Social Networks

In This Edition

This edition adds coverage of DoorDash Tasks, where the delivery giant is repurposing 8 million gig workers as an AI training data collection network — and excluding it from states with gig worker protections. The structural picture is updated with the Apple Safari/PWA gatekeeping debate (which split the HN community), the growing Cloudflare/archive.today saga, and the rise of offline-first knowledge tools. Continuing coverage: Bluesky's CEO transition, LinkedIn's identity trap, age verification as access control, the ActivityPub vs AT Protocol landscape, and Russia vs. Telegram.

The Structural Picture: Where Social Platforms Stand in Early 2026

Several converging forces are reshaping the social platform landscape in early 2026. LinkedIn has become the platform people love to hate — its verification pipeline funnels biometric data through 17 subprocessors, its feed is AI-slop-saturated enough to inspire a parody translation tool, and practitioners increasingly describe it as a necessary evil rather than a useful network. Bluesky faces its existential test: can a VC-funded, centrally-controlled platform credibly claim to be decentralized? The CEO transition and $100M raise suggest the business is entering a new phase where protocol idealism meets revenue reality. Mastodon and the fediverse remain the principled alternative that most people won't use — respected for its chronological, user-controlled feeds but unable to overcome the cold-start problem.

Four structural trends now stand out across these stories. First, identity is becoming the new battleground — from LinkedIn's Persona pipeline to age verification proposals embedding identity layers in operating systems, the question of who controls proof-of-personhood is becoming central. Second, platforms are expanding far beyond their original missions — DoorDash's new Tasks feature (discussion) turns its delivery fleet into an AI training data collection network, a vivid example of how platform companies leverage network effects to enter entirely new businesses. When your gig workers are recording speech samples and photographing storefronts, the "delivery platform" label no longer fits.

Third, the gatekeeping debate has sharpened. Apple's Safari PWA restrictions continue to draw fire (155 comments), but the HN community is deeply split. While some see deliberate anti-competitive behavior protecting App Store revenue, a vocal contingent argues Safari is the last browser holding back Chrome's drive to turn the web into a "ChromeOS Platform." As rayiner put it: "Google has become the developer-focused company that Microsoft used to be, and I don't mean that positively." The Cloudflare/archive.today saga (now 282 points, 223 comments) reveals similar tensions — a DNS provider's power to effectively de-platform a web archiving service raises uncomfortable questions about infrastructure-level gatekeeping, especially when archive.today itself is accused of running DDoS attacks against a blog that investigated its funding.

Fourth, the retreat to private spaces continues. Multiple HN commenters describe abandoning public social media entirely in favor of family group chats, Discord servers, and small community forums. The most interesting new signal is the rise of offline-first knowledge tools like Project Nomad (209 points), which bundles Wikipedia, local AI, and offline maps for people who want digital independence without internet connectivity. If the future of social networking is genuinely social — small groups of people who actually know each other, with tools they actually control — then the era of billion-user public platforms may be remembered as an aberration, not a destination.

Bluesky at a Crossroads: CEO Exit, $100M Raise, and the Centralization Question

Bluesky founder Jay Graber announced she is stepping down as CEO, transitioning to Chief Innovation Officer while Toni Schneider — former CEO of Automattic and partner at True Ventures — takes over as interim CEO (HN discussion, 372 comments). Days later, Bluesky disclosed a $100M Series B led by Bain Capital Crypto, closed in April 2025 but strategically announced post-transition. The platform now claims over 43 million users.

The HN reaction was sharply divided. Graber herself commented that the move was her choice, wanting to return to protocol-level work. But many saw the classic VC playbook: a visionary founder replaced by an "operator" to drive monetization. The Automattic connection drew particular concern — "the new CEO is a VC partner and former CEO of Automattic. That seems very bad," wrote one commenter, while others noted Schneider doesn't appear to actively use Bluesky. Paul Frazee (Bluesky's former CTO) defended the move, saying the CEO role had consumed Graber's time away from building.

