Anthropic's Claude coding agents get isolated microVMs on Amazon Bedrock AgentCore

AWS published a walkthrough showing how to host coding agents on Amazon Bedrock AgentCore Runtime, which assigns each agent session its own isolated microVM with a persistent workspace, secure tool access through AgentCore Gateway, and built-in observability. The pitch — 'it's safe to close your laptop now' — is that long-running coding agents can execute autonomously in the cloud rather than tying up a developer's local machine.

Crucially, the isolation lets teams run multiple agents — Claude Code, OpenAI's Codex, Amazon's Kiro, and Cursor — in parallel without sharing secrets, network ports, or filesystems, addressing a real security and reliability concern as agents gain shell and tool access. Sessions persist, so work can be paused and resumed.

This is the infrastructure counterpart to the week's agentic-coding theme: OpenAI's Codex-first 'superapp,' Google's NotebookLM secure cloud computer, and Anthropic's own long-horizon agent claims. AWS's angle is neutral hosting — supporting rival vendors' agents on a governed AWS substrate with IAM, VPC, and observability.

Anthropic separately featured in a companion AWS post on automating Oracle PL/SQL-to-PostgreSQL migration using Claude Sonnet 4.6 on Bedrock with the Strands Agents framework and the AWS Knowledge MCP Server. Anthropic's Boris Cherny (Claude Code lead) underscored the autonomous-agent push on X, sharing tips for running Opus 'for hours/days' with auto-mode permissions and dynamic workflows. Watch microVM cold-start latency and per-session cost, the practical gating factors for parallel-agent fleets.