AI Instrumentation: Track specific agent type

Everyone can contribute. Help move this issue forward while earning points, leveling up and collecting rewards.

• Close this issue

Summary

We currently track workflow request categories (Chat, Code Review, Software Development, etc.) which tell us the workflow type, but we cannot distinguish which specific agent users invoked within those categories.

For example:

• We know a user ran a "Chat" workflow
• We don't know if they used Duo Agent, a Foundational Agent, a Catalog Agent, or an external agent

Impact

This blocks critical adoption visibility metrics:

• Can't answer: "Which agents are most popular?"
• Can't identify: "Are teams adopting the new Duo Planner agent?
• Can't prioritize: Which agents should be promoted or improved based on usage
• Blocks: Customer-facing adoption dashboards and ROI analytics

What We Need

Add agent-level instrumentation to DAP workflow events so we can:

1. Identify which specific agent (Duo Agent, Foundational Agent, Catalog Agent, External Agent) was invoked
2. Distinguish GitLab-native agents from 3rd party external agents
3. Join this data with other usage metrics (tokens, cost, user activity, etc.) in Snowflake for adoption analysis

New Data Fields Required

1. agent_type (enum)

   • duo_agent - GitLab-built Duo Agent
   • foundational_agent - Foundational model agents (Claude, GPT, etc.)
   • catalog_agent - Custom/community agents from the catalog
   • external_agent - 3rd party agents
   • NULL - For workflows that don't invoke agents

2. agent_subtype (string, optional)

   • For foundational_agent: duo_planner, security_analyst, etc.
   • For catalog_agent: agent_id or agent_name (which specific catalog agent)
   • For external_agent: agent provider or identifier
   • For duo_agent or NULL: leave empty

Data Availability Requirements

Internal Analytics:

• Snowflake data warehouse (joinable with existing DAP workflow data via workflow_id or equivalent unique key)
• Internal dashboards (Tableau, AI Reporting)

External Analytics:

• Billing events - Include agent_type and agent_subtype in billing event payload alongside token consumption, cost, and other request-based metrics
• Analytics dashboard - Customer-facing analytics dashboard (extend existing Duo usage reporting)
• Customer-facing API - Expose via API for customers to query programmatically