Presentation: From AI Agent Demo to Production: Automated Testing

What Presentation actually is

Zhou Yu recently presented a comprehensive framework addressing why AI agents frequently stall in the prototype phase and how simulation-driven testing can move these systems into production. The presentation focuses on solving critical compliance and reliability bottlenecks that prevent multi-turn conversational agents from operating safely in real-world environments. By leveraging research and practical implementations from Columbia University and Arklex AI, the methodology outlines a systematic approach to automated evaluation, shifting AI development away from

InfoQ Homepage Presentations From AI Agent Demo to Production: Automated Testing and Evaluation AI, ML & Data Engineering When AI Accelerates Development, Can Your CI Pipeline Keep Up? Learn how Columbia and Arklex AI use synthetic user personas, trajectory entropy, and automated CI/CD pipelines to evaluate multi-turn agents, catch edge cases before deployment, and scale self-learning workflows in production.

How Presentation works

Bio Zhou (Jo) Yu is a Professor of Computer Science at Columbia University and the Founder of Arklex.ai. Her research has earned several Best Paper Awards at leading NLP conferences, and she was named to the Forbes 30 Under 30 list in 2018.

Why Presentation matters now

About the conference QCon AI is a practitioner-led event focused entirely on the engineering discipline required to scale these workloads safely. See the full write-up from InfoQ via the source link for quotes and complete context.

It provides direct access to the architectural playbooks and failure metrics that peer organizations use in production. INFOQ EVENTS September 17th, 2026, 1 PM EDT Beyond the PR: The New Control Plane for Agentic Software Delivery Presented by: Mohit Suman - Staff Product Manager at Harness October 8th, 2026, 12 PM EDT When AI Accelerates Development, Can Your CI Pipeline Keep Up?

Who Presentation changes things for

I've been working with various open-source models for pre-training, post-training, and so on. One of the things that we noticed today and we have seen in a lot of reviews is like 95% of agents are staying in demos.

What to watch after Presentation

They're very interesting to play with, they're very nice demos to look at, but they can't really create real values, and really making changes within the process we have been experiencing for many years. See the full write-up from InfoQ via the source link for quotes and complete context.