Summary
Flora Liu, an engineer on Notion’s GTM team, recounts how a small team rebuilt Notion’s fragmented go-to-market stack into a unified, agent-ready system over the course of several months. The talk is candid about the mess they started from: customer data spread across Salesforce, Gong, Outreach, ZoomInfo, and Snowflake, with critical context — notes like “the champion just left” or “don’t contact this account” — locked in unstructured Notion docs that automations couldn’t read.
The architecture they built follows four layers: a context layer (Snowflake as warehouse, DynamoDB as a millisecond-latency key-value store serving denormalized customer profiles), a decisioning layer that turns customer events into signals, an execution layer, and a measurement layer. Crucially, they brought all of this — structured product usage data, enrichment data from Clay, and unstructured meeting notes — into Notion itself, so agents, humans, and automated workflows all operate on the same source of truth. Liu describes this as “using Notion to grow Notion.”
A key design principle was owning the contact layer and renting everything else, keeping the lean team focused on what they knew best. The talk is particularly strong on the hard problems of data latency, conflicting records across vendors, and making automations safe when unstructured context can change what an action should be.
📺 Source: AI Engineer · Published August 26, 2026
🏷️ Format: Workflow Case Study
