Uploaded June 2026 | Updated September 2026, 3 weeks ago
Want to understand what's happening inside your AI agents without digging through every trace yourself? Meet PXI (Phoenix Intelligence), the AI engineering agent built into Phoenix. In this walkthrough, we hand PXI an investigation into a real agent and watch how it works.
The idea is simple: instead of digging through traces, prompts, evaluations, and experiments yourself, you hand the investigation to PXI. It's a coding agent pointed at your telemetry. In this demo we point it at Wonder Toys, a toy-store assistant traced in Phoenix, and ask whether its recommendations are any good.
Watch this video to learn:
• What PXI is: an AI engineering agent built into Phoenix (AI agent observability)
• How it walks failing traces toward root-cause hypotheses (LLM tracing and debugging)
• How it stays in your control: assistance is opt-in, it runs on your own model key, and it asks for explicit approval before any state-changing action
A note on expectations: PXI is in beta. It can make mistakes and can hallucinate, especially on long traces or unfamiliar frameworks, so treat its outputs as suggestions. It runs on your own model key (OpenAI, Anthropic or Gemini)
⏱️ Chapters
00:00 Intro
00:36 Debugging with PXI
01:39 The turn to quality
02:48 PXI is observable too
03:24 Setting up your PXI
04:17 Behind the Scenes
05:33 Wrap-up
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
Want to understand what's happening inside your AI agents without digging through every trace yourself? Meet PXI (Phoenix Intelligence), the AI engineering agent built into Phoenix. In this walkthrough, we hand PXI an investigation into a real agent and watch how it works.
The idea is simple: instead of digging through traces, prompts, evaluations, and experiments yourself, you hand the investigation to PXI. It's a coding agent pointed at your telemetry. In this demo we point it at Wonder Toys, a toy-store assistant traced in Phoenix, and ask whether its recommendations are any good.
Watch this video to learn:
• What PXI is: an AI engineering agent built into Phoenix (AI agent observability)
• How it walks failing traces toward root-cause hypotheses (LLM tracing and debugging)
• How it stays in your control: assistance is opt-in, it runs on your own model key, and it asks for explicit approval before any state-changing action
A note on expectations: PXI is in beta. It can make mistakes and can hallucinate, especially on long traces or unfamiliar frameworks, so treat its outputs as suggestions. It runs on your own model key (OpenAI, Anthropic or Gemini)
⏱️ Chapters
00:00 Intro
00:36 Debugging with PXI
01:39 The turn to quality
02:48 PXI is observable too
03:24 Setting up your PXI
04:17 Behind the Scenes
05:33 Wrap-up
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
