A Practical Breakdown of AI Agent Failures in Production

Hosted by Dr. Aki Wijesundara

209 students

Share this lesson

Go deeper with a course

Build AI Agents with n8n
Aki Wijesundara, PhD and Manu Jayawardana
View syllabus

What you'll learn

How to identify common agent failure modes

Learn why agents hallucinate, loop, misuse tools, or fail silently in real production environments.

How to diagnose where agents break

Understand how to reason about failures across planning, tool use, memory, and execution boundaries.

How to design constraints and guardrails

Learn practical techniques to limit agent behavior using structure, stop conditions, and controlled tool access.

How to decide when not to use an agent

Understand when simpler workflows or deterministic systems are the better choice.

Why this topic matters

AI agents often shine in demos but struggle in real world use, looping, hallucinating, misusing tools, or behaving unpredictably. These failures rarely come from the model alone. This session breaks down agent failures as system design problems and shows how better structure, constraints, and safer failure modes lead to more reliable, controllable agents in production.

You'll learn from

Dr. Aki Wijesundara

AI Founder | Educator | Google AI Accelerator Alum

Aki Wijesundara is an AI researcher, educator, and founder with a PhD in Machine Learning and experience at Google's AI Accelerator, where he contributed to advancing real-world applications of machine learning and large language models. As Co-Founder of Snapdrum, he helps organizations design, build, and deploy applied AI systems that move beyond prototypes to production-scale impact.

Aki has taught over 5,000+ engineers, founders, and product leaders worldwide how to build production-ready AI systems using LangChain, RAG pipelines, and MLOps frameworks.

He serves as AI Advisor to the United Nations, contributing expertise on responsible AI implementation and policy.

His work bridges cutting-edge research and practical implementation across text, voice, and multimodal AI.

Alongside Snapdrum, Aki serves as CTO of Krybe, an AI voice automation startup, and CTO of Tilli, an edtech company redefining how learners engage with AI-powered education.

Previous students from

Google
Meta
OpenAI
Amazon Web Services
NVIDIA

Watch this lesson for free

By continuing, you agree to Maven's Terms and Privacy Policy.