Alexa Plus After Nearly a Year: The King of Smart Home That Can't Seem to Leave the House

Amazon's voice assistant revolution has been promised for years — but how much has Alexa Plus actually delivered? In the latest episode of The Vergecast, host David Pierce sat down with journalist Jen Tuohy, who has been testing Alexa Plus for nearly a year, for a candid retrospective on the product. The conclusion is telling: as a smart home assistant, Alexa Plus has become the best in its class — but the moment it tries to step outside the home and become a general-purpose personal assistant, it hits a wall at every turn.

Smart Home Experience: From "Reciting Incantations" to Natural Conversation

Alexa Plus launched with two core promises: freeing users from rigid command syntax so they could talk to their home more naturally, and enabling more complex tasks through simpler interactions. Jen Tuohy believes Amazon has done a genuinely good job on the first goal, and has "only done half the job" on the second.

The biggest experience upgrade comes from voice-created automation routines. Previously, users had to dig into the app and configure everything manually — most people didn't have the patience or the time. Now, you can just tell your Echo Show something like "turn on the lights and lock the door at a set time every night," and the system handles it automatically; if something's off, you can fine-tune it in the app. Jen called this "a genuine game-changer." David shared his own example: setting up a simple routine like "turn off all indoor and outdoor lights after the front door is locked" — something he'd never bothered to do before because it seemed like too much trouble — now takes a single sentence. "That's power," he said.

This is precisely where large language models (LLMs) excel — translating a user's scattered, imprecise expressions into precise instructions a machine can execute. As David put it, the logic is the same as a meeting notes tool that extracts action items: you don't have to "speak the computer's language," and it can generally figure out what you mean.

The role LLMs play here is essentially that of an "intent translation layer." Traditional voice assistants relied on strict intent recognition frameworks — users had to say the exact trigger words and parameters the system expected, and any deviation would break the match. The new generation of LLM-based assistants, pretrained on vast amounts of natural language, has developed a tolerance for ambiguous phrasing. Take "turn on the lights and lock the door at a set time every night" — the system must simultaneously understand the time condition, multi-device orchestration, and the details the user left implicit (what exactly does "a set time" mean?), then map all of it to an automation script that the underlying smart home protocol (such as Matter or Zigbee) can execute. Previously this required the user to configure each step manually in an app. With LLMs, the conversion from "vague description" to "executable rule" happens automatically. This also explains why smart home scenarios are currently one of the smoothest areas for AI assistant deployment — the state space of home devices is relatively bounded, there's more room for error, and the cost of mistakes (a light left on) is far lower than in general tasks (a flight booked incorrectly).

Deep Control and Contextual Understanding — Stuck in the Middle

What Jen is really waiting for — the "next wave" — is deep integration with devices. Right now, Alexa Plus mostly operates at an on/off level: it can set the office lights to purple or a warm sunset tone, but it can't switch the washing machine to "delicate cycle" because the integration simply doesn't go deep enough.

On contextual understanding, Alexa Plus has made real progress. Users can send it a daughter's class schedule and have it automatically added to the calendar, or upload a recipe to a cooking library. Ring camera integration is a standout example: if you miss a doorbell alert, you can ask "when was my package delivered?" and the system will pull up the most recent delivery video — and if it came from Amazon, it'll even tell you what's in the package. Jen's verdict was characteristically blunt: "Creepy, yes. Useful, also yes." That's the eternal balancing act of smart home AI.

David raised a real-world pain point, though: with three kids, laundry is endless, and he still can't get the system to remind him when the washing machine finishes a cycle. In theory, an Echo could listen for the alert tone — but in practice, all these small, long-tail needs remain stubbornly hard to fulfill. "As someone who lives in a house, these little things are going to take a very, very long time to work out."

Step Outside the House and It Falls Apart: Amazon's Walled Garden Problem

The moment a task goes beyond the smart home, Alexa Plus's performance drops sharply. Jen tried using it to order food on Uber Eats, book travel, and buy tickets — and kept running into dead ends, eventually falling back on her phone. "It would've been faster to just use my phone from the start."

Many third-party integrations promised at launch — like booking hotels through Expedia — still haven't materialized. The shopping use case is even more telling: Jen built a grocery list in Alexa, then wanted to order from her usual store, Harris Teeter — only to find she had to transfer each item manually. Amazon will only place the order automatically if you use Whole Foods. "Amazon doesn't acknowledge the existence of other grocery stores. It's Whole Foods or nothing."

Jen issued a pointed warning: if every company locks its AI into its own walled garden, the experience will be terrible for everyone. "This is all just beginning, and everything has to be more open than it's ever been." She believes Alexa was originally the most open smart home platform around, and Amazon should reclaim that DNA.

The concept of the "walled garden" comes from platform economics — it refers to tech companies controlling the entry and exit points of an ecosystem to lock users into their own products and services while raising barriers for third parties. Amazon's predicament here is structural: its core business interests lie in driving users to consume within the Amazon ecosystem (Whole Foods, Amazon's own products), so there's little incentive to integrate competing third parties like Uber Eats or Harris Teeter — and there may even be a business logic to actively suppressing them. This is the same category of anti-competitive concern as Apple's App Store restrictions on competing payment systems, or Google Search's preferential treatment of its own services. For AI assistants, the contradiction is especially acute — the assistant's value comes from "helping users accomplish any task," while the platform's interest lies in "keeping users inside its own ecosystem." These two goals are fundamentally at odds. This is also why the industry generally approaches the idea of "general-purpose AI agents" with caution: when an agent's backend is controlled by a party with its own interests, the credibility of its "general-purpose" promise is inherently questionable.

