Ship Agents: A Virtual Conference

Many AI agent systems fail not because of model quality, but because of weak data architecture, observability, and system contracts. In this talk, I'll share production lessons from building enterprise-scale RAG and agentic platforms, covering intent-driven data design, retrieval reliability, latency/cost tradeoffs, and AI-to-system integrations (MCP/tool calling). Attendees will learn how to design agent systems that remain reliable beyond demos.

I have built and deployed a production grade sales assistant agent at Nvidia which is live and improving over the past 12 months. Sharing some hard learned lessons from the ground. It's not just about a single system deployed once but an entire eco-system which continuously improves as we scale and upgrade and at times reduce and rewire to let the agent do more.

Teams want to price AI agents by resolution, conversation, or task, yet most can't clearly explain what a single outcome costs. This talk covers lessons from building a production system to measure cost per outcome, the real-world challenges encountered, and the design patterns that enabled outcome-based pricing.

AI agents excel at reasoning but remain "socially blind," lacking the "wisdom of the crowd" found in traditional recommendation pipelines. While we can already predict user needs at scale, this intelligence is rarely integrated into LLMs. This talk explores the intersection of generative models and industrial-scale retrieval, evolving static tools into agents that leverage collaborative intelligence to bridge the gap between logical reasoning and human intuition.

As AI agents transition from experimental demos to enterprise production, the "happy path" of autonomous execution often collides with the messy reality of legacy integrations, non-deterministic failures, and "infinite loop" costs. This session moves past the hype to explore the actual state of agentic systems in 2026. We will dissect common production architectures—from single-agent tool loops to hierarchical swarms—and reveal the most frequent failure modes encountered in the field, such as context abandonment and coordination silent-failures. Finally, we’ll discuss the "Engineering Pivot": why teams are trading raw autonomy for deterministic workflows, multi-layer guardrails, and session-level tracing.

AI agents are becoming primary users of developer tools -- and most devtools aren't ready. We built the first benchmark to evaluate agent readiness. Based on five distinct factors, we evaluate how easy it is for AI agents to adopt devtools. After evaluating 40+ tools across search, auth, databases etc., we've uncovered key fundamentals behind great AX. This talk covers the patterns that make tools agent-ready along with actionable insights for any team building for the agentic future.

Context Engineering - Stop Agents from Choking on Their Own Data