Amazon's smarter Alexa slips to 2025 as basic tasks fail

Amazon's ambitious plan to rebuild Alexa into a full-fledged AI chatbot has reportedly hit another wall.

According to Bloomberg, the tech giant has officially pushed the rollout of its next-generation, LLM-powered Alexa into 2025. Originally showcased last September and slated for a fall release, the revamp was meant to deliver conversational, ChatGPT-like intelligence. Instead, fall came and went without Amazon's traditional splashy hardware event, and the company has quietly closed the beta access previously triggered by saying "Alexa, let’s chat."

Slower, wordier, and failing at the basics

According to sources cited by Bloomberg, early testers found the updated assistant frustrating to use. The new Alexa responded slowly, spoke with a stiff tone, and suffered from classic LLM hallucinations. Worse, it developed a habit of showing off.

In one test example, asking where Justin Timberlake and Janet Jackson performed used to yield a straightforward "2004 Super Bowl." The new model instead rambled on with unprompted trivia about the infamous wardrobe malfunction.

More critically, tying a pre-trained language model to the assistant broke its core utility. The new system struggled with basic, deterministic tasks like setting kitchen timers or reliably turning smart lights on and off. While the old Alexa had plenty of flaws, it could consistently control connected gadgets. The upgraded brain made it chatty, but bad at everyday chores.

A lack of vision at the top

Behind the scenes, Bloomberg reports that Amazon CEO Andy Jassy has yet to define a compelling internal vision for an AI-powered Alexa. Publicly, Jassy has talked about re-architecting the assistant's brain, but leadership has struggled to prove what practical value an LLM brings beyond small talk.

Meanwhile, former Microsoft Surface chief Panos Panay has now spent a year running the devices division. While Panay is attempting to bring high-end hardware polish to utilitarian gadgets, Amazon is still reckoning with a decade of distributing dirt-cheap devices that ran stagnant software.

Why it matters

Nobody bought a smart speaker because they wanted a roommate who talks your ear off but refuses to get off the couch to hit the light switch. Pushing the release back to 2025 is an embarrassing delay for Amazon, but shipping an assistant that can't reliably turn off the lights would have been catastrophic.

This mess exposes the hardest problem in consumer AI today: merging the creative, probabilistic nature of LLMs with the strict, deterministic accuracy required for smart home commands. Replacing rule-based voice controls with a language model turns simple instructions into fuzzy gambles.

In our view, Amazon made the right call by pulling the plug on the "Let's Chat" preview before wider exposure. Apple and Google are wrestling with the exact same transition with Siri and Gemini. Until these companies figure out hybrid architectures that keep instant device control rock-solid while layering intelligence on top, these assistants will remain smarter on paper and worse in your living room.