Automate Legacy Web Applications with Amazon Bedrock AgentCore Browser Tool

Enterprises across various industries face challenges in automating legacy web applications that require human-like interactions beyond traditional Robotic Process Automation (RPA). Amazon Bedrock AgentCore Browser Tool, integrated with Strands Agents, addresses this by providing a fully managed, cloud-based browser environment that allows AI agents to interact with legacy interfaces through secure, isolated sessions.

These legacy systems, often built on server-side middleware generating HTML, lack modern APIs, making automation complex due to multi-step workflows, dynamic content, and authentication mechanisms like MFA and proprietary SSO. Traditional RPA solutions struggle with these complexities, leading to brittle automation and manual interventions.

The architecture leverages a managed Chromium instance in the cloud, interacting with legacy applications via WebSocket-based Chrome DevTools Protocol (CDP) connections. It uses Amazon Bedrock foundation models for visual analysis and decision-making, orchestrated by Strands Agents, enabling the AI to navigate, interpret, and modify web interfaces dynamically.

A key feature is human-in-the-loop capability, where operators can oversee and intervene in automation processes, ensuring accuracy and compliance. The system maintains full audit trails by recording interactions and storing session data securely, satisfying regulatory requirements.

The implementation involves components like a React UI, TLS proxy, Python workers running Strands Agents, and the Bedrock Browser Tool, all deployed via Terraform. The solution supports session persistence, resilient UI interaction through semantic locators, and scalable, elastic operation.

This approach significantly enhances automation reliability, scalability, and compliance, making it suitable for complex, regulated enterprise workflows that depend on legacy web applications.