Default to Serverless for AI Agents, Then Let Production Data Decide What Moves
Datadog, Thursday, October 29th, 2026
Datadog recommends evaluating serverless first for AI agent workloads and using production data to decide which steps need warm or dedicated compute.
The post notes agents spend much of their lifetime waiting: in a study of over 35,000 production coding-agent sessions, the median session ran LLM or tool calls only 20% of its lifetime.
It cites Gartner and Google Cloud surveys showing agent adoption outpacing platform readiness, and argues serverless lets teams add agent workloads without managing a cluster.
Longer runs can use durable serverless workflows that save progress and release compute during waits.
It explains how to build an execution profile, compare cost with warm or dedicated capacity, and move only steps whose latency or runtime needs justify it.