Amazon aims at AI agents with Nova Act for browser use
The vendor introduces a new model and SDK that developers can use to build AI agents to complete tasks on the web. But it's not the first -- or even the second -- to do so.
Amazon introduced Amazon Nova Act on Monday, a new AI model that can be trained to carry out agentic actions such as performing search, answering questions or checking out within a web browser. It also introduced the Amazon Nova Act SDK. Both releases keep Amazon as a key player in the agentic market.
Amazon released the new model alongside nova.amazon.com, a new website that enables developers to easily explore Amazon Nova, the vendor's latest generation of foundation models.
AWS first introduced Amazon Nova foundation models at re:Invent 2024 last December. The family of models comes in three sizes: Nova Micro, Lite and Pro. According to Amazon, the models generate text from different modalities.
Developers can also access a research preview of the Amazon Nova Act SDK on nova.amazon.com and use the SDK to build agents that can complete tasks in a web browser. This continues the trend in the generative AI market toward AI agents and agentic systems.
Amazon is also not the first vendor to introduce an agentic browser-use tool. Last fall, Anthropic released a beta of Computer Use, an API that enables developers to direct Claude to use the computer. The API interacts with the computer the way a human does -- by looking at the screen, moving the cursor and typing.
In December, Google DeepMind launched a research prototype called Project Mariner, which Google claimed can understand and reason through information in a user's browser. In January, OpenAI launched Operator in research preview. The tool can perform tasks for users on the web. OpenAI's computer-using agent powers Operator.
Not behind and a developer mindshare
With many other foundation model providers having released computer and web browsing functions, Amazon might appear to be behind in the market. However, according to Bradley Shimmin, an analyst at The Futurum Group, that is not true.
He said Amazon has been studying the benefits of agentic AI for a couple of years and discussing it with analysts, even if it did not outwardly produce a product.
"They were very early to look at the value proposition of agentic processes," Shimmin said. "They've been thinking about this for a while and are not just late to the party."
While there seems to be little differentiation between Amazon Nova Act and other browser-use tools, the SDK takes things a step further, enabling developers to take Amazon's agent APIs, such as search and speech-to-text agents, and embed them into the application they're building, according to Gartner analyst Arun Chandrasekaran.
"Amazon has a lot of mindshare with those developers, so that in some sense differentiates them," Chandrasekaran said.
Plus, Amazon Nova Act shows that Amazon is also becoming more comfortable and bolder with its foundation model, Chandrasekaran said.
"It looks like they're ready to give a lot more branding and prominence for their first-party model," he said. He added that knowing how Amazon plans to position this relative to the Anthropic models would be interesting.
Enterprise need
With Nova Act and the Nova model family, Amazon reflects a need within the enterprise for easy-to-use AI functionality, Shimmin said.
"Enterprises and consumers alike want a solution that isn't just, 'Here's the manual, figure it out,'" he said. "They want something that brings 80-plus percent of what they're after to them on the first pass. And so I see the Nova family of models reflecting that."
This kind of browser or even machine use more broadly can open up many opportunities to automate.
Bradley ShimminAnalyst, The Futurum Group
Shimmin said when it comes to the growth of agentic tools that use the computer, it all comes down to automation.
"This kind of browser or even machine use more broadly can open up many opportunities to automate," he said.
However, those using browser-use tools must attempt to do so safely.
"There's a reason why most of the solutions that we have in the market right now allow for browser use. Try to put that use inside a containerized platform," Shimmin said.
He added that containerized platforms help secure enterprises so that if something goes wrong, enterprise data remains secure. Therefore, enterprises exploring computer-use tools will likely go to partners of foundation model providers such as Azure or AWS rather than the foundation model providers themselves.
Chandrasekaran said enterprises considering different computer-use tools should also make sure that the tools can integrate with their workflow.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.