By Amazon AGI
Jul 16, 2025
We believe that agents will be the atomic building blocks for the new era of collaborative computing. In our view, the most valuable agentic use cases have yet to be built--so we’re laser focused on developing the key capabilities that will enable useful agents at scale.
A few months ago, we launched a research preview of the Nova Act SDK. Since then, we’ve been thrilled to see the community using Nova Act’s capabilities to build complex, reliable agents. Now in a limited preview, we’re ready to work with customers to productize agents. We’ve added new enterprise-grade capabilities through AWS integration to provide developers a pathway to production. This preview represents a significant step forward in our mission to enable developers to build agents that can reliably perform real-world tasks at scale. Although we’re still in the first inning of agent R&D as a field, our early customers are already finding significant value in the workflows they’ve built with Nova Act.
We hear again and again from customers that reliability has been a bottleneck to production-grade agents. We’re excited to see Nova Act achieving 90%+ end-to-end reliability on early enterprise customer use cases, which means that Nova Act is consistently completing tasks ranging from automated quality assurance to complex form handling and process execution. To do this, we’ve extended Nova Act’s industry-leading capabilities on basic UI interactions like date picking and dropdown selection to more complex tasks like multi-element form filling. Since delivering industry-leading performance on benchmarks, like ScreenSpot and GroundUI Web, we've also improved reasoning capabilities to help with robustness to website changes and recovering from mistakes.
With these new capabilities, we’re making it easy for select customers to deploy their agents in production environments. Customers can now use Nova Act with enterprise-grade features like AWS IAM, secure storage and data controls through S3, and integration with the Amazon Bedrock AgentCore Browser for a seamless path to production.
While we love a cool demo as much as anyone, we’ve been inspired by watching Nova Act in action - both in how it helps our customers who are using it for early production use cases achieve meaningful results and how it handles unexpected curveballs in new environments:
One natural language act() prompt led to 93 reliable agent steps: Rackspace
Technology is working with Alvee Health to register members for public benefits
using Nova Act. "Many registration forms for public programs are long and
confusing, so members often don't apply for the help they need," said Nicole
Cook, CEO, Alvee. "With Nova Act, we're not just simplifying paperwork – we're
helping ensure timely, accurate access to the resources that support healthier
lives. We expect this innovation to increase successful benefit registrations by
30%, and improve overall case load by up to 10x, allowing healthcare providers
to focus more on patient care and less on administration. This is a prime
example of how AI can be used to support well-being and improve overall health
for communities.”
We were surprised to find that instructions written for humans seem to work as agent instructions: Tyler Technologies, a leading public sector software provider, is using Nova Act to automate software testing and improve the reliability of its releases. “Nova Act’s natural-language interface lets us convert our existing manual test plans directly into automated suites, without writing a single line of code,” said Franklin Williams, president of data and insights, Tyler Technologies. “This saves us hundreds of hours while expanding test coverage and increasing product quality.”
A single Nova Act script is able to generalize across diverse environments: Navan, a leading travel and expense management platform, is using Nova Act to simplify its travel agents’ workflows by using a single Nova Act script to fill out diverse payment forms from a range of hotel brands. Sarav Bhatia, director of software engineering, Navan, said, “Prior efforts to tackle this use case with other computer use tools failed — Nova Act is able to generalize across different web environments. We’re excited to put our prototype into production and expand to even more vendors, which will allow our small team to meet increasing customer demand.”
Nova Act is trusted to streamline essential administrative workflows: Automation Anywhere, a global leader in agentic process automation, is expanding its automation capabilities through Nova Act to handle time-consuming administrative workflows like credential verification - a critical, repetitive task that’s essential for day-to-day operations. “By deeply integrating Amazon Nova Act into our Process Reasoning Engine (PRE), we’ve unlocked a major leap forward in computer use for enterprise automation,” said Adi Kuruganti, chief product officer, Automation Anywhere. “Our goal-oriented AI agents don’t just mimic clicks, they reason through UI-based processes in real time, navigating complex websites with human-like expertise. This opens the door to automating previously out-of-reach use cases like healthcare program enrollment testing, where accuracy and scale are essential.”
Visible chains of thought build confidence in Nova Act for UI testing: Katalon, an automated software quality platform, has developed a UI interface that converts JSON test cases into Nova Act scripts and executes the QA tests. Said Coty Rosenblath, Chief Technology Officer, Katalon: "Two capabilities stand out for us. First, the visible reasoning feature offers our developers a clear view into the model's decision-making process, making it easier to steer toward successful outcomes. Second, structured data extraction is critical for validating application states in our automated testing workflows. When benchmarked against other solutions, Nova Act consistently delivers higher accuracy. That level of performance builds our confidence in its potential as a core capability in our ecosystem."
Nova Act enables scaling to millions of users: Amazon Alexa+, our next generation assistant powered by generative AI, leverages Nova Act for core browser automation at the heart of the Alexa AI Web Action developer offering. Mark Yoshitake, director and general manager, Alexa AI Developer Technologies, said: “Integrating the Nova Act SDK has allowed us to abstract the primitives of agentic browser use like direct model management, perception, and actuation across a range of web tasks. With the Nova Act SDK handling complex browser interactions, we’ve been able to focus on delivering the consumer-facing primitives like agentic payment, third-party account integration, and other features that make Alexa+ even more powerful for developers. We’re excited about what's to come as we scale these capabilities—this is just the beginning.”
While we're encouraged by early results, we have a lot more work to do. Here are some of the research problems we are excited to solve for Nova Act:
We believe that building useful AI agents requires both technical progress and careful real-world deployment to understand how these systems perform in practice. This launch represents our commitment to advancing both fronts simultaneously. If you’re excited about frontier research that ships quickly into useful agents, join our team!
To get started with the Nova Act SDK, visit nova.amazon.com/act. Once you’re ready to bring your prototype to production, reach out to our team using this interest form for access to our preview.