AWS announces next-generation Amazon Bedrock AgentCore runtime with faster starts, larger sessions, idle-scaling billing and data-level isolation

AWS's own blog says it has updated the runtime in Amazon Bedrock AgentCore. Each agent session still runs in its own Firecracker MicroVM, but sessions can now use up to 8 vCPUs and 32 GB of memory (4x before) and container images up to 10 GB. Sessions are said to start in under 2 seconds regardless of image size or concurrency. Idle agents scale to zero and memory is reclaimed within minutes. AWS also adds committed baseline pricing for always-on agents, with a minimum memory baseline and bursting up to 4x, at Lambda MicroVM rates. The text is cut off before the promised data-level isolation details, so that part is unclear.

Teams already building agents on AgentCore get bigger sessions, faster cold starts and a more predictable pricing option, which could change cost and design choices for long-running or multi-turn agents. Teams weighing managed runtimes against self-hosting, or against rival clouds, now have a few concrete specs to benchmark. This is an upgrade to an existing service, not a new capability, so the effect is incremental. The real cost impact depends on rates and baselines that the extracted text does not state.

The "up to 68% less memory per session" and "up to 15x" figures are AWS's own best-case numbers, with no stated workload or independent test. The framing of security versus speed versus cost as a solved trade-off, the METR, McKinsey and Deloitte adoption statistics and the customer quotes are marketing context, not evidence the update delivers. The headline's data-level isolation claim cannot be assessed because the details are missing.

1. 01Independent benchmarks of cold-start time (under 2 seconds claim) at large image sizes and high concurrency, and real memory savings for multi-turn agents against the 68% claim.
2. 02Published committed-baseline rates and the minimum memory baseline per session, to calculate whether always-on agents are actually cheaper than the previous on-demand pricing.
3. 03The full data-level isolation documentation, plus regional availability and any customer reports of production deployments at thousands of concurrent sessions.