What changed
AWS has announced updates to the AgentCore runtime, a managed compute layer for deploying and running AI agents securely at scale. The new version of the runtime allows for faster session startup times, increased compute scalability, and improved resource efficiency while maintaining strong isolation. Sessions now start up to 15x faster, scale to 4x more compute, and run up to 68% more efficiently, with the isolation boundary fully intact. The runtime also supports sessions with up to 8 vCPUs and 32 GB of memory, with container images up to 10 GB.
What this means in practice
The enhanced AgentCore runtime enables developers to run more complex workloads without waiting for long startup times. It also allows for more flexible billing, as idle agents scale to zero, reducing memory consumption and costs. For multi-turn agents, this means up to 68% less memory consumed per session. Always-on agents benefit from committed baseline pricing, which sets a memory baseline per session and allows for bursts up to four times that baseline on demand. This provides predictable rates at AWS Lambda MicroVM rates, making it easier for organizations to justify production rollouts.
Limitations and availability
The updates are now available for use, with the core guarantee that every session runs in its own hardware-isolated Firecracker micro virtual machine (MicroVM), with its own identity, policy, and audit trail. The changes are designed to support the growing demands of AI agents in production environments, addressing the challenges of balancing performance, cost, and efficiency at scale. The supplied announcement does not specify exact release dates for these updates beyond their current availability.
