Enough chit-chat: Alexa is done waiting for the agentic AI future — I spoke to Amazon's Daniel Rausch to find out more

After years of pronouncements that agentic AI was just around the corner, it suddenly crawled into view this year in the form of OpenClaw, a lightweight, flexible means of orchestrating agents.

And I spoke to Daniel Rausch, vice president of Alexa and Echo at Amazon, to find out how the Alexa team is taking advantage of state-of-the-art models with Alexa+.

OpenClaw gives AI digital “hands” that can interact with your PC, the web, and other agents, the same digital tools that human workers would use. Particularly if you spring for a well-equipped Mac mini (the go-to box as macOS is OpenClaw’s original platform) or a PC with a powerful GPU, the local option can save money in the long run.

But if you’re not careful about permissions and security, configurations can expose passwords and other highly sensitive data.

OpenClaw has already spurred enhancements (like Nvidia’s Nemobot) and rivals (like Perplexity Computer and Claude Cowork Dispatch). These can cost $100 per month or more, and may be well worth it for someone seeking to automate much of a solo business or side hustle.

Yesterday’s friendliness, tomorrow’s foundation

But for mainstream consumers, their best option for using AI to just get stuff done may come via one of the most familiar names in digital assistants.

Rausch explains that Alexa has always been focused on getting things done for customers in the real world.

"There are things we've believed to be true about what we're going for with Alexa. And we definitely believe that being the world's best personal AI assistant is a worthy thing to be building for customers and that we can get a lot of magical things done. The vision for Alexa has been very consistent over time, but the technology has held us back at times from becoming the ultimate assistant."

Now, after an 18-month gestation that rebuilt Alexa’s architecture from the ground up while maintaining compatibility with 97 percent of the hundreds of millions of devices on which it runs, users are engaging with Alexa Plus two to three times more than they did with the original assistant. In an even more encouraging sign, the engagement level continues to grow over time.

For example, Amazon has seen week-over-week interactions growth in such usage staples as listening to music streams as generative AI takes discovery deep into genres. Smart home interactions--one of the few ways first-gen assistants were able to interact with the real world - are up more than 50 percent versus classic Alexa. Rausch attributes this to Alexa+’s greater understanding of intent.

“You can truly say what you want and it happens,” he said.

Unlike many in the current wave of AI start-ups serving enterprises or creating core tech for developers, Alexa has maintained its predominant consumer focus. Today, for example, you can order groceries or an Uber with Alexa+. At Amazon’s Devices and Services event last fall, the company flashed a slide packed with integration partners.

Discovery and danger zones

Rausch also shares that a good deal of the dialogue consumers are leaning into includes other kinds of discovery, asking Alexa about its new capabilities. While such dialogue is also occurring with Gemini, ChatGPT and Claude, Alexa’s voice-first usage model likely leads to a more free-form exploration.

In contrast, other AI chat interfaces have taken to compartmentalizing their capabilities: create an image, help me learn, etc. (Claude’s have taken a looser approach, with an invitation to discuss “life stuff.”)

But while exploration can be rewarding, rabbit holes can lead to dark places. In the last few years, passing the once-formidable Turing Test (which posed whether a computer could convince a human that they were talking to another human) has become child’s play for today’s agents, at least in limited interactions.

Getting deep into a process with some chatbots can trigger a persistent drive to egg you on, to do one more thing. In some cases, humans have formed relationships in which they’ve got engaged to their AI chatbot. Other relationships have turned far more tragic. Social media is an attention sieve compared to what targeted AI can deliver if the business model calls for it.

Rausch acknowledges that he has seen experiences from other companies where the goal is clearly to just continue the interaction. However, while he notes that Alexa will often ask follow-ups, it is not designed to drive engagement for its own sake.

“We want to help customers get something done. That is the design plan of Alexa. All of our work cascades to getting something done,” he says.

As for guardrails to protect against AI that might become manipulative, Rausch points to Amazon’s eleven-year track record in providing safe AI and a study that shows that Alexa has been particularly resilient to distorting the truth.

While much of that was in an era where the most dangerous mistruth Alexa might tell you was about the chance of rain, Rausch touts the work that Amazon’s AI trust and safety team works to discourage isolation and emotional dependence.

He revealed to me: “One of our tenets of trust is that, when you talk to Alexa, she will help you prioritize your ‘people friends.’ Our goal is to help customers make those relationships strong."

This time, it's personal

While OpenClaw may be pushing the envelope far beyond what Alexa can do today, Rausch sees it as sharing the same design goal, but without the need to dedicate a PC, flash a Raspberry Pi, or cash in some crypto to afford more tokens.

While many OpenClaw pioneers can justify the time and expense as helping them automate a small business or work a side hustle, Alexa’s rolled-up sleeves will take on more mainstream tasks. Alexa and Gemini, for example, have recently implemented agents that work asynchronously.

Rausch cites rapidly growing interest in a Ticketmaster-monitoring task that can get an early jump on tickets in high demand.

But if Alexa+ raises the stakes on what the assistant can deliver, it also raises the stakes of what it needs to perform optimally. Last fall’s Echo updates added sensing technology to provide more context around what’s happening in the home.

And while Alexa has no smartphone platform beyond the Alexa app (yet?), Ring and Blink customers can offer it a sense not only of “seeing what I’m seeing,” but of “seeing what I’m not seeing when I’m not there.”

That said, to truly deliver the optimal level of personalization in which AI, particularly AI focused on results, Amazon will have to step up its focus on personal devices. The company took a step toward that last year with the purchase of Bee.

The company’s deceptively simple wristband allows recording of notes and meetings, but does more processing of them into next steps within its app. In addition, it’s been a while since we’ve seen new iterations of the Echo Buds and Echo Frames, both last updated in 2023. Alexa’s first decade was about being everywhere. The roadmap for the second is about getting things done anywhere.

Read more: We tested out the best AI tools you can get right now.