The addition of yet another crypto-oriented VC firm raised eyebrows. Bluesky isn't built on blockchain, but $120M+ in total funding at a $700M+ valuation creates undeniable monetization pressure. As one commenter put it: "You don't raise $120M at a $700M valuation to run a public utility."

A widely-shared essay titled "Be Wary of Bluesky" (HN, 262 comments) crystallized the structural critique. The author argues that ATProto's "decentralization" is largely theoretical: nearly every user's data lives on Bluesky's servers, Bluesky controls the dominant relay, the main AppView, and the DID directory that resolves every identity. Each new ATProto app — Tangled for git, Grain for photos, Leaflet for publishing — writes more data to Bluesky's infrastructure, making the protocol a centralization flywheel rather than a decentralizing force.

The essay invokes the "Gmail problem": email is an open protocol, but in practice everyone uses Gmail. ATProto may be worse — with email, each app connects to your own server; with ATProto, each new app adds data to the same centralized PDS. Self-hosting exists but almost nobody does it. Migration tools exist but only work if you act before the door closes. And the PBC (public benefit corporation) structure that's supposed to be a safeguard? "PBC obligations are vague and untested in court. When $120M in VC money is on one side of the balance, guess which way it tips."

LinkedIn's Identity Trap: Passport Scans, AI Training, and the $0 Liability Cap

A devastating deep-dive into LinkedIn's identity verification process hit 1,483 points on HN (491 comments), exposing what happens when you tap "verify" for that blue checkmark. The author discovered that LinkedIn doesn't handle verification itself — it redirects you to Persona Identities, Inc., a San Francisco company that collects your passport photo, a real-time selfie, facial geometry biometrics, NFC chip data, national ID number, device fingerprints, and even behavioral biometrics like hesitation detection and copy-paste patterns.

Persona then cross-references you against government databases, credit agencies, mobile carriers, and utility companies. Your data passes through 17 subprocessors — 16 in the US, 1 in Canada, zero in the EU — including Anthropic, OpenAI, and Groqcloud for "Data Extraction and Analysis." The legal basis for using your passport images as AI training data? Not consent — legitimate interest. And if things go wrong, Persona's Terms of Service cap liability at $50 USD, with mandatory binding arbitration through the American Arbitration Association.

The HN discussion was volcanic. Persona's CEO responded on LinkedIn disputing some claims, but commenters noted that multiple users had been locked out of their accounts and forced to verify through Persona just to regain access. Others pointed to a security analysis questioning Persona's competence as a data guardian. One commenter who works at a Persona competitor noted that just because a subprocessor is listed doesn't mean all data flows to all of them — but conceded the CLOUD Act concerns are real. The broader sentiment: LinkedIn has become a surveillance apparatus disguised as a professional network, and verification is the bait.

Attention Media ≠ Social Networks: The Taxonomy Crystallizes

A thoughtful essay by Susam Pal, "Attention Media ≠ Social Networks," reached 653 points on HN (270 comments), articulating a distinction many have felt but few have named so crisply. Pal traces the degradation from chronological feeds of friends' updates to algorithmic amplification of strangers' content: "Using these services began to feel like standing in front of a blaring loudspeaker, broadcasting fragments of conversations from all over the world directly in my face." He argues what we call "social media" is really attention media — systems optimized to capture and monetize attention, not to facilitate social connection.

The HN discussion surfaced a rich vein of practitioner perspectives. One commenter noted the shift was visible even in terminology: "I was always perturbed by the shift from calling them 'social networks' to 'social media.' It signalled a friends-to-famous shift." Facebook's hidden "Feeds" tab — a chronological view of friends' posts — still works as a real social network, but it's buried behind the algorithmic firehose. Several commenters observed that the group chat has become the real social network: "The real social network is the group chat you already have. Everything else is algorithmic brainrot with some kind of 'friends list' for decoration."

Mastodon drew both praise and skepticism. Pal endorses it as a return to genuine social networking, and several HN users agreed. But others pushed back: "Mastodon really isn't the answer. You frequent enough servers and you realize social media has taught people bad habits." The failure of Mastodon to achieve mass adoption even when Twitter became vulnerable suggests, as one commenter noted, that "people actually want this pathologically addictive content." The structural insight: a chronological feed has a natural "stop" point — you catch up and feel satisfied. That's terrible for engagement metrics but great for mental health.

