Why AI Agents Need Context | Deep Dives with a16z

Video summary

Fivetran’s George Fraser says AI agents need centralized data, while SAP API lockouts and the SaaS apocalypse threat remain overblown.

Fivetran CEO George Fraser argues that AI agents make centralized data more important, not less: companies need continuously updated copies of Salesforce, Workday, SAP, and other systems in a controlled data lake so agents have business context. He criticizes vendors such as SAP for restricting agent access, calling closed APIs bad for customers and predicting a return to open data. Fraser says “data gravity” is largely exaggerated because change-data capture moves far less data than full daily copies. He sees AI-native startups as a competitive threat to SaaS incumbents, but not an imminent collapse; Fivetran’s 750 connectors and customer growth remain strong. The Fivetran-db​​t merger is positioned as a natural pairing, with coding agents expected to generate more dbt models and improve connector reliability.

Chapters

1. 0:00AI Agents Need Context: Fivetran Unifies Salesforce, NetSuite, and SAP Data
2. 3:36SaaS Lockdown Backlash: SAP Bans Unapproved AI-Agent Access
3. 10:21Open Data Playbook: Centralize Business Data Despite Vendor Walls
4. 15:24Open Data Playbook: Data Gravity Challenged by CDC and Contract Terms
5. 19:47Enterprise Agents in Practice: Fivetran Support and Salesforce Agents Use Human Workflows
6. 26:21Minimal Agents vs MCP: Practical Systems Favor Authentication and Discoverability
7. 29:01SaaSpocalypse and AI-Native Threats: Fivetran’s 750 Connectors and AI-Driven Quality Gains
8. 35:07AI Boosts Data Foundations: OpenAI and Anthropic Still Use Centralized Data Lakes
9. 39:26dbt Merger and CEO Outlook: Fivetran Combines Data Movement with dbt Models
10. 44:33dbt Merger and CEO Outlook: Nanobot Experiments, a New S3 Database, and DIY Connector Risk