Rebuilding Alexa: How Amazon is mixing models, agents and browser-use for smarter AI

Amazon I bet on the interoperability of the agents and the mixture of models to make its new vocal assistant Alexa more efficient, reticating its flagship voice assistant with agenic capacities and tasks for using the browser.

This new Alexa has been renamed Alexa +, and Amazon stresses that this “more” version. For example, he can now proactively say users if a new book by their favorite author is available, or that their favorite artist is in town – and even offers to buy a ticket. Alexa + reason through instructions and “experts” taps in different knowledge bases to answer user questions and perform tasks like “where is the closest pizza? Will my colleagues love him? – Make a reservation if you think they will.

In other words, Alexa + mixes AI agents, the capacities for use of computers and the knowledge he learns from the larger Amazon ecosystem to be what Amazon hopes to be a more capable and more intelligent domestic vocal assistant.

Alexa + currently works on Amazon’s Nova models and models from Anthropic. However, Daniel Rausch, vice-president of Alexa and Echo from Amazon, told Venturebeat that the device will remain “agnostic model” and that the company could present other models (at least models available on the basis of Amazon) to find the best to accomplish tasks.

“”[It’s about] Choose the right integrations to complete a task, determine the right type of instructions, which it takes to finish the task, then orchestrate everything, “said Rausch. “The great thing to understand is that Alexa will continue to evolve with the best models available anywhere on the rocky substratum.”

What is the mixture of models?

The model mixture or the routing of the model allows companies and other users to choose the appropriate AI model to press a query request. The developers are turning more and more towards the mixture of models to reduce costs. After all, not all guests must be answered by a reasoning model; Some models do certain tasks better.

The unit of the cloud and Amazon AI, AWShas long been a supporter of the mixture of models. Recently, he announced a functionality on the basis called intelligent invites Routing, which directs the prompts to the best model and the best model size to resolve the request.

And, it could work. “I can tell you that I cannot say for a given answer from Alexa on a given task which model he uses,” said Rausch.

Agent interoperability and orchestration

Rausch said that Alexa + brings together agents in three different ways. The first is the traditional API; The second consists in deploying agents who can navigate websites and applications such as the use of Anthropic’s computer; The third is connection agents to other agents.

“But in the center of all this, orchestrate through all these different types of experiences are these basic LLM, very capable and ultramodern,” said Rausch.

He added that if a third -party application already has its own agent, this agent can always speak to agents working inside Alexa + even if the external agent has been built using a different model.

Rausch stressed that the Alexa team used the tools and technology of the Substratum Randage, including new multi-agent orchestration tools.

The anthropogenic CPO, Mike Krieger, told Venturebeat that Claude’s even older versions could not accomplish what Alexa + wanted.

“A really interesting” Why now? ” The moment is apparent in the demo because, of course, the models have improved, “said Krieger. “But if you have tried to do it with the Sonnet 3.0 or our level 3.0 models, I think you would have trouble in many ways to use many different tools at the same time.”

Although neither Rausch nor Krieger would confirm which specific anthropogenic model Amazon had used to build Alexa +, it should be noted that Anthropic published the Sonnet Claude 3.7 on Monday, and it is available on brooding.

Large investments in AI

The first brush by users with AI has gone through vocal assistants AI like Alexa, Google Home or even Apple Siri. These allow people to subcontract certain tasks, like lighting the lights. I do not have a Alexa or Google Home device, but I learned how one has one for the stay in a hotel. I could tell the Alexa to stop the alarm, light the lights and open a curtain while being under the covers.

But while Alexa, Google Home Devices and Siri became omnipresent in people’s lives, they started to show their age when the generator became popular. Suddenly people wanted more real -time responses from AI assistants and asked for more intelligent task resolutions, such as adding several meetings to calendars without the need for incentive.

Amazon admitted that the rise of the AI ​​generation, in particular the agents, allowed Alexa to finally respect its potential.

“Until this moment, we were limited by technology in what Alexa could be,” said Panos Panay, Amazon Learn and Services please, during a demo.

Rausch said that hope is that Alexa + continues to improve, add new models and hope to make more people comfortable with what technology can do.