Databases were not designed for this
summary
Databases face concrete changes when AI agents become primary users: enforce statement_timeouts per role and idle_in_transaction_timeout; record soft deletes with deleted_by and delete_reason and use append-only event logs with idempotency keys for high‑stakes writes; create separate connection pools and per‑role connection limits; tag SQL with comments (or proxy rules) for observability and circuit breakers for noisy agents; apply row‑level security, scoped permissions and approval queues so autonomous workloads can’t corrupt or overreach.
AI-written summary · about
chapters (6)
- 0:00The agentic challenge
- 2:31Managing agent writes
- 10:34Handling connection patterns
- 14:11Observability and safety
- 20:07Securing data access
- 25:07Defensive data architecture
related talks
47mThe Agent Era: How AI Agents Are Reshaping Data Platforms
28mSmall Data: Databases in the real world
06mOracle AI Database
24mDatabases beyond the tutorials
34mKeynote: Things Databases Don't Do (but should).
20mOpen Source Database Operations in the Age of AI
58mWhy Databases Cry At Night
21mAgents on GuardRails: Deterministic Safety for Database Operations