How to Build Reliable Long-Running AI Agents | Arize x Temporal @arizeai
How to Build Reliable Long-Running AI Agents | Arize x Temporal  @arizeai
Uploaded September 2026 | Updated November 2026, 3 days ago
What happens when an AI agent runs for hours, calls multiple tools and models, waits on users, and then something fails halfway through?

In this joint webinar from Temporal and Arize, Nancy from Arize and Cecil from Temporal show how to build long-running AI agents that can survive failures while still giving engineering teams the visibility they need to understand, evaluate, and improve agent behavior.

Temporal provides durable execution for agent workflows, preserving state across retries, crashes, timeouts, and long-running processes. Arize AX adds the observability and evaluation layer: tracing model and tool calls, measuring quality, detecting recurring failures, and turning production behavior into data for improving the agent.

The session includes a live demo of an AI shopping assistant built with Temporal and instrumented for AI observability in Arize.

You’ll learn how to:

• Build durable, resumable AI agent workflows with Temporal
• Understand Temporal workflows, activities, workers, and durable execution
• Keep long-running agents alive across crashes, retries, and infrastructure failures
• Instrument agent workflows with OpenTelemetry and OpenInference
• Trace complete agent runs in Arize AX
• Inspect LLM calls, tool calls, sessions, latency, and cost
• Build LLM-as-a-judge evaluators for production agent behavior
• Detect hallucinated or ungrounded product recommendations
• Use Signal to find recurring failure patterns across production traces
• Turn failures into regression datasets and new evaluators
• Let agents propose fixes and open PRs for human review
• Build a continuous loop from production observability to agent improvement
• Handle human-in-the-loop workflows with Temporal
• Trace and evaluate coding agents as well as application agents

The core production challenge has two parts.

First, the workflow has to finish. Temporal preserves workflow state so an interrupted agent can resume instead of restarting expensive work from the beginning.

Second, finishing successfully does not prove that the agent behaved correctly. Arize provides the traces and evaluations needed to understand which path the agent took, which tools it called, what those calls cost, and whether the final behavior met your quality requirements.

Together, durable execution and AI observability create a foundation for agents that can run reliably in production and continuously improve from real-world behavior.

Chapters:
00:00 Why production AI agents need durability + observability
00:54 Meet Arize and Temporal
01:37 From agent prototype to production
03:45 What is Temporal?
07:20 Durable execution explained
09:26 Temporal workflows, activities, and workers
12:18 How Temporal preserves agent state
14:45 Demo architecture: AI shopping assistant
17:05 Live demo: Running the agent with Temporal
22:09 Inspecting agent workflows in Temporal
24:00 Why workflow success isn't enough
27:02 Why traces become the source of truth for agents
28:45 What is Arize AX?
30:41 OpenTelemetry + OpenInference for agent tracing
32:34 Demo: Inspecting agent traces in Arize AX
35:25 Tracking AI agent cost
36:17 Adding evaluations to production traces
37:47 Building a grounding evaluator
41:28 Using Signal to find recurring agent failures
44:28 Turning production failures into regression datasets
45:05 Letting agents propose fixes and open PRs
47:09 The production agent improvement loop
49:01 Q&A: Human-in-the-loop workflows with Temporal
51:54 Q&A: Dynamic agent decisions and tool results
53:17 Q&A: Tracing coding agents
54:08 Closing

🔗 Learn more about Arize AX: arize.com/ax
🔗 Learn more about Temporal: temporal.io
🔗 Explore OpenInference: github.com/Arize-ai/openinference

🔔 Subscribe for more technical sessions on AI agents, evals, observability, and production AI:
youtube.com/@arizeai?sub_confirmation=1

#AIAgents #Temporal #AIObservability
How to Build Reliable Long-Running AI Agents | Arize x TemporalWhen AI Can Write Code, What Are Software Engineers Worth? | CitadelAdd Tracing to Any AI Agent in One Command | Arize AXOne AI Question - how is marketing at Arize using AI, with Chris CooningWhy Your LLM Starts Forgetting InstructionsAI Agent Mastery Certification Course: Module 4 – Tools & MCPBuilding Responsible AI in Highly Regulated Industries | BlackRock | Arize Observe 2026How to Catch AI Agent Failures Fast with Monitors and Alerts | Ep. 12Nous Research’s Hermes Agent: The Case for Open Models in Production | Arize Observe 2026AI Agent Mastery Certification Course: Module 6 – Agent EvaluationFrom 5 Minutes to 2 Days: The Scale of AI Agent AutonomyAI Agents Certification Course: Module 1 – Introduction to Agents
Arize AI |

How to Build Reliable Long-Running AI Agents | Arize x Temporal

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER