternative to traditional checkpointing and event-driven architectures, especially for complex, LLM-driven agentic systems.
A practitioner's guide to deploying a text-to-SQL agent in a real-world business environment. The talk covers the critical lessons learned in moving from concept to production, focusing on the importance of the communication channel (Slack), the necessity of a semantic layer over benchmark scores, and a pragmatic approach to system architecture, testing, and evaluation.