THE WORLD IS NOT STANDING STILL.RSS
BIG CHANGE.
BIG CHANGE, one tap away
BIG CHANGE, one tap away

Make room for the big picture.

Add BIG CHANGE to your home screen or desktop. It opens in its own window, with the latest stories when you’re online.

iPhone & iPad
Open bigchange.ai in Safari. Tap Share, then Add to Home Screen. Turn on Open as Web App if shown, then tap Add.
Android
Open bigchange.ai in Chrome. In the browser menu, choose Install app or Add to Home screen.
Desktop
In Chrome or Edge, use the install icon in the address bar or the browser’s app menu. In Safari on Mac, choose File → Add to Dock.

Markdown edition

# AWS says AgentCore's newer runtime is becoming the default

> Existing runtimes stay on their current platform.

By BIGCHANGE.AI Editorial

Published: 2026-11-13T00:47:38.225Z
Updated: 2026-11-13T00:47:38.225Z
Canonical: https://bigchange.ai/blog/aws-agentcore-runtime-default

![Illustration of two developers reviewing a configuration.](https://bigchange.ai/api/media/file/runtime-v3.png)
AI-generated illustration for BIG CHANGE.

AWS says AgentCore Runtime V2 is becoming the default for new deployments. Its November 12 announcement says existing runtimes stay on their current platform until customers move them.

## The big change

- The stated default is changing, but AWS’s operational guide still names V1. Teams should explicitly choose the intended platform and verify the saved configuration instead of assuming the two descriptions agree.
- A runtime supplies the environment in which agent software executes. Starting that environment and completing the agent’s task are different things; a fast start does not establish a fast or correct answer.
- Storage permissions deserve a separate check from compute isolation. A session’s ability to mount files does not by itself decide which data its agent should be permitted to read or change.

## What changes now

The [November announcement](https://aws.amazon.com/about-aws/whats-new/2026/11/new-agentcore-runtime-default/) lists up to 8 virtual CPUs, 32 GB memory and 10 GB container images. An image packages application software and dependencies. AWS says each session retains a hardware-isolated micro virtual machine; that isolation is not a substitute for tool permissions.

Its [developer guide](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/runtime-how-it-works.html) still says omitting `platformVersion` on creation selects V1. The announcement acknowledges that V2 is not yet the default in every Region. Explicitly setting V1 or V2 avoids relying on the disputed omission behavior. The guide lists ten supported V2 Regions; check the intended location before migration.

## Why initialization matters

V2 was [available in September](https://aws.amazon.com/about-aws/whats-new/2026/09/new-agentcore-runtime-generally-available/). AWS’s [technical explanation](https://aws.amazon.com/blogs/machine-learning/the-new-agentcore-runtime-elastic-optimized-and-consistently-fast-starts/) says it initializes an environment once, snapshots it, and restores that prepared state for new instances instead of repeating initialization.

The guide describes a practical consequence: creation and updates take minutes to prepare the snapshot. The container must finish initialization and report healthy within 120 seconds of startup. Otherwise creation fails. This preparation time differs from starting a later session from the saved state, so deployment timing and invocation timing should be measured separately.

## What the speed evidence measures

September’s AWS test used 5,000 cold invocations per agent, an empty echo application calling no model or tools, and public-internet round trips between two AWS Regions. Across 200 MB–2 GB images, reported 75th-percentile starts were 1.9–2.0 seconds versus 5.4–30 seconds previously. This is startup evidence, not a measurement of full agent work.

The 75th percentile means three quarters of observations fall at or below the value. A production request may also wait for model inference, tools and application logic. Test those steps together before translating a startup figure into a user-facing response-time expectation.

The September article also describes reclaiming unused memory but a higher unit rate. Lower memory consumption alone does not guarantee a lower total bill. Compare representative workloads, not just resource peaks.

## Storage is a separate boundary

October [release notes](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/release-notes.html) already list V2 filesystem mounting through FUSE and session context keys. FUSE lets software work with mounted storage as local files. Context keys let policies restrict sessions sharing an execution role to their own data.

The [November blog](https://aws.amazon.com/blogs/machine-learning/run-agents-securely-and-efficiently-with-the-next-generation-agentcore-runtime/) promises folder-level session scoping through session-labeled credentials by re:Invent; this is future wording, not confirmation of completed rollout.

For a deployment review, ask which storage is mounted, which identity grants access and whether the policy is actually enforced. Keep migration tests focused on observed behavior, permissions and cost.

## Sources

- [AWS November announcement](https://aws.amazon.com/about-aws/whats-new/2026/11/new-agentcore-runtime-default/) — Default and capacity announcement.
- [AWS September release](https://aws.amazon.com/about-aws/whats-new/2026/09/new-agentcore-runtime-generally-available/) — Earlier availability.
- [AgentCore runtime guide](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/runtime-how-it-works.html) — Version selection and region list; describes a conflicting default.
- [AWS release notes](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/release-notes.html) — Earlier storage and session controls.
- [AWS runtime technical explanation](https://aws.amazon.com/blogs/machine-learning/the-new-agentcore-runtime-elastic-optimized-and-consistently-fast-starts/) — Earlier architecture, test and billing.
- [AWS November runtime blog](https://aws.amazon.com/blogs/machine-learning/run-agents-securely-and-efficiently-with-the-next-generation-agentcore-runtime/) — Planned session-folder scoping.