Build and Continuously Improve Agents That Work in Production (with Agents)

# **Frontier models are more capable than ever. So why do so many agents still fall over in production?**

There's still a big gap between what a model can do in a demo and what an agent can reliably deliver at scale. Join Triangle Tech Talks for a practical session on why that gap exists and how teams are closing it.

**About the talk**

**Marc Klingen**, Co-Founder of **Langfuse**, will cover the real challenges of turning capable models into useful agents: tool use, reliability, evaluation, monitoring, and handling how real users actually behave.

Marc will also explain why the Langfuse team started building open-source observability and evaluation infrastructure, what has changed since then, and what state of the art looks like today. Using Langfuse as a working example, he'll show how to:

* 🔍 **Trace agents in production** so you can see what's happening inside them

* 🧩 **Learn from failures** and find out why an agent went wrong

* 📊 **Turn real interactions into evaluation data