Production AI Agents, Beyond the Demo - What breaks when agents meet real customers?

Thursday, July 16 | 1:00 PM PT | Online

AI agent demos are everywhere. Production AI agents are still much harder to get right.

The real challenge begins when an agent leaves the sandbox: when it meets real customers, handles real workflows, makes decisions your team has to stand behind, and creates outcomes someone has to own.

For many engineering and technology leaders, this is the moment where the conversation changes. It is no longer just about what agents can do. It is about what they can be trusted to do reliably, safely, and at scale.

We’ll discuss

1. The gap between demo and production: Where teams are actually running agents today, what remains stuck in pilot mode, and what is keeping promising projects from shipping.
2. What happens when agents meet real users: The issues that only appear outside controlled tests, what breaks first in production, and what those moments teach teams about readiness.
3. Reliability without determinism:How leaders make non-deterministic systems dependable enough for customer-facing workflows, and where they draw the line between human-in-the-loop and full autonomy.
4. Monitoring, evaluation, and trust: How teams know whether an agent is working when there is no single correct output, and how they catch quiet degradation before customers do.

Hosted by

Daniel Monahan has built and shipped production AI agents for real customer-facing workflows, including three Spanish-language voice AI agents used for customer quoting and outbound at Sigo Seguros. His work spans agent architecture, reliability, governance, and the operational patterns required to keep AI agents useful, observable, and controlled in live production environments.

Join this conversation with eng leaders working through one of the most important questions in AI adoption today:

How do we move agents beyond impressive demos and into production systems we can actually trust?