You've connected Cursor, Claude Code, or Copilot to your monorepo. In YouTube demos, it's magical. On your 500,000-file codebase spread across 200 repos, it's… less magical.
The agent suggests renaming a function. Except it can't see the 23 services calling it across other repos. It recommends a refactoring. Except the pipeline failing in prod isn't in its context. It “understands” your architecture. Except it's seen 200 files out of 100,000 and made up the rest.
The problem isn't the model. It's the context. At enterprise scale, agents need more than a bigger window and more than semantic retrieval alone. They need structured, traversable context: a knowledge graph.
In this talk, we'll look at the emerging category of knowledge-graph-backed tools for software teams: what they add beyond code search and semantic RAG, which questions they answer well, and how they connect into agents through interfaces like MCP.
Then we'll close with a short live demo using GitLab Duo Agent Platform's Orbit as one concrete example: the same question asked with and without graph-backed context.
On the agenda: