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
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
![When AI Can Write Code, What Are Software Engineers Worth? | Citadel
When AI can generate code in seconds, what still makes a software engineer valuable?
In this Arize:Observe session, Craig Owenby of Citadel explores how agentic coding tools are changing software engineering, and why the profession still requires far more than producing code.
Craig compares large language models to the printing press. The printing press replaced the manual work of copying books, but it did not replace authors. In the same way, AI coding agents can automate the mechanics of writing code without replacing the judgment, vision, empathy, and experience required to build useful software.
The session covers:
• Why “coder” and “software engineer” are increasingly different roles
• How tools like Claude Code, Codex, Copilot, and Cursor remove traditional barriers to building software
• What the printing press teaches us about AI-assisted development
• Why engineers should avoid competing with AI on raw code generation
• How Sears lost its advantage by competing with e-commerce on the wrong terms
• Why human experience, intuition, and empathy remain essential
• How constraints can improve product and engineering decisions
• Why shipping more features can increase volatility and reduce user trust
• How AI acts as leverage for strong and weak engineering decisions
• Why product direction and problem selection matter more as implementation gets easier
The central lesson: software engineering is not primarily about writing code. It is about deciding what should be built, understanding why it matters, and applying technology with judgment.
When everyone can code, engineers differentiate themselves through their standards, instincts, product sense, and ability to understand the people using what they build. :contentReference[oaicite:0]{index=0}
Chapters:
00:00 When everyone can code, what are engineers worth?
00:46 LLMs and the printing press
01:30 The limits that shaped software engineering
02:42 We finally live in a world where everyone can code
03:45 The existential question for experienced engineers
04:45 Coders versus software engineers
06:05 Don’t make the same mistake as Sears
07:38 Human judgment, experience, and empathy
09:02 Why constraints can produce better software
10:25 Product volatility and the Sharpe ratio
11:49 Software engineering is about solving problems
12:38 LLMs as leverage for engineers
13:45 What sets engineers apart
14:25 Your humanity is the key
🔗 Learn more about Arize: https://arize.com
🔗 Explore Arize:Observe: https://arize.com/observe
🔔 Subscribe for more talks on AI engineering, coding agents, evaluation, observability, and the future of software development:
https://www.youtube.com/@arizeai?sub_confirmation=1
#SoftwareEngineering #CodingAgents #AIEngineering When AI Can Write Code, What Are Software Engineers Worth? | Citadel](https://i.ytimg.com/vi/TllPOmVWF8s/mqdefault.jpg)









