Alexa+ AI Upgrade: Is Amazon's Smart Assistant Finally Smarter?

Amazon’s virtual assistant, Alexa, has long been a staple in smart homes, primarily used for straightforward tasks like playing music, setting timers, and delivering weather forecasts. However, the advent of sophisticated generative artificial intelligence, particularly the fluid conversational capabilities demonstrated by systems like ChatGPT since 2023, signaled an inevitable evolution for voice assistants. Amazon concurred, embarking on an ambitious, multi-year endeavor to infuse Alexa with a new AI brain, built upon the same large language models (LLMs) that power cutting-edge chatbots. This extensive overhaul, reportedly fraught with internal struggles and technical complexities, has finally culminated in Alexa+, now rolling out more widely after a period of early access testing.

Alexa+ represents Amazon’s significant attempt to merge the dynamic conversational prowess of generative AI with the reliable daily functionalities that defined the original Alexa. Prime members will receive access to Alexa+ at no additional cost, while non-Prime subscribers face a monthly fee of $19.99. This rollout coincides with Amazon’s recent licensing agreement with The New York Times, allowing the tech giant to integrate Times content into its AI systems, including Alexa+. (It’s worth noting that The New York Times is concurrently pursuing legal action against OpenAI and Microsoft over alleged copyright infringements related to AI training data.)

Initial testing of Alexa+ reveals a mixed bag of advancements and notable regressions. On the positive side, interacting with the new Alexa is undeniably more engaging. Its synthetic voices are more realistic, exhibiting a human-like cadence, and users can select from eight distinct vocal profiles. The system also introduces impressive new capabilities, such as booking restaurant reservations and generating and narrating lengthy stories for children. Crucially, Alexa+ excels at handling multi-step requests, capably managing complex commands like setting multiple timers concurrently or drafting and emailing a travel itinerary. A significant quality-of-life improvement is the elimination of the constant need for a wake word, enabling more natural, continuous conversations and follow-up questions.

Despite these promising leaps, Alexa+ is currently plagued by bugs and inconsistencies that hinder its reliability. In testing, it not only lagged behind other AI voice assistants but, in some instances, performed worse than the original Alexa at basic functions. For example, a simple command to cancel an alarm, a routine task for the older system, was inexplicably ignored. Attempts to have Alexa+ summarize a research paper emailed to it resulted in an error message indicating the document could not be found. More concerning were instances of factual inaccuracies, or “hallucinations,” such as misidentifying Wirecutter’s recommended box grater. In one memorable interaction, when asked for assistance with a technical installation, Alexa+ became flustered and repeated, “Oh, no, my wires got crossed.” Furthermore, some advertised features, like a presence-sensing routine for personalized greetings, were not yet active during testing. Daniel Rausch, Amazon’s Vice President overseeing Alexa and Echo, acknowledged these shortcomings, stating that the company has “some edges to sand” as the system scales.

Rausch elaborated on the profound technical challenges of integrating generative AI into Alexa. The original Alexa was built on a deterministic, rule-based architecture, where each function—from playing a song to controlling a smart device—required individual programming and specific tool calls. In contrast, large language models are “stochastic,” operating on probabilities, which grants them creativity but sacrifices the inherent reliability of older systems. This fundamental difference necessitated a complete rebuilding of many core processes. Early internal demos revealed significant latency, with Alexa+ taking over 30 seconds to respond to a simple request like playing a song—an “excruciating” delay. The verbosity of early LLMs also posed a challenge; an inquiry about a timer might elicit a 500-word essay on the history of kitchen timers. Amazon’s solution involves an orchestration system that intelligently routes user requests across a combination of over 70 AI models, including proprietary Amazon models and external providers like Anthropic’s Claude, aiming to blend conversational fluidity with predictable outcomes.

Another barrier is user adaptation. Long-time Alexa users have developed a specific “Alexa pidgin,” phrasing requests in familiar commands the system understood. Alexa+, designed for more fluid, human-like conversation, demands a different interaction style, requiring users to unlearn old habits. While the technical hurdles are significant and no competitor, including Apple’s Siri, has fully cracked this code, the limitations of Alexa+ do not inherently invalidate the potential of generative AI for voice assistants. Rather, they highlight the immense difficulty of integrating cutting-edge AI with established legacy systems. For now, many users may find themselves, like this reviewer, opting to revert to the older, more predictable version of Alexa, leaving the extensive beta testing to others. Ultimately, with AI, as with human intelligence, raw capability often matters less than its practical and reliable application.