Add too little oversight and an agent makes a call nobody approved. Add too much and you've rebuilt the manual process you were trying to replace. Human-in-the-loop controls work when agents move at human speed. As agents act faster and across more systems, teams have to decide which decisions need a person, which need a guardrail and which can run on their own. Where should that line sit? This session brings together leaders building and operating agentic systems to break down how they set autonomy levels, design approval points and build guardrails that scale with the agent rather than against it. Through candid examples of where review added real protection and where it only slowed things down, they'll share how they decide how much control to hand over. Key takeaways: → How to classify agent decisions by risk and set autonomy thresholds for each → Where human review adds real safety and where it only adds friction → Guardrail and accountability patterns for agents acting at scale