Atelier is a managed runtime platform for AI agents: users bring repositories, prompts, workflows, and integrations, while the platform handles isolation, orchestration, observability, deployment, and operational controls. This session walks through the AWS architecture behind running agent workloads safely across many tenants, including sandboxed execution, event-driven workflow coordination, secure secret handling, data boundaries, tracing, and cost-aware scaling.

We will cover the design tradeoffs involved in moving from single-tenant experiments to a production-grade agent platform: how to isolate customer workspaces, schedule long-running agent jobs, keep side effects auditable, observe failures across distributed services, and ship safely with infrastructure as code.

Attendees will leave with practical patterns for building and operating cloud-native AI platforms on AWS, plus lessons learned from running Atelier in production.