Uploaded August 2026 | Updated September 2026, 2 weeks ago
When an autonomous AI agent fails in production at 2 AM, finding the bad output is only the beginning. You can't read an agent's logic through raw code like traditional deterministic software; traces are the new source of truth. Finding the failure pattern, tracing it to the underlying root cause, writing a fix, and verifying that the change improved the entire agent harness traditionally takes days of manual human triage.
In this technical deep dive, we demonstrate how to invert the debugging loop using Arize AX. Learn how to shift your engineering team from first responders to reviewers by building an automated, self-improving system loop that moves from production trace evidence directly to a verified pull request.
Chapters:
00:00 Introduction & The Production Agent Gap
02:31 Why Traces (Not Code) Are the Source of Truth
06:17 Live Demo: The Bottleneck of Manual Agent Triage
09:57 Inverting the Loop: Evidence, Context & Triggers
14:50 Live Demo: Automated Root Cause Analysis & Pull Requests
20:52 Agent Studio: Building Custom Workers, Presets & Skills
23:46 Case Study: How Signal Automatically Fixed Arize's AI Agent
33:24 Live Q&A: Costs, Agent PR Acceptance & Evals Strategy
Try Arize AX Free: arize.com/products/ax/?utm_source=youtube&utm_medium=social&utm_campaign=q22026-webinar-vibes-to-production-na&utm_content=youtube-description
Resources:
🔬 Phoenix (open source): phoenix.arize.com
đź“– OpenInference: github.com/Arize-ai/openinference
đź“– Phoenix docs: docs.arize.com/phoenix
How long does it currently take your engineering team to move from spotting a production agent failure to shipping a fix? Let us know in the comments below!
Subscribe to Arize AI and hit the notification bell for production agent engineering breakdowns!: youtube.com/@arizeai?sub_confirmation=1
#AIEngineering #AIAgents #AgentEvals
When an autonomous AI agent fails in production at 2 AM, finding the bad output is only the beginning. You can't read an agent's logic through raw code like traditional deterministic software; traces are the new source of truth. Finding the failure pattern, tracing it to the underlying root cause, writing a fix, and verifying that the change improved the entire agent harness traditionally takes days of manual human triage.
In this technical deep dive, we demonstrate how to invert the debugging loop using Arize AX. Learn how to shift your engineering team from first responders to reviewers by building an automated, self-improving system loop that moves from production trace evidence directly to a verified pull request.
Chapters:
00:00 Introduction & The Production Agent Gap
02:31 Why Traces (Not Code) Are the Source of Truth
06:17 Live Demo: The Bottleneck of Manual Agent Triage
09:57 Inverting the Loop: Evidence, Context & Triggers
14:50 Live Demo: Automated Root Cause Analysis & Pull Requests
20:52 Agent Studio: Building Custom Workers, Presets & Skills
23:46 Case Study: How Signal Automatically Fixed Arize's AI Agent
33:24 Live Q&A: Costs, Agent PR Acceptance & Evals Strategy
Try Arize AX Free: arize.com/products/ax/?utm_source=youtube&utm_medium=social&utm_campaign=q22026-webinar-vibes-to-production-na&utm_content=youtube-description
Resources:
🔬 Phoenix (open source): phoenix.arize.com
đź“– OpenInference: github.com/Arize-ai/openinference
đź“– Phoenix docs: docs.arize.com/phoenix
How long does it currently take your engineering team to move from spotting a production agent failure to shipping a fix? Let us know in the comments below!
Subscribe to Arize AI and hit the notification bell for production agent engineering breakdowns!: youtube.com/@arizeai?sub_confirmation=1
#AIEngineering #AIAgents #AgentEvals
