Spying on AI Agent (For Good): Observability for Hidden Conversations Tweet

Proposal: (エントリー済み)

AI agents don’t just answer questions they talk to each other, call APIs, and move data in ways that often remain invisible. These hidden conversations can quickly become reliability and compliance risks: runaway API calls, unmonitored data exchanges, or silent failures that only surface when users complain.

In this talk, Kaiwalya will explore how observability practices can shine a light on these invisible workflows. You’ll learn how to trace agent-to-agent interactions, capture telemetry from API calls made by LLMs, and correlate usage patterns with system health.

Through a live demo, we’ll show how to:
- Instrument AI agent communications without changing code.
- Monitor and visualize agent-driven API traffic in real time.
- Detect bottlenecks, failures, and unexpected dependencies.
- Feed this insight back into SRE workflows for reliability and governance.

If AI agents are the new microservices, observability is the only way to keep them reliable, secure, and predictable. Come see how to turn black-box AI behavior into clear, actionable insights.