According to AWS, each agent session still runs in its own hardware-isolated Firecracker microVM with its own identity, policy and audit trail. Sessions now scale up to 8 vCPUs and 32 GB of memory, support container images up to 10 GB, and start in under 2 seconds regardless of image size and concurrency, so teams no longer need warm pools to manage cold starts.
AWS also changed billing. The runtime reclaims memory within minutes of an agent releasing it, and an idle agent scales to zero. Always-on agents can use a committed baseline priced as a minimum, with the ability to burst up to four times that baseline on demand.
Agents can now mount and save work to a filesystem such as an Amazon S3 bucket with FUSE support, so data used during a session stays under the same controls. Customers quoted in the post, including Coinbase, Zuora and Okta, report faster cold starts and lower memory use in their own tests.
Why it matters if you build agents
Developers running agents with bursty or idle periods may be able to drop warm pools and pre-provisioned capacity, since startup is now stated to be under 2 seconds.
Agents that pause between turns will be billed on memory in use rather than wall-clock time, which changes cost estimates for multi-turn workloads.
Agents that read files from S3 can keep that data inside the session's trust boundary, though per-session storage scoping is only promised for a later release.
Get stories like this every weekday
One short email with the day's agent tooling news.