We ask Gemini and Alexa to track cats and give advice

Episode Summary

Executive Summary: The episode reviews Google’s Gemini for Home and Amazon’s Alexa Plus, arguing that generative AI has improved conversational flexibility but made smart-home control slower, less predictable, and sometimes worse. The hosts conclude that no company has yet solved the core problem: a reliable, useful, privacy-respecting ambient assistant that can safely orchestrate home devices, calendars, and information without becoming noisy surveillance or an upsell machine.

Main Topics: Gemini for Home vs. Alexa Plus (Priority: 5/5): The hosts compare Google’s and Amazon’s new AI voice assistants, finding Alexa Plus more conversational and integrated, while Gemini for Home remains slower and more fragmented, especially for smart-home control. Speed, reliability, and determinism (Priority: 5/5): A central critique is that generative AI introduces latency and unpredictability into tasks that need to be fast and exact, like turning on lights, setting timers, or controlling scenes. User experience and product design friction (Priority: 4/5): The discussion highlights confusing naming, wake words, separate modes (Gemini for Home vs. Gemini Live), and the need to use awkward phrasing that undercuts the promise of natural language. Monetization and ads degrading the experience (Priority: 4/5): Alexa Plus is portrayed as increasingly commercial, with constant product-selling behavior and ads on Echo Show devices, while Google’s future monetization is still unclear. What smart homes should become (Priority: 4/5): The hosts debate whether the future is one all-knowing assistant or a split system where home orchestration, personal assistant, and camera intelligence are separate layers. Privacy, surveillance, and ambient sensing (Priority: 5/5): The conversation warns that camera-driven smart homes could create a surveillance nightmare if homes become constantly aware of movement, location, and behavior across devices. Limits of Apple and OpenAI entering the space (Priority: 3/5): They argue that Apple and OpenAI face major ecosystem, platform, and integration hurdles, making it unlikely they can quickly deliver a truly comprehensive smart-home assistant.

Key Arguments: Generative AI makes assistants more conversational, but smart-home use cases need deterministic behavior, not creative variability. Gemini for Home can handle some chained commands better than before, but it is notably slower and less reliable than the classic Google Assistant. Alexa Plus is better at conversational back-and-forth and calendar summaries, but it frequently tries to sell products and inject commerce into the experience. The core promise of “natural language” is undermined when users still must speak in stilted, exact phrases to get basic actions done. Smart-home assistants need to integrate with many different systems and permissions, which makes a single all-powerful assistant technically and politically hard to build. Camera-based ambient awareness may improve context, but it also raises serious privacy and surveillance concerns. Apple may have broader HomeKit adoption, but Siri still struggles with basic smart-home commands and account/context handling. OpenAI may be strong at agents and reasoning, but building a full smart-home platform from scratch would require massive infrastructure and ecosystem work. The current value proposition is weak because the upgraded assistants are slower, pricier, and not yet clearly more useful than the old versions.

Data Points: Testing duration: about a year - Alexa Plus had been tested by one speaker for roughly a year, long enough to see major changes and ongoing issues. Podcast date: Thursday, August 20th, 2026 - The episode begins with the Vergecast news roundup dated August 20, 2026. Image/video leaks: 3 clips - The Verge’s news segment mentions three clips of apparent Grand Theft Auto VI footage leaking online. DJI import ban workaround: $659 - A Verge reporter bought an Osmo Pocket 4 from Temu for $659 despite the FCC import ban context. DJI workaround delivery time: 4 days - The banned gadget was reportedly delivered to Oregon in four days. AI tool count at company: 67th AI tool - A sponsor read jokes that a team had added its 67th AI tool and 67th security blind spot. Subscription price: $20 a month - The hosts note that the new smart-assistant tiers require paid subscriptions, including a $20/month model. Alexa shopping behavior: every single Echo Show screen - One speaker complains that ads now appear on every Echo Show screen in her home. Three-timer test: 3 timers at once - Both assistants successfully set three timers in one example, showing progress on multi-command chaining. Timer durations: 25, 7, 12, and 30 minutes - Example command: potatoes for 25, steak for 7, frozen pizza for 12, and a wine reminder in 30 minutes. Calendar events listed: 4 events - Both assistants summarize a day with four scheduled items, though work calendars remain unsupported. HomeKit camera recognition example: 3 cats - Google reportedly cannot reliably distinguish among the speaker’s three cats.

Pivotal Quotes: "When I ask my smart home to turn on the lights three times, I want it to do that three times." — Jennifer Patterson-Tui: Explaining why generative AI’s non-determinism is a bad fit for smart-home commands that must be exact and repeatable. "Alexa is always trying to sell me things." — Jennifer Patterson-Tui: Describing how Amazon’s monetization model bleeds into the smart-home experience and makes the assistant feel commercially invasive. "I just don't know that anyone's ever really going to be able to do it." — Nili: Summarizing skepticism that any company can build a fully reliable, all-purpose home assistant that handles everything well.

Implications: Smart-home AI is improving at conversation, but usefulness still lags because homes need speed, consistency, and privacy. The likely future is specialized tools and camera-driven context, not one perfect assistant.