Everyone is building AI agents. Far fewer are engineering the loop that makes an agent more than a clever one-shot prompt. The real power—and the real difficulty—of agentic systems lives in the feedback loop: perceive, reason, act, observe, learn, repeat. Get the loop right and you get systems that adapt, self-correct, and compound in capability over time. Get it wrong and you get runaway costs, infinite retries, silent drift, and confident-but-wrong actions at scale.
This session treats the agentic loop as a first-class engineering discipline. We'll break down the anatomy of a well-engineered loop—grounding agents in real signals, designing termination and escalation conditions, managing context and memory across iterations, controlling cost and latency per cycle, and building in human checkpoints for high-impact decisions. I'll use autonomous quality engineering as a concrete, battle-tested case study: agents that generate, execute, and self-heal work while reasoning over production telemetry, delivering measurable gains alongside real governance challenges. You'll leave understanding why the loop—not the model—is the unit of design, with practical patterns for building agentic systems that are reliable, observable, and accountable rather than impressive demos that fall apart in production.