Dynatrace widens AI Observability ingest to coding agents, Python and OpenInference

OneAgent adds Python instrumentation without code changes

Dynatrace says a large share of production LLM applications are written in Python and call AWS Bedrock, OpenAI or Azure OpenAI, or are built with LangChain. OneAgent now instruments these applications directly, capturing logs, traces and LLM-specific telemetry through a setting rather than a code change.

Once enabled, teams can capture prompt and completion content to debug AI and LLM interactions, analyze model behavior, and assess the performance and cost impact of LLM calls. The data appears next to the rest of the application telemetry, so an LLM call is, in Dynatrace's words, "just another span in a trace".

OneAgent also ships experimental Python sensors for a wider set of providers and frameworks:

• Anthropic
• OpenAI agents
• Google ADK
• Claude Agent SDK
• Microsoft Foundry
• Mistral AI
• LangChain/LangGraph
• Pydantic AI
• LiteLLM
• Cohere
• CrewAI
• Groq
• Haystack
• Ollama

Each sensor is switched on through its own feature controller in the OneAgent settings, so teams can enable only the providers they use. Dynatrace cautions that attribute collection is best-effort, and that completeness and schema might change in future releases without prior notice.

OpenInference telemetry is ingested directly

OpenInference is an open semantic convention and instrumentation standard for AI applications, built on OpenTelemetry standards and originally developed by Arize. According to Dynatrace, it auto-instruments 33+ frameworks, including LangChain, LlamaIndex, CrewAI, AutoGen, OpenAI and Anthropic, and captures LLM calls, agent runs, tool calls and retrieval steps as distributed traces without code changes.

Dynatrace states that Arize donated the specification to the OpenTelemetry project in late May 2026, that the OpenTelemetry governance board accepted it, and that the merge landed in mid-June. Dynatrace now ingests OpenInference telemetry and normalizes its attributes to the gen_ai.* convention that its AI Observability is built on. Teams that already ship OpenInference spans can use that instrumentation in Dynatrace unmodified, and the data supports the same performance and cost monitoring, root cause analysis across agent and tool chains, and dt-evals quality scoring as native gen_ai.* telemetry.

Coding agent monitoring and Claude Platform support

Dynatrace points to coding agents such as Claude Code, Codex (OpenAI), Cursor, Google Antigravity and GitHub Copilot, which it says have moved from experiments to essential daily tools for many engineering teams. Leaders, it says, often cannot tell which agents are used, what they cost, how reliable they are and where they fail.

AI Observability can capture OpenTelemetry signals from these agents to track token consumption, cost, session activity and tool behavior in real time. Dynatrace also says it now supports Claude Managed Agents and Claude Platform tracing and is officially listed as an observability partner on the Claude Platform. With native OpenTelemetry export, traces can be sent from the Claude Console to Dynatrace without code changes, covering agent sessions, model calls, tool execution, token consumption and latency.

The company argues that because agent telemetry sits alongside production traces, services and infrastructure, teams can link agent activity to the commits, pull requests and downstream production behavior that followed.

OpenLLMetry support ends, and what it means for teams

Dynatrace is announcing end of life for OpenLLMetry support. It will continue through the end of January, and from February 1 AI Observability will support three ingest paths: OpenTelemetry-based instrumentation, OpenInference-instrumented applications, and OneAgent. The company recommends that teams using OpenLLMetry choose a path and plan their migration now, and says an EOL notice will soon be published in the Dynatrace documentation.

For teams running mixed AI stacks, the change means each tool does not need its own route into observability. Dynatrace says it provides a common foundation across supported sources without forcing standardization on a single vendor format.