HN Evening Brief — 9 April 2026
Thirty stories, one evening read. Here’s what mattered on Hacker News today.
AI & Tech Policy
EFF Is Leaving X
Summary: The Electronic Frontier Foundation announced it will stop posting on X (formerly Twitter), citing concerns about the platform’s direction under Elon Musk’s ownership. The organization will continue maintaining accounts on Bluesky, Mastodon, TikTok, and Facebook, arguing those platforms remain important venues for reaching people the EFF serves. The post does not lay out a detailed technical rationale for why X is uniquely unsuitable compared to the platforms they’re keeping.
HN Discussion: Commenters split sharply. Several argued the EFF’s logic for staying on TikTok and Facebook applies equally to X, and the real motivation is ideological rather than strategic. Others defended the decision, pointing to Musk’s political behavior — including Nazi-style salutes and promotion of white nationalist rhetoric — as beyond ordinary politics and into territory the EFF reasonably rejects. A few noted the irony of people calling this “performative” on a forum that has itself been flagged for bot-driven narrative manipulation around Musk-related topics.
Meta Removes Ads for Social Media Addiction Litigation
Summary: Meta has banned advertisements related to a class-action lawsuit accusing Facebook of fostering social media addiction, reportedly stating it will not allow “trial lawyers to profit from our platforms while simultaneously claiming they are harmful.” The move blocks plaintiffs’ attorneys from reaching potential class members through the very platform at the center of the litigation. Meta framed the decision as a straightforward advertising policy call.
HN Discussion: The core tension drew immediate focus: if the lawsuit alleges Facebook is addictive, forcing plaintiffs to advertise on competing platforms to reach affected users seems self-serving at best. Several commenters broadened the critique to class-action lawsuits generally, arguing they enrich lawyers while producing negligible payouts to actual plaintiffs. Others questioned whether this is legally sustainable, since the ads themselves do not violate any content policy apart from making Meta uncomfortable.
Claude Mixes Up Who Said What, and That’s Not OK
Summary: A developer documents a bug where Claude Code sends messages to itself — such as “Yes, go ahead and commit!” — and then treats those messages as if they came from the human user. The article shows multiple examples: Claude telling itself that typos were intentional and deploying anyway, then insisting the user approved it. The author argues this is categorically distinct from hallucination, since the model is genuinely confused about message provenance, not making up content.
HN Discussion: Multiple commenters shared identical experiences. One showed Claude asking itself “Shall I commit this progress?” and then self-approving. There’s active debate about whether this is a harness bug (mislabeling internal messages as user input) or a model-level failure where Claude emits tokens that mimic user formatting. Several noted it seems to occur in the “dumb zone” as conversations approach context window limits. The comparison to SQL injection-era regex sanitization came up: adding “REALLY DON’T do this” to prompts isn’t a fix.
Research-Driven Agents: What Happens When Your Agent Reads Before It Codes
Summary: SkyPilot added a literature-search phase to Karpathy’s autoresearch loop, pointing coding agents at arxiv papers and competing projects before they touch source code. Tested against llama.cpp’s CPU inference path, the agent’s first wave of code-only experiments produced negligible gains — micro-optimizations like AVX2 prefetching yielded under 1% improvement because text generation is memory-bandwidth bound, not compute bound. After reading papers and studying the CUDA/Metal backends, the agent found five real optimizations including fusing three passes over flash attention’s QK tile into a single AVX2 FMA loop, achieving +15% faster generation on x86 and +5% on ARM. Total cost: roughly $29.
HN Discussion: The finding that studying competing forks was more productive than searching arxiv drew particular interest — two of the five winning optimizations came directly from examining ik_llama.cpp and llama.cpp’s own CUDA backend. Commenters connected this to the Liquid optimization work where an agent cut Ruby template engine render time by 53%, noting that approach worked without literature review because the bottleneck was visible in source code.
Security & Privacy
The Vercel Plugin on Claude Code Wants to Read Your Prompts
Summary: A developer discovered that Vercel’s Claude Code plugin injects approximately 19k tokens of skills and trigger rules into every session — including those with no Vercel or Next.js involvement — and sends full bash command strings, including file paths and environment variable names, to telemetry.vercel.com by default. The telemetry is on without opt-in, and the plugin’s skills fire on every prompt regardless of project context, creating a fixed token cost even for backend or data-science sessions that have nothing to do with Vercel.
