Your agent can reach your data and still get it wrong. Vector search hands it a slice, Text2SQL hands it another, and neither tells it what is actually relevant or how the pieces connect, so the answer comes back confident and wrong. Zach Blumenfeld's argument in this hands on workshop is that the missing piece is not a better model or a better query. It is context, and context comes in shapes. He builds three reusable graph shapes on top of lakehouse data with Neo4j, each answering a question the agent cannot ask a table. Trees give a table of contents, so the agent can navigate what is even there. Communities surface themes, the patterns nobody named. Paths and cycles trace connections, how entities, documents, and records actually relate. The shapes are portable to BigQuery, Databricks, or Snowflake, and you leave with the code to run them on your own data and agents. Speaker info: - https://www.linkedin.com/in/zachblumenfeld/ - https://graphacademy.neo4j.com/courses/workshop-lakehouse Timestamps: 0:00 - Introduction: context comes in shapes, not queries 6:25 - The three graph shapes to build 10:56 - Environment setup: Codespaces and Neo4j 21:47 - Schema, shared terms, and join paths 45:25 - Shape 1: a table of contents for your data (trees) 50:30 - Building the containment tree and links 59:48 - Serving the graph to the agent over MCP 1:12:21 - Q&A: naming and the containment shape 1:22:54 - Shape 2: surfacing themes with communities 1:24:24 - Community detection with Leiden 1:31:09 - Hierarchical communities and naming themes 1:43:22 - Shape 3: connections and cross links 1:50:20 - Watching the agent use outlines and themes 1:55:05 - Wrap: theme types and real time linking