A speed-run through the fundamentals of graphs for AI engineers, covering how to build effective graphs from unstructured text using schema-first approaches and entity resolution, and then exploring powerful graph-native algorithms like Personalized PageRank, Shortest Path, and Subgraph Matching, all illustrated with real-world applications to create smarter, cheaper, and more reliable AI solutions.
Matt Palmer discusses how code, empowered by AI, is becoming the fastest and most efficient medium for technical content creation, emphasizing that while code is cheap, structure and conscientiousness are now the most valuable assets for producing high-quality output. He predicts the rise of "content engineers" who leverage declarative pipelines to automate technical communication.
Armanas Povilionis introduces Froglet, an open-source protocol addressing the critical need for verifiable agent-to-agent collaboration in AI workflows, especially in scientific research. Froglet enables agents to discover, transact with, and receive tamper-proof receipts for services across organizational boundaries, transforming them into 'executive chefs' managing distributed supply chains rather than isolated 'cooks' with more tools. The protocol streamlines interactions into a signed flow (Descriptor -> Offer -> Quote -> Deal -> Receipt) and integrates with diverse agent harnesses and execution environments, abstracting complex operations for LLM-driven agents.
AI teams are deploying advanced agent systems without the fundamental safety mechanisms (feature flags, canaries, kill switches) that web teams adopted over a decade ago. This oversight leads to critical incidents like data deletion and financial loss. This talk introduces six agent-specific feature flag types—for prompts, tools, models, memory, autonomy, and sub-agents—and outlines a practical playbook for secure AI deployment, emphasizing the critical role of a pre-wired kill switch to manage the high blast radius of AI agents.
Most AI agents lack a "save button" for persistent state, making debugging and optimization challenging. This talk introduces the concept of a durable runtime with checkpointing, allowing developers to replay agent executions, perform "what-if" analyses (e.g., swapping models, mocking tools), and conduct cohort-based evaluations using production data. The open-source tool Kitaru is demonstrated as a solution to build, debug, and improve agents at scale by capturing full execution state, not just traces.
An Oxford-style debate exploring the gap between the hype and practical reality of "loops" in AI/ML development. Experts discuss their history, optimal anatomy, future role in software factories, and challenges like security, economic viability, and the imperative for strong engineering discipline.