Live Nov 18 · Cohort 3Nov 18, 19:00 MSK — free hands-on AI agents live workshop · Live Cohort 3 starts Nov 23Join Nov 18 Live →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableAWS Machine Learning Blog→ original

Amazon Bedrock AgentCore Runtime: Isolated microVMs for Claude Code, Codex and Cursor

AWS has launched Amazon Bedrock AgentCore Runtime, a cloud infrastructure for coding agents. Each session gets an isolated microVM, so Claude Code, Codex, Kiro and Cursor can run in parallel without sharing secrets or file systems. The headline feature: the agent keeps working after you close your laptop. Delegate a task, go to dinner, and the results are already in the workspace in the morning.

AI-processed from AWS Machine Learning Blog; edited by Hamidun News
Amazon Bedrock AgentCore Runtime: Isolated microVMs for Claude Code, Codex and Cursor
Source: AWS Machine Learning Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceAWS Machine Learning Blog ↗
Editor-in-Chief
One microVM per sessionisolation at the virtual machine level, not just a container
Gateway for toolsone protected channel instead of raw credentials inside the agent
Persistent workspacecontext, files and cache live across session runs

AWS has introduced Amazon Bedrock AgentCore Runtime, a cloud execution environment for AI coding agents. Close your laptop and come back to the task in the morning: the agent keeps working in the cloud on its own, and the results wait for you in a persistent workspace.

What is AgentCore Runtime

Amazon Bedrock AgentCore Runtime is managed AWS infrastructure that gives each AI agent session its own isolated microVM. The platform handles the full session lifecycle, from launch to saving results. Files, cache, environment variables and the entire working context persist between sessions and are not reset when you close the browser or lose your connection.

Popular coding agents are supported: Claude Code, OpenAI Codex, Kiro and Cursor. They can run in parallel in a single AWS account, each in its own microVM, with no overlap in data, ports or secrets. AgentCore Runtime is built on top of the Bedrock infrastructure and integrates with the platform's other services, including tool access through Gateway. For teams that do not want to be tied to a single provider, support for agents from different ecosystems on one platform is especially convenient.

Isolation and security

The key principle of AgentCore Runtime is full isolation between sessions. No agent can see another's secrets, ports or file system. In team development, where several engineers run different agents under one account, this matters: even with parallel use, there is no risk of accidentally leaking tokens or of working directories conflicting.

Access to external tools (APIs, databases, git repositories, corporate services) goes through AgentCore Gateway: a single entry point with authorization management and full logging of every call. The agent works with minimal privileges and requests only what the current task needs.

  • One microVM per session: isolation at the virtual machine level, not just a container
  • Gateway for tools: one protected channel instead of raw credentials inside the agent
  • Persistent workspace: context, files and cache live across session runs
  • Built-in observability: tracing, logs and metrics out of the box, with no extra setup
  • Parallel execution: several agents at once without resource conflicts

The end of the open-laptop era

Before AgentCore, coding agents ran locally, directly in the terminal or browser. Hours-long codebase analysis, large refactoring and test generation all required keeping the machine on. A dropped connection or an accidentally closed tab meant lost progress.

This is especially painful for long tasks: a full security scan of a large repository, writing documentation for hundreds of functions, or migrating legacy code. Such tasks used to be either split into parts, losing context, or left running on a server by hand, with homemade scripts and no guarantees. AgentCore automates exactly this "invisible" part.

Now you delegate the task to the agent, close the laptop and go to dinner. The agent works in an isolated microVM on the AWS side, and in the morning the results are already in the workspace, waiting for review. For teams, this means horizontal scaling without extra complexity: several developers run different agents at the same time, while Gateway manages access centrally.

What it means

AgentCore Runtime is a serious step by AWS toward turning AI coding agents from desktop tools into full cloud infrastructure. For teams already using Claude Code, Codex or Cursor, it is the next logical step: delegating not only the tasks but also the time they take to run. The "close the laptop, done by morning" pattern is becoming the standard rather than the exception.

Story

Key Questions Answered

What is AgentCore Runtime?

Amazon Bedrock AgentCore Runtime is managed AWS infrastructure that gives each AI agent session its own isolated microVM. The platform handles the full session lifecycle, from launch to saving results. Files, cache, environment variables and the entire working context persist between sessions and are not reset when you close the browser or lose your connection. Popular coding agents are supported: Claude Code, OpenAI Codex, Kiro and Cursor.

What are the key details on Isolation and security?

The key principle of AgentCore Runtime is full isolation between sessions. No agent can see another's secrets, ports or file system. In team development, where several engineers run different agents under one account, this matters: even with parallel use, there is no risk of accidentally leaking tokens or of working directories conflicting.

What are the key details on The end of the open-laptop era?

Before AgentCore, coding agents ran locally, directly in the terminal or browser. Hours-long codebase analysis, large refactoring and test generation all required keeping the machine on. A dropped connection or an accidentally closed tab meant lost progress. This is especially painful for long tasks: a full security scan of a large repository, writing documentation for hundreds of functions, or migrating legacy code.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…