Epsilon, a data, technology, and marketing-services company under Publicis Groupe, has implemented Amazon Nova Act to scale its intent-driven UI testing capabilities across its product teams. Facing the challenge of rapid application changes and managing a vast number of consumer profiles and brand relationships, Epsilon sought a UI testing solution that could maintain pace without creating additional maintenance overhead for its teams.
The Chenab Times has learned that Epsilon’s Agentic Platform team, comprising 50 engineers, developed a shared infrastructure designed to enable other product teams to build, deploy, govern, and operate AI agents safely. This platform provides a skills-based agent builder, federated identity, pre-flight authorization, Model Context Protocol (MCP) integration, and cross-account provisioning. When a product team adopts AI tooling, they gain access to these capabilities without needing to reconfigure identity, security, browser infrastructure, or model integration.
One of the managed capabilities delivered through this platform is UI testing. By leveraging Amazon Nova Act and the Amazon Bedrock AgentCore Browser Tool, Epsilon’s platform team was able to offer resilient, intent-driven testing as a service. This integration allows for testing that is more resilient to cosmetic UI changes and can be authored using natural language instructions, significantly reducing the burden compared to traditional selector-based testing.
Addressing the Challenge of Enhanced Coverage
Epsilon’s existing UI testing practices, built on a Playwright-based regression suite, were effective for deterministic checks but had limitations. These scripted tests were tightly coupled to specific selectors and page structures, making them susceptible to breaking due to minor cosmetic UI alterations, even when core functionality remained intact. Furthermore, the need for multiple product teams to achieve comprehensive testing coverage across their applications led to duplicative efforts in setting up browser infrastructure, managing dependencies, and integrating with identity systems.
The platform team’s objective was to layer intent-based testing atop the existing Playwright regression suite. This was to be delivered as a multi-tenant platform service, accessible through the same API surface that teams used for other agent capabilities. The goal was to enhance testing coverage without replacing the established, functional regression backbone.
Choosing Amazon Nova Act and Bedrock AgentCore Browser Tool
Epsilon evaluated several options based on two key requirements: natural-language test authoring and managed browser infrastructure that would align with their multi-tenant platform model. The combination of Amazon Nova Act and the Amazon Bedrock AgentCore Browser Tool met both criteria.
The natural-language steps feature allows test instructions to be written in plain English, making them more resilient to UI changes and accessible to a wider range of users, including those without deep engineering backgrounds. Steps like “Confirm the rewards balance is visible” are designed to be more durable than those relying on brittle CSS selectors.
The use of AWS-managed, ephemeral browser sessions via Amazon Bedrock AgentCore Browser Tool eliminates the need for teams to manage browser binaries or display servers. This capability also includes features such as live view for human observation or takeover, session recording to Amazon S3, and AWS CloudTrail auditing, all provided without additional configuration.
Authentication is handled through AWS Identity and Access Management (AWS IAM), aligning with Epsilon’s existing security framework. The solution is also cross-account friendly, allowing teams to run tests within their own AWS accounts under scoped roles, ensuring isolation and auditability through AWS Security Token Service (AWS STS) AssumeRole provisioning.
Solution Architecture and Implementation Details
Epsilon’s UI testing solution comprises three primary components: a platform API for workflow creation, a background task runner for execution, and the integration of Nova Act and Amazon Bedrock AgentCore Browser Tool to drive browser interactions.
The platform architecture follows a clear pipeline. Users create a workflow via the platform API, defining a target URL and a series of natural-language steps. Upon creation, the platform registers a Nova Act workflow definition on AWS. During execution, a background task runner, utilizing Celery for asynchronous job processing, initiates an Amazon Bedrock AgentCore Browser Tool browser session. Each step is then executed through Nova Act via a Chrome DevTools Protocol (CDP) WebSocket connection, incorporating conditional logic where necessary. The system captures screenshots for each step, tracks pass/fail status, computes an overall pass rate, and stores all results, including JSON data and screenshots, in Amazon S3.
The UI testing capability is exposed as a standard agent with two distinct skills: ui-test-recorder for interactively exploring scenarios and recording natural-language steps, and ui-test-executor for running these recorded steps and returning structured results. This approach ensures that onboarded teams consume UI testing in the same manner as any other platform agent, eliminating the need for them to manage additional infrastructure.
Core Integration and Example Workflow
The integration between Nova Act and Amazon Bedrock AgentCore Browser Tool is described as concise. It involves opening a managed browser session, connecting via the CDP WebSocket, and driving actions using natural language. This forms the foundation, which Epsilon enhances with its multi-tenant orchestration, result capture, and observability layers.
An illustrative example workflow demonstrates the automation of a ‘Create an Agent’ flow. This workflow encompasses login, form entry, conditional handling, scrolling, and verification. Key differentiating patterns of Nova Act highlighted include conditional logic, such as skipping a tour box if it appears, and scrolling until a specific field is visible. These dynamic actions are handled naturally through language understanding, a departure from traditional frameworks that would require explicit branching logic.
Best Practices for Implementation
The Epsilon team has outlined several best practices derived from their experience. These include writing test steps as intents rather than mechanics to ensure resilience to UI redesigns, isolating the Nova Act runtime in a dedicated service or container to prevent dependency conflicts, and reusing recorded trajectories to reduce cost and latency by skipping redundant LLM processing during repeat executions.
Designing for multi-tenancy from the outset, utilizing cross-account AssumeRole with scoped IAM policies for isolation and auditability, and leveraging managed observability features like live view and session recording are also emphasized. Finally, accounting for the viewport behavior of Amazon Bedrock AgentCore Browser Tool when prompting steps is noted as important for correct handling of content on different parts of the page.
Results and Impact
Epsilon’s UI testing capability is now operational across multiple teams, with its multi-tenant architecture proven in production. Internal benchmarks indicate a significant improvement in efficiency. A representative 18-step workflow took approximately 5–10 minutes to author with Nova Act, compared to 30–90 minutes for an equivalent Playwright script of 80–150 lines of code. This translates to an estimated 4-6x reduction in authoring effort.
Furthermore, saved trajectories achieve a 100 percent pass rate across repeated runs, and the use of natural language has lowered the barrier to test authoring, enabling team members beyond core test engineers to contribute. The integration with AWS CloudTrail and session recording has also provided a built-in audit trail, simplifying debugging and compliance. Crucially, the solution offers zero per-team infrastructure overhead, as onboarded teams access UI testing through the same API surface as other platform agents, without the need to manage browser binaries or display servers.
“Nova Act let us turn UI testing from a brittle, selector-chasing chore into a managed platform capability. Our tests now describe intent in plain English, run on AWS managed browsers, and every product team we onboard gets reliable testing without additional configuration.”
— Epsilon Agentic Platform Team
Epsilon’s implementation demonstrates how intent-based browser automation can be used as a managed service, offering efficiency gains and broader accessibility for quality assurance processes.
❤️ Support Independent Journalism
Your contribution keeps our reporting free, fearless, and accessible to everyone.
Or make a one-time donation
Secure via Razorpay • 12 monthly payments • Cancel anytime before next cycle
(We don't allow anyone to copy content. For Copyright or Use of Content related questions, visit here.)
The Chenab Times News Desk
