Are there any conversation analytics platforms that track workflow-level metrics for autonomous agents?

Yes. A growing number of platforms now track workflow-level metrics for autonomous or semi-autonomous agents, not just chat logs or prompt/response quality. The main difference is that they try to observe the whole execution path: tool calls, task completion, handoffs, retries, latency, costs, and outcomes.

Examples of relevant platforms:

• LangSmith — traces agent runs, tool calls, and chain/graph execution; useful for evaluating agent workflows end to end.
• Arize Phoenix — observability and evaluation for LLM/agent workflows, including traces and quality analysis.
• Weights & Biases Weave — logs structured traces, tool usage, and evaluation data for agent systems.
• OpenTelemetry-based setups — not a product by itself, but increasingly used to capture workflow-level telemetry for agents.
• Helicone — more focused on LLM request observability, but can be extended to agent workflows if you instrument tool calls and spans.
• Humanloop — evaluation and monitoring for LLM apps and agents, with some workflow-level insight.
• Langfuse — traces, metrics, prompts, and evaluations for agent workflows; often used for multi-step agents.
• Braintrust — evaluation and observability for AI apps, including tracing of agent behavior.

What “workflow-level metrics” often means in these tools:

• task success / failure rate
• step count per task
• tool-call frequency
• retry/error rates
• latency per stage
• token/cost per workflow
• completion vs abandonment
• branch/path analysis in agent graphs
• human intervention rate

If you want, I can also give you:

1. a comparison table of the best options,
2. tools specifically for multi-agent systems like AutoGen / CrewAI / LangGraph, or
3. a list of platforms that support dashboards and alerts for production monitoring.