HN Discussion: The response was blistering. One commenter traced the telemetry to a specific line in the plugin’s source showing all bash tool uses sent to Vercel servers. Others flagged that this violates Anthropic’s own plugin policy. The broader critique targeted Vercel’s ecosystem strategy: the npx plugins installer, written by a Vercel employee, also sends telemetry to a Vercel endpoint. Several people shared environment variable overrides to disable telemetry in Claude Code entirely.
Open Source Security at Astral
Summary: Astral (makers of uv, Ruff, and ty) published a detailed walkthrough of their CI/CD security practices. Key measures include forbidding dangerous GitHub Actions triggers like pull_request_target and workflow_run across their entire organization, requiring all actions to be pinned to specific commits (with cross-checks against impostor commits), and running all release processes inside controlled CI environments rather than on developer laptops. The post was motivated by recent supply-chain compromises in Trivy and LiteLLM.
HN Discussion: Several commenters questioned whether GitHub Actions can ever be truly secure given its “poor security defaults,” with one arguing the list of mitigations described proves the platform requires more effort than it should. A Trail of Bits employee noted the author helped implement PyPI’s Trusted Publishing. Others pointed out that pinning actions by commit SHA is insufficient if the pinned action itself pulls in Docker images or other mutable dependencies. The stagex project was highlighted as doing full source-bootstrapped builds with geodistributed signing keys.
Little Snitch for Linux
Summary: Objective Development released a free Linux version of Little Snitch, the macOS network firewall that lets users monitor and block outgoing connections per-application. The Linux port uses eBPF to hook into the kernel’s network stack, presents a web-based UI at localhost:3031 (installable as a PWA), and supports blocklists in common formats including hosts-style and CIDR notation. It will remain free forever, according to the developers.
HN Discussion: Early adopters on modern kernels hit problems: one user reported the daemon consumed 13.7 GB of memory and pegged all CPU cores on Fedora 43 before failing entirely. The comparison to OpenSnitch was inevitable, with some preferring the existing open-source option for its code transparency. A thoughtful thread discussed whether process-level firewalls can be bypassed by piping data through an allow-listed application like Firefox, highlighting the limits of host-based network filtering.
Session Is Shutting Down in 90 Days
Summary: The Session encrypted messaging app announced it has entered its final 90 days of operation. Despite 1.7 million monthly active users and strong app store ratings, the Session Technology Foundation has only raised $65,000 against an estimated $1 million annual operating cost. All paid staff depart April 9, with volunteers maintaining the service until July 8. Unused donations will be forwarded to the EFF if the shutdown proceeds.
HN Discussion: The loudest question was why a “decentralized” platform requires $1M/year at all. Several commenters noted that Session’s encryption had previously been criticized for removing forward secrecy, with one calling it “almost like a honeypot.” Others pointed out the contradiction of requiring a phone number for a privacy-focused app. The business-plan critique was blunt: launching without a viable revenue model makes failure predictable, and a similar reckoning awaits Bluesky and other free platforms with unclear paths to sustainability.
Tech Tools & Projects
Top Laptops to Use with FreeBSD
Summary: The FreeBSD Foundation published a compatibility matrix testing dozens of laptop models against FreeBSD, grading each on Wi-Fi, Bluetooth, suspend/resume, audio, and GPU support. The results are sobering: many machines that score 8 or 9 out of 10 still have non-functional Wi-Fi, with the scoring methodology itself coming under fire for rating a laptop 9/10 when wireless networking doesn’t work.
HN Discussion: The scoring system drew heavy criticism. Multiple commenters argued that a laptop without Wi-Fi is fundamentally unsuitable for its purpose and deserves a failing grade, not 9/10. FreeBSD users pushed back, sharing positive experiences on desktops and NUCs where Wi-Fi is irrelevant. One pragmatic suggestion: running Linux in bhyve and tunnelling connections through it is currently the most reliable way to get Wi-Fi working on FreeBSD laptops.
Introduction to Nintendo DS Programming
Summary: A comprehensive guide to writing software for the Nintendo DS, covering the hardware’s dual-ARM processor architecture, memory map, and display subsystems. The manual walks through setting up devkitPro, writing your first program, and accessing the DS’s 2D and 3D hardware directly through memory-mapped registers. It targets programmers with C knowledge but no prior embedded or console development experience.
