BUILDING RELIABLE LLM AGENTS: WHAT WORKS AND WHAT DOESN’T

​2025 is shaping up to be the year of AI agents – but are we ready for it?
The race to deploy LLM agents is accelerating, but even the biggest players – Anthropic, Microsoft, Amazon, Airbnb, and Dropbox – are running into real-world production hurdles. These aren’t hypotheticals; they’re drawn straight from case studies in the LLMOps database—a collection of over 300 deployments from the last two years. And when LLM agents fail, they don’t just fizzle – they hallucinate, misfire API calls, and sometimes take entire workflows down with them.

​Join Alex Strick van Linschoten (ZenML) for a live-streamed conversation with Hugo Bowne-Anderson as they break down how companies are navigating the harsh realities of deploying agents in production. This session digs into the architectures that are succeeding—and the disasters that happen when things go wrong.

​In this live episode, we’ll cover:

• ​The Agent Reality Check:
  Why most production LLM deployments still lean on structured workflows over fully autonomous agents – and why that’s often the right call.

• ​Architectural Patterns from the Field:
  How Anthropic, Dust.tt, Amazon, and Klarna are using ReAct, RAG, and microservices to build agents that don’t derail in production.

• ​Hard-Won Lessons from Production:
  What happens when agents hallucinate, how scaling introduces fragility, and how teams like Dropbox and Slack are addressing observability and security gaps.

• ​Actionable Takeaways:
  How to integrate LLMs into your pipelines as predictable tools, while avoiding the common pitfalls that lead to unreliable agents.

​If you’re building LLM-powered systems or scaling agents in 2025, this session will equip you with the insights you actually need.

​

​About Alex Strick van Linschoten

​Alex is a Machine Learning Engineer at ZenML, where he researches the realities of deploying LLM agents at scale. His latest work on LLM Agents in Production distills insights from hundreds of deployments, offering practical guidance for teams navigating LLMOps.

​

​Alex’s focus lies at the intersection of LLM architectures, observability, and operational challenges—helping teams transition from ambitious demos to stable, production-grade systems.