The $500 Alexa Tablet: Laying the Groundwork for "Personal Context"

Amazon has been on a hardware release spree lately — new Ring devices, new Kindles, and the most surprising entry: a $500 Alexa tablet. It's not a locked-down Fire tablet but a full Android tablet, going head-to-head with the iPad.

In Jen's view, this device — led by hardware enthusiast Panos — is Amazon's attempt to fill in the personal context data gap. Alexa Plus can reportedly read screen content with the user's permission and embed itself deeply at the OS level — a level of ambition, similar to Siri or Gemini, that Amazon has never attempted before. But she's skeptical: tablets tend to be companion devices to computers, rarely serving as someone's primary work interface. In the end, it might be better suited as a kids' tablet than as the platform for a full personal assistant vision.

David offered a reminder, though: Amazon was ahead of its time on rings, glasses, and several other hardware categories. Now that genuine market interest has arrived and the technology has matured, those dusty experiments could very well make a comeback.

"Personal context data" is the central battleground of the current AI assistant arms race. Personal context refers to the collection of user behavior data an AI can access — calendars, email, location history, purchase records, app usage patterns, and more. Without this data, an AI assistant can only respond to explicit commands in the moment. With it, an assistant can proactively anticipate needs — for example, automatically alerting users about traffic before they leave, or suggesting a restaurant reservation based on their calendar. Google's Gemini is deeply embedded in Android and Gmail; Apple's Apple Intelligence runs through the iOS system layer. Both are building moats around this contextual data. Amazon has long reached users primarily through Echo devices in the home, with almost no visibility into what users do beyond the front door. If this Android tablet can let Alexa read screen content and embed itself at the OS level, it's a critical step toward closing that data gap — but it also implies a deeper privacy trade-off, which connects directly to the trust crisis discussed at the end of this article. The "Panos" mentioned is Panos Panay, the hardware designer who led Microsoft's Surface product line before joining Amazon in 2023 to head its devices division.

Apple Is About to Enter the Field — Approaching the Same Problem from the Opposite Direction

The episode also looked ahead at Apple's rumored smart home launch (reportedly possibly on the 13th). Leaked images surfaced of smart home devices made by LG with Apple branding — resembling a Nest thermostat in appearance.

Jen's primary focus is on family-facing Siri AI. She argues that smart displays are old news — the real differentiator is "the interface as AI." Not just a touchscreen, but Apple Intelligence deeply embedded in the home, genuinely understanding and controlling all devices rather than just toggling them on and off. Apple's recent addition of scene-description features for HomeKit Secure Video, and Bloomberg's reporting on a new camera that "captures data without recording," are both pointing in this direction.

An interesting contrast emerges: Amazon started from the smart home and is struggling to expand toward a personal assistant; Apple is starting from personal devices and trying to push into the home. Multi-user environments and the risk of agentic AI acting first and asking questions later both make the home a far more complex space than a smartphone.

Technology Is Maturing — But Trust Is Eroding

The two eventually landed on the most delicate tension in the entire industry: privacy and trust. Apple must walk a tightrope between "useful features," "protecting privacy," and "but there's a camera" — extraordinarily difficult to pull off. Jen used Gmail as an analogy: people accepted it because it genuinely worked, and privacy was a trade-off worth making. Personally, she leans toward a future where smart homes run entirely locally, even if the upfront costs are higher and local AI capabilities haven't yet caught up with the cloud.

David's reflection closed out the episode: "Everything we've been waiting ten years for is becoming real right as the technology actually starts to work — and people are starting to realize that maybe they never actually wanted any of this." Even Apple, long the most trusted name in tech, is seeing that trust eroded. The widening gap between technological maturity and user willingness is arguably the most thought-provoking question in the tech industry right now.

On-device / local processing refers to AI inference tasks that run entirely on the user's device, without uploading data to cloud servers. The privacy advantages are obvious: data never leaves the device, and in theory cannot be collected by the manufacturer or exposed in a server breach. Apple's Apple Intelligence has partially achieved this architecture, deploying smaller models locally on iPhone and only invoking the cloud when necessary via "Private Cloud Compute" — with a commitment that cloud-side processing doesn't retain user data. However, local processing faces two real-world bottlenecks. First, compute limitations: consumer-grade chips can run far smaller models than cloud-side superscale systems, with a correspondingly lower capability ceiling. Second, cost: on-device AI chips drive up hardware prices, which is why high-performance local AI devices tend to be significantly more expensive. Jen's comment that "the upfront investment is higher and local AI capabilities haven't yet caught up with the cloud" is an accurate description of these constraints. As chip performance continues to improve — Apple Silicon, Qualcomm Snapdragon X series, and others — the capability ceiling for local AI is rising fast, but compared to cloud models at the scale of GPT-4, the gap will persist in the near term.