Uploaded August 2026 | Updated November 2026, 1 day ago
When an autonomous AI agent fails in production, developers almost always blame model intelligence. But what happens when the model knows exactly what to say, yet your system still breaks?
In this episode, Laurie Voss (Head of Developer Relations at Arize AI) breaks down why the agent harness, specifically code-level guardrails, is the most critical boundary layer in modern software engineering. From voice agents starting to speak over themselves to financial evals completely failing due to temporal context bias, Laurie unpacks why prompts aren't enough and how to build true self-improving software loops.
Chapters:
00:00 Introduction
00:25 How autonomous are agents becoming?
00:58 What matters most in an agent harness?
01:18 What happens when the evaluator is wrong?
01:50 What does a self-improving loop look like?
Resources:
🔬 Phoenix (open source): phoenix.arize.com
🔗 Arize AX: arize.com
📖 OpenInference: github.com/Arize-ai/openinference
📖 Phoenix docs: docs.arize.com/phoenix
Subscribe to Arize AI to watch the full "Rise of the AI Engineer" interview series! 🚀: youtube.com/@arizeai?sub_confirmation=1
#AIEngineering #AIAgents #ArizeAI
When an autonomous AI agent fails in production, developers almost always blame model intelligence. But what happens when the model knows exactly what to say, yet your system still breaks?
In this episode, Laurie Voss (Head of Developer Relations at Arize AI) breaks down why the agent harness, specifically code-level guardrails, is the most critical boundary layer in modern software engineering. From voice agents starting to speak over themselves to financial evals completely failing due to temporal context bias, Laurie unpacks why prompts aren't enough and how to build true self-improving software loops.
Chapters:
00:00 Introduction
00:25 How autonomous are agents becoming?
00:58 What matters most in an agent harness?
01:18 What happens when the evaluator is wrong?
01:50 What does a self-improving loop look like?
Resources:
🔬 Phoenix (open source): phoenix.arize.com
🔗 Arize AX: arize.com
📖 OpenInference: github.com/Arize-ai/openinference
📖 Phoenix docs: docs.arize.com/phoenix
Subscribe to Arize AI to watch the full "Rise of the AI Engineer" interview series! 🚀: youtube.com/@arizeai?sub_confirmation=1
#AIEngineering #AIAgents #ArizeAI










