Why a Multi-Day AI Workflow Needs Different Infrastructure Than a Chatbot
A lone agent handling customer queries can run in a serverless environment with short-lived sessions. Three agents collaborating on a creative workflow that spans several days — sharing context and building on each other's output — can't: serverless sessions that cap at a few hours don't cut it. That mismatch is the case AWS makes for two compute options on Amazon Bedrock AgentCore. MicroVMs are the serverless option, with fast cold starts, session isolation and consumption-based pricing. Runtime Instances are newer: AWS-managed EC2 infrastructure for persistent, long-running agent workflows. Both use the same runtime APIs, but instances add multi-day sessions, GPUs, persistent volumes, and the ability to colocate several agents on one machine. The trade is legible — keep the shared runtime interface, buy duration and hardware. Whether a given pipeline needs it depends on whether the work is genuinely multi-day or merely multi-step.
AI-generated discussion · Different perspectives
The load-bearing word here is 'persistent.' Most agent demos are multi-step, not multi-day. The infrastructure question only gets interesting when state has to survive the agents going to sleep and coming back to each other's files.
Persistence may also inherit a new class of failure: whatever a bad run wrote to the shared volume yesterday is what the other two agents open today. Long-lived sessions buy continuity and consequences.