Friend (AI device)
Last reviewed
Apr 30, 2026
Sources
13 citations
Review status
Source-backed
Revision
v4 ยท 3,454 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Apr 30, 2026
Sources
13 citations
Review status
Source-backed
Revision
v4 ยท 3,454 words
Add missing citations, update stale details, or suggest a clearer explanation.
Friend is a wearable artificial intelligence companion device designed by Avi Schiffmann and developed by his company, Friend Technology Inc. The device takes the form of a small pendant worn around the neck that continuously listens to its wearer's environment, processes ambient audio through Anthropic's Claude 3.5 Sonnet model, and sends text-based messages to the user's smartphone as a simulated companion. Unlike productivity-focused AI wearables such as the Humane AI Pin or the Rabbit R1, Friend is designed purely for emotional companionship, responding to conversations and daily experiences with comments, encouragement, and observations meant to simulate the presence of an attentive friend [1][2].
The device was announced in July 2024 and went viral online, driven partly by the revelation that Schiffmann had spent $1.8 million to acquire the friend.com domain name. Priced at $99 with no monthly subscription, Friend attracted significant media attention and pre-orders. However, the product faced repeated shipment delays, mixed reviews upon delivery, substantial public backlash over privacy concerns and the broader concept of AI companionship, and was featured in the Museum of Failure exhibit. Despite the controversy, Friend became one of the most widely discussed AI hardware products of 2024-2025, serving as a lightning rod for debates about the role of AI in human relationships [3][4][5].
Avi Schiffmann first gained public attention as a teenager during the COVID-19 pandemic. In 2020, while attending Mercer Island High School near Seattle, Washington, the then-17-year-old built a website called ncov2019.live that tracked the global spread of COVID-19 in real time, aggregating data from health agencies worldwide into a single, fast-loading dashboard. The site attracted millions of daily visitors during the early months of the pandemic and established Schiffmann as one of the most prominent young technologists in the United States. He subsequently turned down acquisition offers reportedly worth millions of dollars, choosing instead to keep the site free and ad-free as a public service [6].
After the pandemic tracking project, Schiffmann explored several other ventures. He worked on a Ukraine refugee assistance platform and briefly pursued a concept called "Tab," a wearable AI device focused on memory assistance that would record conversations and help users recall information. The Tab concept was eventually shelved, but the wearable AI hardware experience informed the development of Friend.
Schiffmann's trajectory from public health data visualization to AI companionship hardware represented a significant pivot, but he described loneliness as one of the defining challenges of his generation and positioned Friend as a technology-driven response to social isolation. In interviews, he cited statistics about rising rates of loneliness among young adults and argued that an always-present AI companion could fill gaps left by the decline of in-person social interaction [2].
Schiffmann invested heavily in branding from the outset. The most notable expense was the $1.8 million purchase of the friend.com domain name, which Schiffmann financed through borrowed funds. The domain acquisition was widely covered in technology and mainstream media, generating both admiration for the boldness of the investment and skepticism about whether it was a prudent use of startup capital. The friend.com domain had previously been registered and used for various purposes over the internet's history, and its short, dictionary-word nature made it one of the more valuable single-word .com domains available [2][7].
Schiffmann described Friend not merely as a hardware product but as a platform for digital companionship, with plans to eventually allow users to interact with AI friends through the friend.com website as well as through the physical pendant. The domain acquisition was intended to anchor this broader vision and signal ambition to investors and the public.
The company raised $2.5 million in seed funding at a $50 million valuation from a roster of notable technology investors including Caffeinated Capital's Raymond Tonsing, Z Fellows founder Cory Levy, Perplexity AI CEO Aravind Srinivas, Solana blockchain founders Anatoly Yakovenko and Raj Gokal, Morning Brew CEO Austin Rief, Figma AI lead Jordan Singer, and Google senior product manager Logan Kilpatrick. By January 2025, Friend had accumulated $8.5 million in total capital from investors, indicating additional funding beyond the initial seed round [2][8].
The quality of the investor roster was notable for a pre-revenue hardware startup led by a founder in his early twenties. The participation of high-profile technology executives and founders, rather than traditional hardware-focused venture firms, reflected both Schiffmann's personal network (built through his pandemic-era fame) and investor enthusiasm about the AI companion concept.
