Last updated: July 5, 2026
About AI Coding Insights
AI Coding Insights helps engineering organizations understand the adoption, impact, and cost of AI coding assistants. It captures Claude Code and Cursor activity as OpenTelemetry spans and connects agent sessions to pull requests, enabling you to measure cycle time from first prompt to merged PR.
Why Measure AI Coding Activity
Most organizations track AI coding spend but lack visibility into whether that investment translates to faster delivery. Cost metrics alone do not tell the full story. A model that requires more interactions but costs less per token may be more cost-effective than a premium model.
To understand the real impact, you need to trace the complete workflow from the initial prompt to the merged pull request and measure how cycle time changes over time.
What You Can Measure
AI Coding Insights provides visibility across four key areas:
- Track Cost: Spend by model, team, and user in real time.
- Track Adoption: Active developers, new users, and usage patterns from occasional to power users.
- Measure Productivity: Pull requests with AI assistance and cycle time from session to merged PR (requires GitHub integration).
- Analyze Tools & Skills: See which tools, MCP servers, and skills developers use.
- Explore Sessions: Individual sessions with prompts, tool calls, MCP server usage, and full conversation history.
The Overview tab provides an at-a-glance summary of all these metrics in one place, pulling headline numbers from adoption, productivity, cost, recent sessions, and technical usage.
How Collection Works
AI Coding Insights uses the open-source Dash0 agent plugin to capture telemetry:
- Claude Code: The Claude Code plugin hooks into Claude Code's extension API to emit OpenTelemetry spans
- Cursor: The Cursor plugin includes a lightweight binary that captures Cursor activity
- Spans: Each interaction generates a chat span with token usage, along with child spans for tool executions. All spans follow OpenTelemetry GenAI semantic conventions.
- Transport: Spans are sent to Dash0 via OTLP over HTTPS
Because the plugin uses OpenTelemetry standards, both Claude Code and Cursor sessions appear as the same signal type in Dash0. Prompts and tool outputs can be filtered at the source for privacy. Token counts, model names, durations, and tool names are always captured.
What Data Is Required?
Different combinations unlock different insights:
| Data Source | Cost | Adoption | Productivity | Tools & Skills | Sessions |
|---|---|---|---|---|---|
| Claude Code or Cursor telemetry | ✓ | ✓ | ✗ | ✗ | ✗ Basic metrics only |
| Telemetry + Dash0 agent plugin | ✓ | ✓ | ✗ | ✓ | ✓ Full conversation replay |
| Telemetry + plugin + GitHub integration | ✓ | ✓ | ✓ Cycle time from session to merged PR | ✓ | ✓ |
The plugin is recommended because it captures the complete set of attributes needed for all metrics views. The GitHub integration connects agent sessions to pull requests, enabling cycle time measurement.
Getting Started
- Follow Set Up AI Coding Insights to connect Claude Code or Cursor to Dash0.
- Read Key Concepts to understand how sessions, prompts, and teams map onto the data.
- Explore the tabs to investigate your team's usage: Track Cost, Track Adoption, Measure Productivity, Analyze Tools & Skills, and Explore Sessions.
Further Reading
- Track Cost: See spend by model, team, and user.
- Track Adoption: Understand who uses AI coding tools and how adoption is spreading.
- Measure Productivity: Connect spend to velocity with cycle time metrics.
- Analyze Tools & Skills: See which tools, MCP servers, and skills developers use.
- Explore Sessions: Drill into individual sessions to understand detailed usage.