AWS introduces agent-driven synthetic monitoring with the Amazon Nova Act model and Amazon Bedrock AgentCore
AWS described an architecture for synthetic monitoring of web applications using the Amazon Nova Act model and Amazon Bedrock AgentCore. The agent tracks the UI via screenshots instead of DOM selectors, which is meant to reduce the fragility of automated tests.
Amazon (AWS) published a procedure on its Machine Learning Blog for synthetic monitoring of user journeys (login, purchase, form submission) using an agent built on the Amazon Nova Act model and running in the Amazon Bedrock AgentCore Runtime environment. The goal is to replace traditional tools like Selenium or Playwright, which rely on DOM selectors and CSS classes and break easily with any UI change, which according to the article leads to high costs for maintaining test scripts.
Amazon Nova Act is a multimodal language model that, according to the company, evaluates the UI based on screenshots of the screen rather than element identifiers in the page code. This is meant to allow it to adapt better to visual changes without the need to modify scripts. The company states that in early deployments with enterprise customers, the model achieved accuracy of over 90% on browser workflows, i.e., tasks involving browsing and controlling the web interface. The source recommends that teams verify the accuracy on their own websites and supplement the monitoring with retry logic for cases where the agent does not adapt correctly on the page.
The described architecture combines Amazon Nova Act, Amazon Bedrock AgentCore Runtime, and Amazon EventBridge Scheduler, which triggers the agent according to a schedule. The agent defines user journeys as sequences of actions (the act() function) and checkpoints that explicitly verify the result (the act_get() function with a boolean schema) — for example, that search results were displayed or that the cart contains items. According to the testing described in the source, a typical six-step journey takes 2 to 4 minutes depending on page load times. The solution is intended for e-commerce, financial services, travel, SaaS, and healthcare; AWS provides a sample repository with the complete implementation.
The rest of the article describing detailed deployment setup and scheduling was not available. For details, see the source article.
Why it matters
For companies running web applications, this is an alternative to traditional end-to-end testing that is meant to reduce the cost of maintaining test scripts when the interface changes. Teams responsible for website reliability (SRE, QA) can consider deploying agentic monitoring instead of paid premium modules of commercial monitoring platforms, but they must account for verifying the accuracy on their own website, since the stated over 90% is a company claim from early deployments, not an independently verified benchmark.
Two audiences, two different impacts
What this means
For individuals
Developers and QA engineers working on end-to-end testing of web applications can try this approach as a replacement for selector-based scripts that require frequent fixes.
For a business
Companies running customer-facing websites (e-commerce, finance, SaaS, travel, healthcare) can reduce the cost of maintaining monitoring scripts and detect failures in key user journeys faster, but the claimed accuracy of over 90% is so far an AWS claim from early deployments, not an independently verified figure.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.