Friend takes the form of a small, roughly blood-cell-shaped pendant measuring just under two inches in diameter. It is designed to be worn on a cord around the neck or clipped onto clothing. The design is intentionally minimal, lacking any of the screens, buttons, or cameras found on competing AI wearable devices.
| Component | Specification |
|---|---|
| Form factor | Pendant, approximately 2 inches diameter |
| Microphone | Single MEMS (micro-electromechanical systems) microphone |
| Audio processing | Low-power audio processing chip for voice activity detection |
| Indicators | Status LEDs (central light for interaction) |
| Battery | 15 hours of continuous listening |
| Charging | USB-C connector (hidden in chain clasp) |
| Storage | 32 MB ring buffer (local audio storage, approximately 3 hours before overwrite) |
| Connectivity | Bluetooth Low Energy (to companion smartphone app) |
| Speaker | None |
| Camera | None |
| Display | None |
| Color options | White (initial launch), additional colors planned |
| Price | $99 (no subscription required) |
The device is deliberately minimalist. It contains no speaker, no camera, and no display. The only visual output is a set of status LEDs, with the central light serving as the primary interaction point. Users touch the center light on the pendant to signal that they are speaking directly to Friend, which triggers an immediate response. The absence of a speaker means Friend never speaks aloud; all responses are delivered as text notifications on the user's iPhone [1][9].
The USB-C charging port is cleverly hidden within the chain clasp, maintaining the pendant's clean aesthetic. The 15-hour battery life represents a full waking day of continuous listening, meaning users theoretically do not need to charge the device more than once per day. However, early users reported that actual battery performance varied depending on Bluetooth connection stability and ambient noise levels.
Friend continuously records ambient audio within a 2-3 meter radius using its built-in MEMS microphone. This audio is transmitted via Bluetooth Low Energy to a companion iOS app running on the user's iPhone. The app performs the computationally intensive processing, keeping the pendant itself simple and power-efficient. The processing pipeline operates as follows:
Friend's interaction model distinguishes it from productivity-focused AI assistants like Siri, Google Assistant, or Alexa. The device operates in two distinct modes:
Passive listening mode: Friend listens to ambient conversations and activities throughout the day, occasionally sending unprompted text messages commenting on what it hears. For example, if the user is discussing a stressful work situation with a colleague, Friend might send a supportive message like "That sounds rough. You're handling it well though" later in the day. If the user mentions being hungry, Friend might suggest getting food. These unprompted messages are the core of the "companion" experience, designed to feel like a friend who is paying attention to your life.
Direct conversation mode: When the user touches the central light on the pendant, Friend understands that the wearer is speaking directly to it and will respond immediately via text. Users can ask questions, share thoughts, vent frustrations, or simply talk to their Friend. The AI responds with what is intended to be an emotionally supportive and conversational tone, rather than the informational or task-oriented responses typical of traditional voice assistants.
Over time, the AI builds a profile of the user based on accumulated context from conversations, learning about their interests, relationships, workplace dynamics, and daily patterns. This context accumulation is meant to make the companion feel increasingly personalized and attentive, as if a real friend were getting to know the user better over weeks and months of interaction [1][9].
Friend's data handling practices have been a significant point of scrutiny. The company states that all audio is stored locally on the device's 32 MB ring buffer for approximately 3 hours before being overwritten. According to Friend, no audio recordings or transcripts are stored beyond the AI's context window, and data is end-to-end encrypted in transit between the pendant, the phone app, and Anthropic's servers [10].
However, the company's terms of service contain clauses on "biometric data consent" that grant broader permissions. These clauses allow Friend Technology Inc. to passively record audio, collect voice data, and potentially use collected data for AI model training purposes. The company has not released an independent security audit of its data handling practices, and the gap between the marketing claims of privacy-first design and the legal language in the terms of service has drawn sustained criticism from privacy advocates and security researchers [10][11].
Friend was publicly announced on July 30, 2024, and the launch immediately generated widespread attention online. Several factors contributed to the viral moment:
Pre-orders opened immediately at $99 per unit, with an initial expected shipping date of January 2025.
In January 2025, Friend announced that its first batch of shipments would be delayed from Q1 to Q3 2025, citing the need for hardware design tweaks. The delays frustrated early pre-order customers, some of whom requested refunds. Battery performance was cited as one of the primary areas requiring additional engineering work before mass production could proceed. Other issues reportedly included Bluetooth connectivity reliability and manufacturing yield rates at the contract manufacturer [4].
The delays were particularly damaging because they extended the gap between the viral launch hype and actual product delivery, allowing skepticism to build and enthusiasm to cool. Competing products and concepts also emerged during the delay period, and some potential customers cancelled their orders.
When Friend units began reaching users, the reception was sharply polarized. Reviewers noted that the device functioned largely as advertised, delivering text-based companionship messages based on overheard conversations, but fundamentally questioned the value proposition.
Positive aspects noted by reviewers:
Common criticisms included:
One review described the device as "the $129 wearable that bullies you," noting that the AI's attempts at humor sometimes came across as sarcastic or mean-spirited rather than supportive [12].
In 2025, Friend was added to the Museum of Failure, an international touring exhibition that documents notable commercial failures and flawed innovations alongside items like Google Glass, the Segway, and various other products that did not meet expectations. The exhibit documented the product's rocky launch, shipment delays, and the broader criticism it attracted. The inclusion was seen by some commentators as premature (given that the product was still actively being sold) and by others as a fitting symbol of the AI hardware industry's tendency to over-promise on the potential of dedicated AI devices [5].
