AWS updates AgentCore runtime with faster starts and usage-based billing
AWS released a next-generation version of AgentCore runtime, the managed compute layer inside Amazon Bedrock AgentCore that runs agent sessions. Each session still runs in its own Firecracker MicroVM with a separate identity, access policy and audit trail. AWS says sessions now launch up to 15 times faster, start in under two seconds regardless of image size or concurrency, scale to 8 vCPUs and 32 GB with container images up to 10 GB, and consume up to 68% less memory per session. Idle sessions now scale to zero, and a committed baseline option uses AWS Lambda MicroVM rates. Sessions can also mount and save files to storage such as an Amazon S3 bucket via FUSE.
Why it matters
Teams running long, multi-step agents no longer need pre-provisioned warm pools to avoid slow starts, and they pay for the memory agents actually use rather than idle waiting time. Mounting files inside the session also keeps an agent's data under the same controls as its compute.
Sources
- PrimaryAWS Machine Learning
Written by AIToolMaps from the source above and checked against it sentence by sentence.