When Your Users Are Agents: Lessons from Running Serverless Postgres

Most systems are designed with human users in mind: predictable session lengths, expected traffic spikes, and query patterns shaped by application code. But what happens to your core infrastructure when some of your most active users are autonomous AI agents?

Agents don’t get bored or discouraged. They fan out, retry, and generate non-deterministic, dynamically constructed queries from partial context. The result is traffic that is bursty, expensive, difficult to predict, and sometimes just plain weird.

In this talk, we’ll explore the infrastructural shifts required when your users are no longer just humans, but AI agents. Drawing on hard-won lessons from running a Serverless Postgres platform at scale, we will look at how agentic workloads break traditional assumptions about connection pooling, query execution, resource management, rate limiting, and observability. You will leave with a practical framework for hardening your infrastructure to support autonomous AI workflows without compromising reliability, and some good war stories that will help you remember all this advice.

Key Takeaways:

• Recognizing Agentic Traffic Patterns: Understand how autonomous AI workflows change connection lifecycles, concurrency, retry behavior, and query patterns compared to traditional application traffic.
• Building Guardrails for Critical Infrastructure: Learn how to protect Postgres under unpredictable workloads using controls like timeouts, scoped permissions, rate limiting, and error handling. We’ll discuss when to use each technique and how each one can backfire when applied too bluntly.
• Monitoring and Debugging Agent Workloads: Learn how to use database-side signals to catch agent-driven problems before they become expensive incidents. For example, an agent retrying the same failed SQL thousands of times.