AWS AgentCore enables a live AI browser in a React app with three lines of code

Amazon Bedrock AgentCore has gained a new component, BrowserLiveView, for React applications that solves one of the most common problems with autonomous AI browser agents — transparency. The component displays a real-time video feed of the agent’s browser session inside a React application, with three lines of JSX.

Why this matters

When an AI agent autonomously acts in the browser — filling in forms, clicking links, navigating between pages — the user has reason to be suspicious: is it doing what I actually asked for? Might it accidentally do something wrong? BrowserLiveView solves that problem by letting the user watch every action of the agent live. This builds trust in delegated tasks and enables supervised workflows where the user can intervene in real time without leaving the application. As a bonus: it creates an audit trail with visual documentation — useful for compliance.

How it works technically

The architecture has three components. The React application contains BrowserLiveView and receives SigV4-presigned URLs. The application server manages sessions and generates presigned URLs. AWS Cloud hosts isolated browser sessions via Bedrock AgentCore. Streaming happens via the Amazon DCV protocol, and — this is important for performance — the DCV stream goes directly from AWS to the client’s browser, bypassing the application server. This minimizes latency and reduces the load on your own backend infrastructure.

Implementation in 3 steps

1. Start a session and generate a Live View URL through the AgentCore API
2. Render the stream in a React component with viewport dimensions matching the browser session
3. Connect an AI agent that drives browser actions via the Playwright Chrome DevTools Protocol

The sample application in the documentation demonstrates an agent navigating Wikipedia while the user watches, along with an activity log showing which decisions and tools are in progress. For teams already using Bedrock AgentCore, this is probably the fastest path to a production-ready “AI browser concierge” experience — without self-hosting remote browser infrastructure.