Jaeger Previews AI-Assisted Troubleshooting and GenAI Observability for OpenTelemetry Traces at KubeCon | CNCF

Enroll your company as a CNCF End User and save more than $10K in training and conference costs
Posted on November 10, 2026 by Jaeger Maintainers
CNCF projects highlighted in this post
New experimental capabilities bring natural-language troubleshooting, native MCP tools and GenAI-aware trace visualization to Jaeger
As AI applications move into production, developers and platform teams face new challenges in understanding how complex distributed systems behave. Traditional applications already generate traces containing hundreds or thousands of spans, making it difficult to identify performance bottlenecks and troubleshoot failures. AI applications introduce another layer of complexity, with model calls, agent decisions, retrieval operations and tool execution creating new observability demands.
At KubeCon + CloudNativeCon North America 2026, Jaeger, a CNCF-graduated distributed tracing project, is previewing its first integrated AI-assisted troubleshooting and generative AI (GenAI) observability experience.
The experimental capabilities, available in Jaeger v2.22, combine an in-product AI assistant, native Model Context Protocol (MCP) tools, reusable diagnostic skills and specialized GenAI trace visualization. Together, these capabilities help engineers investigate complex systems using the OpenTelemetry data they already collect, while maintaining visibility into the evidence behind AI-generated insights.
Investigating distributed traces has traditionally required engineers to manually navigate spans, identify dependencies and piece together the events leading to a failure. As applications grow more complex, this process becomes more time-consuming.
Jaeger’s new AI capabilities introduce a guided approach to troubleshooting, helping engineers identify relevant trace information without manually inspecting every span.
The preview introduces several capabilities:
A central design principle is progressive disclosure, which allows AI agents to retrieve only the trace evidence needed for each investigative step rather than sending an entire trace to a language model.
This approach helps keep investigations focused, limits unnecessary data retrieval and allows engineers to inspect the supporting evidence behind an assistant’s findings.
“Traces contain the evidence engineers need, but large traces can bury that evidence across hundreds or thousands of spans,” said Yuri Shkuro, creator and maintainer of Jaeger. “Jaeger’s AI capabilities give AI agents focused tools to search, map topology, find the critical path and inspect only the spans that matter. Engineers can see the tool calls and evidence behind the answer while using the OpenTelemetry data they already collect.”
As organizations deploy AI-powered applications, observability must extend beyond traditional services and infrastructure to account for the behavior of models and autonomous agents.
Unlike conventional application workflows, AI systems may involve multiple model interactions, retrieval steps and external tool calls before completing a task. Understanding these interactions is essential for diagnosing failures, evaluating performance and identifying unexpected behavior.
Jaeger’s new GenAI trace visualization builds on OpenTelemetry’s GenAI semantic conventions, enabling teams to examine AI activity alongside traditional distributed application traces.
With these capabilities, engineers can investigate how models, agents and tools interact, inspect token consumption and trace the sequence of operations behind an AI application’s behavior.
For organizations already using Jaeger and OpenTelemetry, this provides a path to extending existing observability practices to AI workloads without introducing a separate tracing system.
“AI applications add model calls, token use, retrieval and tool execution to an already complex production stack,” said Jonah Kowall, Jaeger maintainer. “Jaeger makes those interactions visible alongside the rest of the distributed system, helping teams extend their existing observability practice instead of creating another isolated data silo.”
Jaeger’s approach to AI-assisted troubleshooting reflects the cloud native community’s commitment to open standards, interoperability and vendor-neutral infrastructure.
The new architecture brings together MCP, Agent Client Protocol (ACP), AG-UI and OpenTelemetry to provide a modular foundation for connecting AI assistants with distributed tracing data.
Rather than tying troubleshooting capabilities to a single model or proprietary agent framework, Jaeger’s design supports replaceable components and extensible diagnostic workflows.
Jaeger works with any ACP-compatible agent, and agents that propagate trace content appear in Jaeger’s own traces, allowing engineers to observe the assistant’s behavior within Jaeger itself.
This architecture supports three important goals:
By combining these principles, Jaeger provides an experimental foundation for AI-assisted operations that keeps engineers in control of the investigation process.
The need for more effective AI-assisted troubleshooting comes as AI adoption accelerates across software development and cloud native infrastructure. AI workloads are part of cloud native production environments. The CNCF Annual Survey 2025 found that 66% of organizations already hosting GenAI workloads use Kubernetes for some or all inference workloads. We can expect this to continue to grow.
For SREs, platform engineers, developers and AI application teams, these trends reinforce the need for observability tools that can support both conventional distributed systems and emerging AI workloads.
Jaeger’s new capabilities are designed to address this challenge by making complex trace investigations more accessible, providing visibility into AI application behavior and keeping troubleshooting grounded in inspectable telemetry.
Jaeger’s AI-assisted troubleshooting and GenAI observability capabilities are available as an experimental preview, givin