Reconstruction of Alexa: Like Amazon models, agents and browser use for more intelligent AI

Amazon Betting on the interoperability and the modelmic from Agents to make its new Alexa-language assistant more effective and to convert its flagship voice assistant with agent functions and browser use tasks.

This new Alexa was renamed Alexa+, and Amazon emphasizes that this version “does more”. For example, the users can now proactively tell you whether a new book of your favorite author is available or that your favorite artist is in the city – and even offer to buy a ticket. Alexa+ reasons through instructions and blocks “experts” in different knowledge base to answer user questions and do tasks, such as “Where is the next pizza place for the office? Will my employees like it? – Make a reservation if you believe that you will do it. ”

In other words, Alexa+ combines AI agents, computer functions and knowledge that learns from the larger Amazon ecosystem in order to be what Amazon hopes is a more capable and more intelligent home -voice assistant.

Alexa+ is currently running in the Nova models and models from Amazon from Amazon Anthropic. Daniel Rausch, Vice President of Amazon and Echo, told Venturebeat that the device remains “model agnostic” and that the company could introduce other models (at least existing models Amazon’s basic rock) to find the best for the completion of tasks.

“(It is about) the selection of the right integrations to do a task, find out the right kind of instructions what is necessary to actually do the task and then the whole orchestrating,” said Rausch. “The most important thing that can be understood is that Alexa will continue to develop with the best models everywhere on the basic rock.”

What is model mix?

Model mixture or Model routing Let’s go companies and other users choose the corresponding AI model for queries for queries. Developers are increasingly turning to the model mix to reduce the costs. After all, not every request has to be answered by an argumentation model. Some models carry out certain tasks better.

Amazon’s Cloud and AI unit, AWShas long been a supporter of the model mix. It recently terminated a feature on the threat with the name Intelligent structureleads the requests to the best model and model size to fix the query.

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

Agent interiority and orchestration

Rausch said Alexa+ brings agents together in three different ways. The first is the traditional API; The second is the provision of agents that can navigate on websites and apps as possible The computer use of Anthropic; The third is the connection between agents with other agents.

“But at the center of the whole, these basic lines, which are very capable of being state -of -the -art LLMs, are orchestrating,” said Rausch.

He added, if an application of third -party providers already has its own agent, this agent can still speak to the agents who work in Alexa+, even if the external agent was created with another model.

Rausch emphasized that the Alexa team used the tools and technology of DRKONCHE, including new ones Multi-agent orchestration tools.

Anthropic CPO Mike Krieger told Venturebeat that even earlier versions of Claude will not be able to achieve what Alexa+ wants.

“A really interesting ‘why now?’ The moment is shown in the demo, because of course the models have gotten better, ”said Krieger.

Although neither intoxication nor warriors would confirm which specific anthropic model Amazon was used to build Alexa+, it is worthwhile to point out that anthropic Published Claude 3.7 Sonett On Monday and it is available on the basic rock.

Big investments in AI

The first brush of many users with AI came through AI language assistants such as Alexa, Google Home or even Apple’s Siri. These people let people outsource some tasks, such as switching on light. I don’t have an Alexa or Google home device, but I have learned how comfortable one can be when staying in a hotel recently. I was able to tell the Alexa that she should stop the alarm, switch on the light and open a curtain while she was still under the cover sheets.

But while Alexa, Google Home devices and Siri became omnipresent in people’s life, they showed their age when the generative AI became popular. Suddenly people wanted more real-time words from AI assistants and demanded more intelligent task resolutions, e.g.

Amazon admitted that the rise of gene AI, especially agents, made it possible to finally recover his potential.

“Until this moment we were limited by the technology in what Alexa could be,” said Panos Panay, SVP from Amazon, during a demo.

Rausch said that the hope is that Alexa+ will improve further, add new models and hopefully more people familiarize with what the technology can do.