AWS Bedrock AgentCore preview keeps web apps live while coding agents edit them
Amazon Web Services’ machine learning blog announced a preview capability on Amazon Bedrock AgentCore runtime where coding agents can keep a working web application running inside an isolated agent session while product teams review changes in near real time. According to AWS, Session Endpoints assign each session an authenticated URL that forwards HTTP and WebSocket traffic to allowed ports, so edits made where the agent runs can surface in the browser within seconds instead of waiting on separate deploy cycles. The post walks through keeping a Vite site live during agent-driven edits. That design targets the friction of copying builds elsewhere before every review. The announcement is a product preview, not general availability, and the available excerpt does not establish pricing, latency guarantees, port limits, session lifetime, or independent benchmarks for the claimed speedup.
AWS Bedrock AgentCore preview keeps web apps live while coding agents edit them
AWS ML announced a preview where coding agents build working web apps and a live preview stays connected to where the agent edits code. AWS says each change can show up in the browser within seconds when the app keeps running on AgentCore runtime.
Key takeaway
Session Endpoints on AgentCore runtime could shrink agent web-app review cycles from deployment speed to seconds, but operational limits remain undocumented in the excerpt.
What happened
AWS ML announced a preview in which coding agents build working web apps and a live preview stays connected to the session where the agent edits code, with AWS saying each change can show up in the browser within seconds when the app keeps running on AgentCore runtime.
The AWS ML post describes Session Endpoints that give each agent session in an isolated microVM an authenticated URL forwarding HTTP methods, paths, query strings, bodies, and WebSocket connections to allowed ports, including a walkthrough that keeps a Vite website running while the agent saves files.
Evidence
AWS ML announced a preview where coding agents build working web apps with a live preview tied to where the agent edits code.
AWS ML · attributed
AWS ML announced a preview where coding agents build working web apps and a live preview stays connected to where the agent edits code.
AWS says browser updates can appear within seconds when the app keeps running on AgentCore runtime.
AWS ML · attributed
AWS says each change can show up in the browser within seconds when the app keeps running on AgentCore runtime.
Session Endpoints expose the running app via an authenticated URL forwarding HTTP and WebSocket traffic to allowed ports.
AWS ML · attributed
Session Endpoints expose that session via an authenticated URL forwarding HTTP and WebSocket traffic to allowed ports.
AgentCore runtime runs each agent session in its own isolated microVM.
AWS ML · attributed
AgentCore runtime , a capability of Amazon Bedrock AgentCore, runs each agent session in its own isolated microVM.
The post demonstrates keeping a Vite website running in the agent session for live review.
AWS ML · attributed
In this post, you deploy a coding agent that keeps a Vite website running in its session, then open one
Why it matters
Cloud teams building agent-assisted UIs may spend less time on deploy-only review loops if in-session previews hold up in production-like security and scaling constraints.
Limits and uncertainties
The AWS ML post is a product preview excerpt and does not establish pricing, general availability, latency guarantees, port restrictions, session lifetime, or multi-user access details.
The promotional post provides no independent benchmarks quantifying the claimed seconds-level review speedup.
Practical implications
Operators evaluating Bedrock AgentCore should pilot Session Endpoints against their port, auth, and isolation requirements before replacing existing preview or staging deploy workflows.
Builders wiring coding agents should confirm whether their stack (for example a Vite dev server) can stay running inside an AgentCore session under allowed ports.
What to watch
Whether AWS publishes GA terms, session limits, and security guidance beyond authenticated URLs and isolated microVMs for Session Endpoints.
Customer reports on real-world latency and reliability when agents save files during live browser sessions on AgentCore runtime.
