Epsilon’s Agentic Platform team introduced a new managed UI testing capability that uses Amazon Nova Act together with the Amazon Bedrock AgentCore Browser tool. The service sits alongside their existing Playwright regression suite and exposes intent‑driven, natural‑language test steps through the same platform API that product teams already use for other AI agent capabilities.
Why Intent‑Driven Testing Matters
Traditional Playwright tests are tightly bound to selectors and page structure, so cosmetic UI changes often cause false failures. By allowing test steps to be expressed in plain English, the Nova Act integration tolerates visual tweaks while still verifying functional outcomes. This reduces the churn of test maintenance and opens the testing surface to non‑engineers who can author or review scenarios without learning selector syntax.
Architecture Overview
The solution consists of three logical pieces:
- Platform API – a REST‑style endpoint that product teams call to create a test workflow definition.
- Background task runner – a worker process that pulls workflow requests from a queue and orchestrates execution.
- Nova Act & Browser integration – code that invokes Nova Act to interpret natural‑language steps and drives the Bedrock AgentCore Browser tool to run them in isolated, ephemeral browser sessions.
Each test runs in a separate AWS account or role via standard STS AssumeRole calls, preserving tenant isolation while sharing the same managed service.
Implementation Details
Key implementation points drawn from the source:
- Test steps are submitted as plain‑English strings; Nova Act parses them into actionable browser commands.
- The Bedrock AgentCore Browser tool provides a fully managed browser environment—no binaries, no display server, and automatic session teardown.
- Sessions can be recorded to Amazon S3 and audited through AWS CloudTrail without additional configuration.
- Live view and human takeover are available for flows that require interaction with CAPTCHAs or multi‑factor authentication.
- Authentication relies on the standard AWS credential chain (task role), eliminating the need for separate API keys.
The platform exposes an 18‑step workflow that demonstrates end‑to‑end test creation, execution, result collection, and cleanup, illustrating how a product team can trigger a test with a single API call.
Operational and Security Considerations
From an ops perspective, the managed browser eliminates the operational burden of maintaining browser versions, drivers, and headless display infrastructure. Because each session is short‑lived and isolated, resource leakage is minimal.
Security implications include:
- Use of IAM‑based authentication aligns with existing credential management practices and avoids credential sprawl.
- Cross‑account execution via STS assumes a role with scoped permissions, preserving tenant isolation.
- All browser activity is logged in CloudTrail, providing an immutable audit trail for compliance or incident investigation.
- Session recordings stored in S3 should be protected with bucket policies that reflect the originating tenant’s access requirements.
Practitioners should verify that role trust policies, session tags, and S3 bucket ACLs are correctly scoped to prevent inadvertent data exposure.
Related CloudNinjas coverage: AWS.
What This Means For Practitioners
Teams that already run Playwright regression suites can extend coverage with intent‑driven tests without adding browser management overhead. Evaluate whether natural‑language test authoring fits your workflow, and review IAM role designs to ensure cross‑account isolation. Monitor CloudTrail and S3 access logs to confirm that session data remains confined to the appropriate tenant. Finally, consider the cost model of managed browser sessions versus self‑hosted alternatives before scaling the service across multiple product teams.