Priya Venkatasubramanian

Columnar layouts beat row stores on scans but lose on writes and point lookups.

Simplicity comes with costs in reprocessing, late data, and query serving.

Shadow traffic lets you validate pipeline changes against real production data with zero user risk.

Agents burn through database budgets in seconds, requiring rethinking rate limits entirely.

Agents need semantic and tool-result caching to match their bursty, low-repetition query patterns.

Agent queries demand stateful backends, but stateless designs hide costs until concurrency scales.

Databases built for human queries can't match the billing model that agentic workloads demand.

Malicious instructions hidden in data can redirect an agent's database queries.