What does AI agent tracking actually look like in practice? In this clip from our Superweek Session, Jordan Peck runs a live demo using a pretend agentic travel assistant built on Claude, similar to the Kayak example from earlier in the session.
The demo shows a hybrid tracking approach in action:
- Client-side events capture what the user sends (e.g. "book me a flight from London to Paris")- Agent self-tracking fires events on every decision the agent makes, including why it made that decision, its confidence level, and which tools it called
- Tool execution events capture the inputs, outputs, and results of each tool the agent uses
- Agent context is attached to every single event, enabling things like A/B testing different models or system prompts
The result: a rich, structured event stream that gives you both the business analytics and the decision-level visibility that other approaches miss.
🎥 Watch the full session: https://snowplow.io/events/track-ai-agents-in-2026