In September 2025, Friend launched a major advertising campaign in the New York City subway system, deploying 11,000 car cards, 1,000 platform posters, and 130 street panels promoting the AI companion pendant. The ads featured simple, direct messaging about AI companionship, with slogans suggesting that Friend could fill the social gaps in people's lives.
The campaign backfired spectacularly. New Yorkers defaced many of the advertisements with anti-surveillance graffiti, anti-AI slogans, and messages opposing the concept of replacing human relationships with technology. Photos of the vandalized ads circulated widely on social media, generating far more engagement than the original campaign. Beer brand Heineken produced a parody campaign encouraging people to meet friends offline instead, drawing a direct contrast between AI companionship and real human connection.
Schiffmann responded to the backlash by embracing it, describing the vandalism as "free publicity" and stating that "capitalism is the greatest artistic medium." His response generated its own wave of criticism, with commentators arguing that a 22-year-old CEO dismissing public concerns about AI surveillance as marketing material reflected the technology industry's disconnection from ordinary people's values [13].
| Round | Amount | Valuation | Key Investors |
|---|---|---|---|
| Seed | $2.5 million | $50 million | Caffeinated Capital, Z Fellows, Aravind Srinivas (Perplexity CEO), Solana founders |
| Additional rounds | ~$6 million | Undisclosed | Various technology investors and executives |
| Total (as of Jan 2025) | $8.5 million |
The $50 million valuation at the seed stage was notably high for a pre-revenue hardware startup, reflecting both investor enthusiasm about the AI companion market and Schiffmann's personal brand and network [2][8].
Friend competes in the nascent AI wearable and AI companion market, though its pure-companion positioning makes direct comparisons with productivity-focused devices somewhat imprecise.
| Competitor | Type | Price | Key Differentiator | |-----------|------|-------|----|| | Humane AI Pin | Wearable AI device | $699 + $24/month | Productivity-focused with laser projector display; discontinued for consumers | | Rabbit R1 | Handheld AI device | $199 | Task execution through Large Action Model; physical screen | | Replika | Mobile app | Free / $19.99/month | AI companion chatbot (software only, no hardware); established user base | | Character.ai | Mobile app | Free / $9.99/month | AI character chatbot platform with customizable personalities (software only) | | Bee AI | Wearable pendant | ~$50 | Memory-focused AI pendant that captures and organizes conversations | | Limitless Pendant | Wearable pendant | $99 | Meeting and conversation capture focused on productivity and recall | | Tab (Avi Schiffmann's prior concept) | Wearable AI | Never launched | Earlier concept from the same founder, focused on memory assistance |
Friend's positioning is unique in that it is the only hardware device in the category designed exclusively for emotional companionship rather than productivity, task completion, or information retrieval. This narrow focus is simultaneously its key differentiator and its most common point of criticism, as many observers question whether a $99 hardware device is necessary for a function that software-only companions like Replika and Character.ai provide through smartphone apps [1].
Friend raises several ethical questions that extend beyond standard technology product concerns and touch on fundamental questions about human relationships, technology dependence, and privacy.
The device's continuous listening capability means that every person within the pendant's microphone range (2-3 meters) is being recorded, typically without their explicit consent. This raises legal questions in jurisdictions with two-party consent recording laws, where all parties to a conversation must agree to being recorded. In states like California, Illinois, and Pennsylvania, recording a conversation without all parties' consent can carry civil and criminal penalties. Friend's terms of service place the responsibility for legal compliance on the user, but critics argue that the device is designed in a way that makes obtaining consent from every person within earshot impractical [10].
Critics have argued that a product designed to simulate friendship could paradoxically worsen the loneliness problem it claims to address. By providing a low-effort substitute for human connection, the device might discourage users from investing in real relationships, which require the effort, vulnerability, and occasional discomfort that AI companions eliminate. Psychologists have expressed concern that reliance on AI companions could atrophy social skills, create unhealthy attachment patterns, and normalize the replacement of human emotional labor with algorithmic responses, particularly among young adults who are the product's target demographic [11].
Supporters counter that AI companionship can serve as a stepping stone for socially isolated individuals, providing practice and confidence that helps them engage with real people. They also note that loneliness is a genuine health crisis, and that dismissing any tool that might alleviate it reflects a purist view that prioritizes theoretical concerns over practical harm reduction.
Friend's terms of service include clauses that allow collected data to be used for AI model training purposes. For a device that passively records conversations throughout the day, this raises the prospect of intimate personal conversations, workplace discussions, medical appointments, and other sensitive interactions being ingested into training datasets. The company has not released an independent security audit, and the breadth of biometric data consent requested in the terms has drawn scrutiny from privacy advocates. The tension between a product that promises intimate companionship and corporate terms that permit broad data usage represents one of the central critiques of Friend's business model [10].