AWS case study: Epsilon’s intent-driven UI testing with Amazon Nova Act and AgentCore Browser Tool
AWS describes how Epsilon built a managed UI-testing service on Nova Act and Bedrock AgentCore Browser Tool, with workflow APIs, step-level execution, screenshots, and multi-tenant isolation; the source is company-reported implementation detail, not independent validation.
What happened
AWS’s machine-learning blog says Epsilon, a Publicis Groupe company, uses Amazon Nova Act and Amazon Bedrock AgentCore Browser Tool for intent-driven UI testing. The reported design has three parts: a platform API that accepts a target_url and natural-language steps, a Celery-based background runner, and a browser integration that executes steps over CDP WebSocket connections. AWS says the system records screenshots, per-step pass/fail status, an overall pass rate, results JSON, and session IDs back into the platform. The post also says Epsilon recommends isolating the runtime in a dedicated service, reusing recorded trajectories for repeat runs, and using cross-account AssumeRole for multi-tenant isolation and auditability. This is a company publication, so it supports the described implementation and guidance, not independent performance claims.
Why it matters
FACT: AWS says Epsilon’s UI-testing service lets product teams submit a target_url and natural-language steps through a platform API, then runs them on managed browser infrastructure with step-level pass/fail tracking and screenshots. INFERENCE: For developer platforms, the practical choice is whether to centralize brittle browser automation as a shared service rather than let each team maintain its own selectors and test harnesses.
Engineering impact
FACT: The reported design combines a platform API, Celery-based execution, and Nova Act running over CDP WebSocket sessions in an AgentCore Browser Tool browser, while also recording screenshots, JSON results, and session IDs. INFERENCE: Teams evaluating a similar stack should judge it on orchestration simplicity, artifact traceability, and whether the runtime can stay isolated enough to avoid dependency conflicts and tenant leakage.
Leadership impact
FACT: AWS reports Epsilon recommends cross-account AssumeRole with scoped IAM policies, dedicated runtime isolation, and trajectory reuse for repeated runs. INFERENCE: Leadership deciding on platform investment should treat this as an operating-model choice: shared test automation can reduce per-team reinvention, but it also requires explicit governance for identity, auditability, and service ownership.
Editorial judgment
This changes the decision for teams evaluating browser automation platforms: the relevant tradeoff is managed, multi-tenant test execution versus building browser, identity, and observability plumbing themselves. The value is concrete, but the evidence remains a vendor-published case study rather than independent benchmarking.
Watch next
FACT: The source says repeat runs can replay saved trajectories to reduce LLM thinking, and that the managed browser tool provides live view, session recording, and CloudTrail integration. INFERENCE: The next decision to watch is whether teams need evidence from their own applications on repeat-run cost, browser-session reliability, and whether the built-in audit surfaces satisfy internal compliance reviews.
Limitations & uncertainty
The evidence is a company-published case study, so it supports AWS and Epsilon’s described implementation and recommendations, not independent validation of performance, reliability, or broad applicability across other environments.