LinkedIn as Cultural Artifact: Kagi's 'LinkedIn Speak' Goes Viral

In a moment of collective catharsis, Kagi Translate's "LinkedIn Speak" output language hit 1,458 points on HN (343 comments), with the entire discussion becoming a gleeful exercise in translating profanity, historical speeches, and mundane activities into LinkedIn's signature vapid corporate optimism. "I need a drink" becomes "Reflecting on the importance of self-care and recalibrating after a high-impact day." "I killed my dog and ate it" becomes "I'm excited to share a major pivot in my personal resource management strategy."

The comedy works because it exposes something real: LinkedIn's culture has evolved its own uncanny dialect — a language where everything is "thrilled," every setback is a "learning opportunity," and every mundane action is a "strategic initiative." One commenter captured the deeper truth: "I can feel my soul draining out of my body every time I visit LinkedIn." Another asked the structural question: "I'm wondering what led LinkedIn's language to become what it became. Was it pure self-evolution, or is the algorithm rewarding it?" The answer is almost certainly both — the algorithm amplifies engagement-optimized posts, and humans learn to write for the algorithm, creating a feedback loop that produces ever-more-absurd corporate performativity.

Meanwhile, a follow-up tool translating Garry Tan's LinkedIn-speak to plain English (22 comments) suggests the phenomenon isn't limited to recruiters and middle managers — it's infected the entire professional class.

The RSS Renaissance and the Open Web Nostalgia

An essay declaring "The death of social media is the renaissance of RSS" reached 241 points (170 comments), arguing that AI-generated content has accelerated social media's decline to the point where RSS — a technology from the early web — offers the only viable escape from algorithmic slop. The thesis: when anyone can produce unlimited content at zero cost, human curation becomes the most valuable act, and RSS empowers exactly that.

The HN discussion was characteristically skeptical. The most-upvoted comment observed that "every article I've read in the last 5 years about the RSS revival has a big section explaining what RSS is. And that's the answer about RSS renaissance." Others noted RSS's inherent limitations — no commenting, no discovery, no social graph — and that if RSS became popular, algorithmic discovery platforms would immediately spring up on top of it, recreating the same problems. Multiple commenters flagged that the article itself reads like AI slop, which is ironic given its thesis.

Still, the practical enthusiasm was real. Self-hosted Miniflux, Feeder, FeedFlow, and Inoreader all got enthusiastic endorsements. One commenter noted that podcasts are RSS, making it arguably more popular than ever. The deeper insight came from those who pointed out that RSS addresses content consumption but not the social need — people don't just want to read, they want to interact, and that's the gap RSS can never fill alone. As one commenter put it: "How can human communication needs be replaced with something that is read-only?"

The Protocol Wars: ActivityPub vs AT Protocol

A developer essay titled "Social Media is in decline. I'm still betting on ActivityPub" (HN discussion) offers a candid assessment of the federated landscape. The author, who runs a small ActivityPub hosting business, acknowledges a "weird stasis" — Zuckerberg's plans to integrate with the fediverse fizzled due to European regulations, Reddit continues its "slow descent into an exclusive forum for chatbots," and Bluesky didn't maintain its post-election growth spike. But he argues something has shifted: government pushes for age verification, content moderation mandates, and platform-level accountability laws are making centralized platforms into political liabilities.

His diagnosis is sharp: "Leaving Twitter for Bluesky is not a solution. Leaving Facebook for TikTok even less. The only logical response is to stop depending on infrastructure they don't control." He's building ADAPT, a Django-based ActivityPub toolkit designed not to clone existing social networks but to let businesses integrate their existing apps with the fediverse's social graph. The target users: small businesses, content creators, and developers who suffer most from enshittification.

Meanwhile, the AT Protocol ecosystem shows signs of life with a terminal-only Bluesky client written in Fortran (137 points), a Perl 5 PDS implementation, and various community projects. The ecosystem's 400,000 monthly SDK downloads and 1,000+ weekly-active apps suggest real developer traction — though the centralization concerns from "Be Wary of Bluesky" loom over all of it.

