If you were underwhelmed by Amazon’s next-gen voice assistant, [Alexa+](https://gizmodo.com/heres-everything-alexa-can-supposedly-do-2000568873), there may be good news: Amazon is reportedly already working on taking things up a notch—or at least *trying *to. According to documents seen by [Business Insider](https://www.businessinsider.com/amazon-moonraker-project-alexa-agentic-cost-2026-7), Amazon is in the process of developing an Alexa project codenamed “Moonraker,” which aims to imbue Alexa with more agentic capabilities.
While Alexa+ can already handle some tasks (ordering food online, for example), Moonraker—which is the name of a James Bond film from 1979, by the way—is reportedly aiming to expand those capabilities. One of the intended improvements of Moonraker is the ability to handle more sophisticated multistep commands, like “book me a ride and text my friend,” allowing users to cram more into a single query.
There’s not much detail about the Moonraker project outside of its goal for dialing up the AI agent of it all, but Business Insider notes that it’s shaping up to be costly. Planning documents suggest that Moonraker could be the company’s “highest cost” initiative, projecting upwards of $100 million in GPU costs in 2026.
Even without much detail, the interest in spending even more time and money on Alexa says a lot about how the current crop of voice assistants has been received. While companies like Google and Amazon have billed their new voice assistants as being more sophisticated and intuitive, that sentiment hasn’t exactly been echoed by most people using them.
I haven’t gotten a chance to put Alexa+ through its paces, but I have used Google’s next-gen Gemini for Home at length and have been pretty consistently underwhelmed. Don’t get me wrong, it does improve on some aspects of the smart home experience, like being able to understand natural language and handle multistep commands more adeptly, but the pain points are about the same as they were before. Queries are misunderstood or misheard, information can be inaccurate, and there’s overall just a lot of variability between interactions, which doesn’t instill trust or compel me to use Gemini for Home more than I used Google Assistant.
And sure, making Alexa or Gemini for Home more agentic could add some utility, but if the results are as finicky as they are in the current iteration, would anyone really trust new models or features to take the wheel? Amazon, for its part, seems to think they might, which means its penchant for pouring money into Alexa looks like it may continue forth. And if Apple’s Siri AI is as useful as some early test results would suggest, the race for less sh*tty voice assistants could wage for a long, long time.