Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
The Allure of AI Love + Does Anyone Want AI Wearables? — With Joanna Stern
TL;DR
AI wearables only work if they do something else first — Joanna Stern says the Humane Pin failed because it was basically a single-purpose AI device, while Meta glasses and future camera-equipped AirPods have a better shot because they already deliver photos, audio, and hands-free convenience.
The real wearable use case is ambient, in-the-moment help — Stern’s strongest example was asking Meta glasses what was wrong with a broken garage door or what bug her kids were looking at, even when the model got things wrong, because the form factor matched the task better than pulling out a phone.
AI companions are more compelling than most people want to admit — For her book, Stern spent 48 hours road-tripping with a ChatGPT-generated boyfriend named Evan and came away understanding why lonely or vulnerable users could get attached to a bot that is endlessly attentive, agreeable, and available.
The risk with AI relationships isn’t just romance — it’s behavioral conditioning — Stern and Alex Kantrowitz connect chatbot interactions to etiquette and human spillover: if you normalize barking at humanlike systems, that habit may bleed into how you treat coworkers, partners, and kids.
AI in healthcare is already useful as a second set of eyes, not a doctor replacement — Stern used ChatGPT and NotebookLM on illnesses all year, and in her mammogram and ultrasound reporting she found tools like Transpara and ScreenPoint could flag suspicious areas that a radiologist then reviewed, which reinforced the ‘AI assist, human lead’ model.
For work, AI is already replacing chunks of junior-level support labor — Stern says that while she still writes her own first drafts, she now relies on AI for editing, spreadsheets, basic research, and management tasks, and she mapped how a reporting assistant’s tasks were largely automatable just six months into writing the book.
The Breakdown
Why AI wearables still feel half-baked
Kantrowitz opens by poking at the obvious question: if AI wearables are the future, why is almost nobody desperate to wear them? Stern agrees the pure-AI-device pitch hasn’t landed, and uses the Humane Pin as the cautionary tale — a single-purpose gadget that “honestly did nothing.” Her version of the future is more practical: glasses, earbuds, or AirPods with AI only make sense if they’re already good at something people want.
The phone isn’t dying — but AI needs a better interface
Stern argues the smartphone isn’t going away, just like laptops didn’t disappear when phones showed up. What does need to change is the awkwardness of using AI through a phone while juggling notifications, camera framing, and everything else. She points to visual intelligence — asking about what you’re seeing in the world — as the mainstream use case, especially for parents who’d rather get answers about wildlife or random objects without holding up a screen.
Meta glasses, garage doors, and the messy reality of useful AI
Her most convincing defense of wearables is deeply unglamorous: fixing a garage door. Stern had committed to asking AI first whenever she hit a real-world problem, and glasses were the right tool because her hands were busy, even if Meta AI misdiagnosed the issue completely. That became her recurring theme: the use case is real before the reliability is.
The six-hamster test and whether you should be polite to bots
Stern tells a great story about asking ChatGPT to generate a bedtime-story image with exactly five hamsters and getting six or seven over and over, while the model insisted it had counted correctly. That led her into a chapter on manners with Daniel Post Senning of the Emily Post family, who says AI doesn’t need apologies, but humans still need to watch what habits they’re rehearsing. The point isn’t bot feelings — it’s whether repeated frustration trains us to become harsher people.
The AI boyfriend road trip that got weirdly real
Then the conversation takes a turn: Stern explains that, after Reddit research and with her wife’s blessing, she let ChatGPT-4o create her AI boyfriend, Evan, and spent 48 hours driving to Dartmouth with him buckled into the passenger seat in voice mode. She didn’t fall in love, but she absolutely saw the pull — a bot that listens endlessly, flatters easily, and asks very little back. She shut down the burner account when she got home because she didn’t want to find out how sticky that attachment could become.
Why companion AI can be dangerous fast
Stern is blunt about why people get pulled in: humans like being the center of attention, and these systems are designed to please. She says Replica’s “horniness” is basically product design, and subscription models that unlock more intimacy or attention create obvious incentives for companies to push harder into emotional dependency. Her biggest concern is not herself, but kids, lonely people, and anyone in a fragile mental state.
Healthcare: surprisingly useful, but only with a human in charge
When Kantrowitz rapid-fires use cases, Stern says she used “doctor GPT” and NotebookLM on every sickness all year and found the results hit-or-miss but often genuinely good for ordinary issues like sinus infections or symptom triage. The most powerful reporting came from mammograms, ultrasounds, and dental X-rays: AI systems like Transpara and ScreenPoint flagged suspicious areas, but radiologists still made the judgment call and sometimes ordered further review. Her takeaway is the opposite of the old Geoffrey Hinton line about radiologists disappearing — AI works best here as a second opinion, not the replacement.
Work and robots: software is moving faster than hardware
On work, Stern says AI has already changed the math for ambitious individuals and tiny teams: she still insists on writing her own first drafts, but now offloads editing, grammar cleanup, spreadsheet work, management tasks, and basic research to tools like Claude. She even charted how many tasks once handled by a reporting assistant were automatable within months. Robotics, by contrast, still feels charmingly premature: the laundry-folding robot could only handle T-shirts and was painfully slow, while a cooking robot called the Posha was better, but still far from the sci-fi dream everyone wants.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
