The Artificial Intelligence Podcast with Dr. Tony Hoang
Interview #107 Helen Gu, Founder & CEO, InsightFinder AI
Helen Gu, Founder & CEO at InsightFinder AI, breaks down why most AI agent failures are silent: no error code, every dashboard green, just a confident wrong answer. She explains why read-only access is mostly safe while any write access needs guardrails and a human who stays accountable, how one mistake travels down a chain of agents from a misrouted task or a corrupted upstream price, and why debugging multi-agent systems echoes the move from mainframes to distributed systems. Helen also dives into why you can't use ChatGPT to grade ChatGPT, how runaway token use and rising latency expose agents stuck in loops long before anything errors, and how teams can catch a vendor's model update by watching the distribution of outputs shift. Finally, she shares why AI agents, like distributed systems, can never be proven 100% reliable, and why the monitoring should be done by other agents.
Listen to this episode:
Be a guest on the show
Helen Gu of InsightFinder AI joined Dr. Hoang for this conversation. If you lead AI work at your organization, apply to be a guest, or have your PR team apply for you.
The AI Decision Course: 20 courses and 271 short lessons for directors, VPs and executives. One lesson from each course is free to read.