Uploaded June 2026 | Updated September 2026, 2 weeks ago
As AI agents become more autonomous, traditional observability approaches are reaching their limits. The challenge is no longer just collecting traces and logs—it’s understanding why agents behave the way they do.
In this session, Nate Slater from AWS explores how agentic AI is fundamentally changing observability. Drawing on his experience building AI-powered root cause analysis systems, Nate explains why traditional monitoring tools struggle with probabilistic systems, multi-agent workflows, temporal drift, and the explosion of telemetry generated by AI agents. He argues that the future of observability requires agents that can reason about other agents, transforming observability from passive data collection into active analysis and diagnosis.
The talk covers the evolution from traditional application monitoring to AI-native observability, the challenges of debugging agentic systems at scale, and how technologies like OpenTelemetry, AWS Bedrock Agent Core, and Arize observability tools can work together to provide visibility into increasingly autonomous systems. Learn why recording telemetry is no longer enough—and why reasoning over observability data is becoming essential for enterprise AI adoption.
Timestamps
00:00 Why Observability Gets Harder with AI Agents
00:52 Traditional Tracing and Debugging Challenges
02:16 Using AI to Understand AI Systems
02:50 Introduction and Background
04:21 Correlation vs. Causation in Observability
06:18 New Challenges in Agentic Systems
06:46 Probabilistic Behavior and Error Propagation
07:30 Temporal Drift and Agent Memory
08:23 The Scale Problem: Thousands of Agents
08:56 From Recording Data to Reasoning About It
09:46 Why Traditional Observability Falls Short
10:23 Causal Reasoning and Intent-Aware Detection
12:19 Why Observability Matters More in the Agent Era
13:44 Enterprise Risk and Agent Liability
14:48 Building Agent Observability with AWS and Arize
16:34 From Observability to Reasoning
17:40 AI Agents Debugging Other AI Agents
19:37 The Future of Agent Observability
20:04 Key Takeaways
Key Takeaways
• AI agents introduce new observability challenges including probabilistic behavior, compounding errors, temporal drift, and massive increases in telemetry volume.
• Traditional observability tools focus on collecting data, but future systems must be able to reason about that data and identify causal relationships.
• Enterprises will need agent-based observability systems to understand, monitor, and govern increasingly autonomous AI workflows.
• Observability is becoming a prerequisite for safely deploying large-scale agentic systems in production environments.
• The future of AI operations involves agents observing, evaluating, and debugging other agents.
--
#AIAgents #Observability #AWS #LLMOps #AgenticAI #OpenTelemetry #AIEngineering #PlatformEngineering #AIInfrastructure #ArizeObserve
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
As AI agents become more autonomous, traditional observability approaches are reaching their limits. The challenge is no longer just collecting traces and logs—it’s understanding why agents behave the way they do.
In this session, Nate Slater from AWS explores how agentic AI is fundamentally changing observability. Drawing on his experience building AI-powered root cause analysis systems, Nate explains why traditional monitoring tools struggle with probabilistic systems, multi-agent workflows, temporal drift, and the explosion of telemetry generated by AI agents. He argues that the future of observability requires agents that can reason about other agents, transforming observability from passive data collection into active analysis and diagnosis.
The talk covers the evolution from traditional application monitoring to AI-native observability, the challenges of debugging agentic systems at scale, and how technologies like OpenTelemetry, AWS Bedrock Agent Core, and Arize observability tools can work together to provide visibility into increasingly autonomous systems. Learn why recording telemetry is no longer enough—and why reasoning over observability data is becoming essential for enterprise AI adoption.
Timestamps
00:00 Why Observability Gets Harder with AI Agents
00:52 Traditional Tracing and Debugging Challenges
02:16 Using AI to Understand AI Systems
02:50 Introduction and Background
04:21 Correlation vs. Causation in Observability
06:18 New Challenges in Agentic Systems
06:46 Probabilistic Behavior and Error Propagation
07:30 Temporal Drift and Agent Memory
08:23 The Scale Problem: Thousands of Agents
08:56 From Recording Data to Reasoning About It
09:46 Why Traditional Observability Falls Short
10:23 Causal Reasoning and Intent-Aware Detection
12:19 Why Observability Matters More in the Agent Era
13:44 Enterprise Risk and Agent Liability
14:48 Building Agent Observability with AWS and Arize
16:34 From Observability to Reasoning
17:40 AI Agents Debugging Other AI Agents
19:37 The Future of Agent Observability
20:04 Key Takeaways
Key Takeaways
• AI agents introduce new observability challenges including probabilistic behavior, compounding errors, temporal drift, and massive increases in telemetry volume.
• Traditional observability tools focus on collecting data, but future systems must be able to reason about that data and identify causal relationships.
• Enterprises will need agent-based observability systems to understand, monitor, and govern increasingly autonomous AI workflows.
• Observability is becoming a prerequisite for safely deploying large-scale agentic systems in production environments.
• The future of AI operations involves agents observing, evaluating, and debugging other agents.
--
#AIAgents #Observability #AWS #LLMOps #AgenticAI #OpenTelemetry #AIEngineering #PlatformEngineering #AIInfrastructure #ArizeObserve
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