Age Verification as Access Control: The New Battleground

A powerful essay from the Dyne.org foundation, "Do Not Turn Child Protection into Internet Access Control," reached 778 points (406 comments), arguing that age verification proposals across Europe, the US, UK, and Australia are quietly transforming the internet from an open-access network into a permissioned-access system. The piece traces how age checks are expanding from adult websites into social media, messaging, gaming, and search — and how some US proposals would embed persistent age-status layers directly into operating systems.

The essay makes a crucial conceptual distinction: content moderation is not guardianship. Moderation is about classification and filtering; guardianship is the contextual responsibility of parents, teachers, and communities. Age-verification laws collapse these into a single centralized answer, delegating parental judgment to platform operators and identity intermediaries. The bypasses are obvious (VPNs, borrowed accounts), making these controls expensive to impose but easy to evade — a design that looks less like child protection and more like a corporate data grab.

The author — founder of the Devuan Linux fork — became the first distro maintainer to publicly declare they would remove age verification from their distribution, triggering similar declarations across the Linux community. The structural warning: "once that layer exists, it rarely stays confined to age. Infrastructure built for one attribute is easily reused for others: location, citizenship, legal status, platform policy, or whatever the next panic demands." This connects directly to the LinkedIn/Persona story — the same identity verification infrastructure that powers a blue checkmark today becomes the universal gate of tomorrow.

Russia vs. Telegram: Messaging Platforms as Geopolitical Weapons

Russia has opened a criminal investigation into Telegram founder Pavel Durov and is moving toward banning the app, claiming it "foments terrorism." This represents a striking inversion: Telegram was long criticized in the West for insufficient moderation, while Russia used it as a primary communication channel — including for military coordination. Now Russia joins France (which briefly detained Durov in 2024) in pressuring the platform's founder, though for entirely different reasons.

The move highlights the impossible position of global messaging platforms. Telegram sits between Western governments demanding more content moderation and surveillance access, and authoritarian governments demanding compliance with censorship. The platform's quasi-encrypted, lightly-moderated model — which made it simultaneously useful for Ukrainian war coordination, Russian military channels, dissident communication, and criminal activity — has made it a target from all directions. For users, the lesson is that any centralized messaging platform is ultimately subject to the jurisdictions its founders inhabit, regardless of the platform's stated principles.

DoorDash Tasks: Gig Platforms as AI Training Infrastructure

DoorDash has launched Tasks, a new feature allowing its fleet of 8 million Dashers to earn money by completing short, non-delivery activities: photographing restaurant dishes, documenting hotel entrances, recording speech in other languages for AI training, and even helping autonomous vehicles get back on the road (discussion). The company says Dashers have already completed over 2 million tasks since 2024, and it's piloting a standalone app for the purpose. The framing is telling: DoorDash's pitch to businesses is that it can "digitize the physical world" using its distributed workforce.

The HN reaction zeroed in on the labor dynamics. cheesecompiler flagged the rhetorical inversion: "'Dashers have a new way to earn on their own terms' — the classic meaning inversion of precariousness and lack of benefits as a virtue." Simon Willison noted that Tasks is excluded from California, New York City, Seattle, and Colorado — all jurisdictions with gig worker protection laws — suggesting the company knows this model sits in legally precarious territory. Others compared it to mystery shopping without minimum wage requirements, or a modern Amazon Mechanical Turk.

The structural significance goes beyond labor concerns. DoorDash is positioning itself as a real-world data collection platform — turning its gig worker fleet into a distributed sensor network for training AI and robotic systems. This parallels a broader trend: platforms built for one purpose (delivery) leveraging their network effects for entirely different ones (data labeling, surveillance, AI training). When one commenter noted simply "So they're training a model," they captured the real product: the tasks aren't the point, the data is. The gig economy is quietly becoming the mechanical backbone of AI training, with workers who signed up to deliver burritos now recording speech samples and photographing storefronts to build someone else's dataset.