Amazon arrives at the AI Agent party with Nova Act, a browsing agent SDK

Stop the clocks once more.

The AI world has changed, again.

This time it's the 800lb gorilla in the form of Amazon arriving at the party with a series of fascinating developer previews based on enabling a highly focused, intensively trained web browsing agent.

It's called Nova Act and it's Amazon's offering to enable you to build systems, processes and capabilities to help people get things done, accurately and effectively.

If you've been watching and playing with the likes of OpenAI's Operator, I'd suggest that this is similar in context, although it's a very different approach.

The Amazon team have really, really focused the models' approach to making sure it doesn't get stuck on some of the more complicated aspects of the browser experience, including for example, date picking, drop-downs, pop-ups and so on.

The team points out:

Nova Act is focused on reliable building blocks that can be composed into more complex workflows. Many agent benchmarks measure model performance on high-level tasks, where state-of-the-art models achieve 30% to 60% accuracy on completing tasks in web browsers. But agents must be reliable to be truly useful — we’ve focused on scoring >90% on internal evals of capabilities that trip up other models, like date picking, drop downs, and popups, and achieve best-in-class performance on benchmarks like ScreenSpot and GroundUI Web which most directly measure the ability for our model to actuate the web.

What this means is that you can essentially string together a series of tasks – including adding in a bit of Python code as needed – to get something done. One of the team highlights Nova Act ordering him the same salad every week, for example.

You can sit and watch the agent do its thing, or you can opt to have it run in 'headless' mode – whereby it just executes the task you've defined without any need to watch the browser activities.

Nova Act is the first step in our vision for building the key capabilities that will enable useful agents at scale. This is an early checkpoint from a much larger training curriculum we are pursuing with Nova models. To truly make agents smart and reliable for increasingly complex multi-step tasks, we think agents need to be trained via reinforcement learning on a wide range of useful environments, not just via supervised fine-tuning with simple demonstrations into an LLM. 

From these statements and the surrounding commentary across the various videos and releases, you can begin to get a picture of what the Amazon team is thinking in terms of scale and capability. It's impressive, compelling and exciting.

If you're in the United States, you can go and find out a lot more at https://nova.amazon.com – otherwise, if you're external (like me) you'll need to sit and wait before you can start playing with it.

One to watch!