# CoreWeave Agent Lens: Intelligent agent observability at enterprise scale

## Executive summary

CoreWeave Agent Lens is an observability platform designed for enterprise-scale AI agents, addressing the challenge of fragmented workflows and the inability to manually monitor millions of production interactions. It provides a single, continuous view of agent activity, turning production traces into actionable insights to continuously improve agent performance. The platform allows different roles—from product managers to AI engineers—to view the same data, facilitating rapid diagnosis, failure clustering, and fix validation.

## Key takeaways

- Centralized Observability: Agent Lens provides a single view of every trace, regardless of which prompt, harness, or model was used, eliminating the need to piece together separate tools for observability, evaluation, debugging, and testing. (0:39)
- Automated Failure Diagnosis: The system automatically surfaces failures, clusters similar issues, and ranks them by frequency and impact, ensuring teams know exactly where to focus their efforts. (0:54)
- Role-Specific Views: The dashboard offers tailored views: the Conversation tab (for PMs/Domain Experts), the Spans view (for AI Engineers to see agent invocations/tool calls), and the Timeline view (for operational sequencing). (3:43)
- LLM-Powered Tagging and Alignment: Tags can be applied manually or automatically using LLM judges. The platform ensures that LLM judges maintain consistency with domain expert criteria through dedicated alignment testing. (4:15, 5:32)

## Technical details

- Agent Lens Dashboard Metrics: The dashboard provides a time-filtered overview including conversation/turn counts, error counts, tag counts, cost, latency, model usage, and tool usage. (3:25)
- Failure and Intent Clustering (Insights): Agent Lens automatically clusters turns to provide in-depth insights into agent failures and user intent, helping teams understand what customers are actually asking the support agent. (6:13)
- Tagging Mechanism: Tags track agent behavior and can be generated using Coreweave Forge's in-app AI research agent, Arya, which analyzes traces to identify helpful new tags. (5:15)

## Practical implications

- Reduces the time spent on QA and refinement for AI applications by providing continuous, automated observability.
- Enables teams to move faster into production by mitigating the risk of regression caused by fragmented development tools.
- Provides a structured workflow for diagnosing issues: finding a problem, fixing it, and confirming the fix, all within one interface.

## Topics

AI Observability, LLM Agents, Build Engineering, Continuous Integration (CI), Data Tracing, Agentic Workflow, Coreweave Agent Lens, Coreweave Forge

Source: https://www.youtube.com/watch?v=z8lBbtEqbL0