Agentic AI in Production: From Prototype to Reliable Systems
Presentation on the engineering challenges of deploying agentic AI systems in production — covering evaluation, monitoring, failure modes, and the gap between demos and reliable production systems.
Abstract
Agentic AI systems look impressive in demos. They fall apart in production. This talk covered the practical engineering disciplines needed to ship agentic AI that actually works — and keeps working — in real-world environments.
Key Topics
- The Evaluation Pyramid for agentic AI: retrieval → generation → system-level → human
- Common failure modes in multi-agent LangGraph pipelines
- LangSmith for production tracing and debugging
- CI/CD quality gates: how to enforce evaluation thresholds before deployment
- Monitoring embedding drift, hallucination rates, and latency in live systems
Event
Convergence AI Dallas is a major DFW AI conference bringing together AI practitioners, researchers, and industry leaders. The March 2026 edition was held at the Irving Convention Center over two days (March 30–31).