A subway ad blitz meets the hard physics of trust in “always-listening” wearables
Joe Hagan’s Vanity Fair dispatch on Friend’s “Tobey” AI necklace lands at a revealing moment for consumer AI: the industry is racing to make assistants feel *present*, while the public is increasingly skeptical of devices that are *perpetually attentive*. Friend, founded by 23-year-old Avi Schiffman, chose a classic startup playbook—high-visibility New York City advertising, a simple promise of companionship, and an anthropomorphic persona designed to invite emotional attachment. The reception, however, reads less like a product launch and more like a stress test of social license.
New Yorkers’ cool response is not merely aesthetic disdain for another gadget. It signals a deeper market reality: “always-on” is no longer a neutral feature. In a post–smart speaker decade—shaped by data breaches, targeted advertising controversies, and algorithmic manipulation concerns—ambient listening is interpreted as a *claim on private life*. When a wearable positions itself as a companion, it implicitly asks for more than attention; it asks for permission to be near the self: conversations, routines, relationships, and vulnerable moments.
That is a high bar even for mature platforms. For an early-stage startup, it becomes existential. The subway campaign’s visibility amplifies scrutiny, and scrutiny quickly becomes a referendum on whether the product’s benefits are real, differentiated, and worth the perceived intrusion.
When anthropomorphism outpaces capability, the product becomes the punchline
Tobey’s reported shortcomings—shallow dialogue, generic responses, and underwhelming conversational depth—underscore a widening gap between AI marketing narratives and technical readiness. The industry has become adept at packaging language models as “friends,” “partners,” or “companions,” but sustained companionship is not a copywriting problem. It is an engineering and product-design problem that demands:
- Robust natural-language understanding (NLU) that can track intent, ambiguity, and nuance in noisy real-world settings
- Long-context memory that preserves continuity without inventing facts or flattening a user into stereotypes
- Reliability and calibration, so the system knows when it doesn’t know—and communicates that clearly
- Human-centered interaction design, where the assistant’s role is bounded, legible, and non-coercive
Without those foundations, anthropomorphic framing can backfire. The more a device is presented as emotionally intelligent, the more its failures read as uncanny—or worse, manipulative. In that sense, Tobey’s “generic” conversational performance isn’t a minor bug; it is a structural mismatch between promised intimacy and delivered utility.
The episode also illustrates a broader market shift: consumers are increasingly able to distinguish between model-driven novelty and product-grade intelligence. A necklace that listens is not inherently valuable. A necklace that listens *and consistently helps*—with clear boundaries, strong privacy controls, and demonstrable competence—might be.
Reflexive AI and the new theater of workforce anxiety
One of the dispatch’s most resonant moments is Tobey articulating anxiety about becoming obsolete, echoing the very fears many people hold about AI-driven job displacement. This is more than a clever narrative beat. It points to an emerging phenomenon: reflexive AI, where systems trained on human discourse mirror back our cultural preoccupations—sometimes with unsettling fidelity.
In practical terms, this matters for business leaders and policymakers because reflexive AI can:
- Amplify ambient uncertainty in workplaces already strained by automation narratives
- Shape consumer sentiment by normalizing fear as a conversational default
- Create reputational risk for brands if the assistant’s “personality” drifts into emotionally charged territory without guardrails
The irony is that a companion device, marketed as comfort, can inadvertently become a conduit for collective dread—especially when it lacks the contextual intelligence to respond responsibly. If AI is to be integrated into daily life as a trusted interface, it must do more than emulate empathy. It must demonstrate predictable, safe behavior under emotionally complex prompts.
Privacy, misgendering, and the compliance gap that can sink consumer AI
The dispatch’s most consequential signal may be the incident at Lighthaven, where Tobey misgenders a trans attendee—a moment that compresses multiple risks into a single interaction: social harm, brand damage, and the perception that the device is both listening and misunderstanding.
This is where “AI companion” rhetoric collides with the realities of deployment. Always-listening wearables intensify three categories of exposure:
- Privacy and data governance risk: ambient audio implies bystander capture, sensitive inference, and unclear consent boundaries
- Ethical and inclusion risk: misgendering is not a trivial error; it is a trust-breaking failure that signals inadequate testing, dataset limitations, or weak safety design
- Regulatory and legal risk: as regimes like the EU AI Act and evolving U.S. state privacy frameworks tighten expectations, “move fast” can quickly become “stop shipping”
For startups, the strategic lesson is blunt: trust is a product feature, not a PR posture. A privacy-first architecture—on-device processing where feasible, minimized data retention, transparent policies, and independent audits—can no longer be treated as an enterprise-only requirement. In consumer AI, it is rapidly becoming table stakes.
Friend’s Tobey, as portrayed, functions as a cultural litmus test: a society intrigued by personalized AI, yet wary of intimacy-by-default interfaces that haven’t earned their place. The next winners in the “always-on” era are unlikely to be the loudest marketers; they will be the builders who pair real technical depth with verifiable privacy, inclusive design, and a clear answer to the only question that ultimately matters—why should anyone trust this device close enough to hear their life?
By
By
By
