What happened?

In a blog post published on August 13, 2026, Amazon Web Services (AWS) announced a new service called Amazon Bedrock AgentCore Browser Tool. The tool aims to automate legacy web applications—still used in healthcare, manufacturing, retail, and financial services—that lack modern application programming interfaces (APIs), using AI agents.

Bedrock AgentCore Browser Tool runs a cloud-managed Chromium browser instance through Playwright integration and WebSocket-based Chrome DevTools Protocol (CDP) connections. This means the target application only needs to be accessible via HTTP or HTTPS, regardless of its underlying technology stack. The tool works alongside Strands Agents for model-driven workflow orchestration and integrates with Amazon Bedrock foundation models.

Why does it matter?

In an example scenario shared by AWS, a large insurance company reportedly processes tens of thousands of operational changes manually each year in its policy management systems. Such legacy systems often rely on HTML interfaces generated by server-side middleware and involve multi-step workflows, dynamic form validations, and session-dependent state management.

Traditional robotic process automation (RPA) tools tend to be fragile against layout changes in these interfaces and struggle to handle exceptions. According to AWS, the solution offers session-isolated security, IAM access controls, and full audit trails, aiming to meet compliance requirements such as GDPR, HIPAA, and financial regulations.

What's known

  • The solution has a four-component architecture consisting of a React-based operator interface, a TLS termination proxy, a Python worker running Strands Agents, and the Bedrock AgentCore Browser Tool.
  • Amazon Cognito manages operator login and provides JSON Web Token (JWT) authentication across the system.
  • Financial regulations require records to be retained for six years, with the last 90 days instantly accessible; AWS states this requirement was factored into the architecture.
  • The full source code of the reference implementation, along with Terraform deployment files, was made available on GitHub.

What's next?

AWS states in the blog post that the architecture is offered as a reference implementation that organizations can adapt to their own legacy systems. The company says the solution aims to boost automation return on investment while preserving human oversight, emphasizing that low-confidence transactions should still be routed to human operators.