Amazon plans to reposition its digital voice assistant Alexa as an AI "agent" that can handle practical tasks like restaurant recommendations or controlling smart home devices. However, the redesign is hampered by technical challenges such as "hallucinations" and slow response times, according to Rohit Prasad, head of the Artificial General Intelligence (AGI) team at Amazon. "Hallucinations must be almost zero," said Prasad.

The transformation requires replacing the previous algorithm with powerful generative AI models. For two years, Amazon has been working on equipping Alexa with these models, capable of generating responses and addressing numerous third-party services. The complexity lies in integrating legacy systems and meeting high reliability and security standards.

Prasad stated that the recent launch of the in-house developed Nova models aims to optimize speed, cost, and reliability. However, further steps are required, such as the implementation of child safety filters and the integration of individual applications like Ring doorbells.

Former employees report that Alexa's original software was cumbersome and difficult to adapt. The integration of generative AI was associated with numerous organizational and technical hurdles. Thomas Lindgren, co-founder of Swedish Alexa partner Wanderword, said: "We are waiting for details. Initially, Amazon was more open.

Amazon is currently examining monetization models such as a subscription or revenue sharing. Despite setbacks and massive layoffs in 2023, the strategy remains clear: Prasad emphasized that Amazon's AI development aims to "deliver customer value and impact, which is more important than ever in the era of generative AI.