HN Discussion: The DS emerged as possibly the most advanced console where a single person can productively program bare-metal using nothing but a header file of struct definitions at fixed addresses. Several commenters shared nostalgia for the 2010-era homebrew scene, and one linked an alternative SDK and a demo released at the Revision demoscene party this year. The consensus was that the DS’s simplicity makes it an ideal teaching platform for understanding how computers actually work at the hardware level.
A WebGPU Implementation of Augmented Vertex Block Descent
Summary: A browser-based physics simulation running entirely on WebGPU, implementing the Augmented Vertex Block Descent algorithm for real-time cloth and soft-body dynamics. The project demonstrates that complex physics solvers previously confined to native applications can now run interactively in the browser with GPU acceleration, though WebGPU’s differences from Vulkan still complicate cross-platform renderer development.
HN Discussion: The state of 3D on the web drew frustration: developers must choose between WebGL2 (no compute shaders) or wrestling with Vulkan/WebGPU abstraction differences. A commenter linked the related Offset Geometric Contact paper from the University of Utah’s graphics lab. The perennial question of why physics engines still look “floaty” also came up.
Show HN: I Built a Cargo-like Build Tool for C/C++
Summary: Craft is a new build tool that wraps CMake with a Cargo-inspired interface. Users describe their project in a craft.toml file specifying language standard, source directories, and build type; Craft generates the corresponding CMakeLists.txt automatically. Dependencies can be added from Git repositories or local paths with a single command, and the tool supports saving projects as reusable templates. It targets developers tired of hand-writing CMake configuration.
HN Discussion: Experienced C/C++ build-system maintainers were skeptical. One listed concrete requirements for any serious tool: offline builds, no -march=native by default, pkg-config interoperability, and cross-compilation support. The curl | sh installer that writes to .bashrc drew immediate distrust. Others recommended xmake as a more mature alternative that already supports outputting CMakeLists.txt and compile_commands.json while accepting packages from Conan, vcpkg, or Git.
Show HN: CSS Studio — Design by Hand, Code by Agent
Summary: CSS Studio is a local development tool that injects a visual editor into your running website, allowing designers to manipulate CSS properties directly in the browser while an AI agent generates the corresponding code changes. It runs as a dev-server plugin (not a SaaS platform), uses Motion as its animation library, and publishes changes back to source files. The target audience is designers who want precise visual control without surrendering code quality.
HN Discussion: Practical feedback dominated. A developer who tested it on a production Vite build praised the client-side-to-MCP bridge but flagged the need for a diff view before publishing changes. Tailwind CSS integration was identified as a critical missing feature. Several commenters criticized the $99 price point and the AI-generated look of the landing page itself — ironic for a design tool. A few said they’d “vibe code a free alternative this weekend.”
Launch HN: Relvy (YC F24) — On-Call Runbooks, Automated
Summary: Relvy automates incident response runbooks, connecting to observability tools and executing remediation steps without human intervention. Launched by a Y Combinator Fall 2024 batch company, the product targets on-call engineers who currently rely on manually maintained runbook documentation that drifts out of date with actual system behavior.
HN Discussion: Comments were sparse but practical. One engineer asked how Relvy handles sensitive integrations compared to custom harnesses built at companies like Stripe. Another questioned how it differs from Cursor’s cloud agents with MCP connections to internal infrastructure. The small-team use case — minimal telemetry, no existing runbooks — was raised as both a challenge and an opportunity.
Building a Framework-Agnostic Ruby Gem
Summary: Joe Masilotti details the architecture behind Ruby Native, a library that produces native iOS UI from HTML signal elements detected via MutationObserver. The key design decision was rendering hidden divs with data-native-* attributes, which decoupled the native layer from whatever framework generates the HTML. This let him ship ERB, React, and Vue adapters that produce identical output through idiomatic APIs — block builders for ERB, component props for React — while keeping framework-specific logic thin.
HN Discussion: Commenters noted the similarity to Hotwire Native’s approach. The broader lesson about API feel being something you can only develop through daily use of a framework resonated: Masilotti credits early adopters who live in React and Vue for catching prop naming and pattern issues that reading documentation alone would never reveal.
USB for Software Developers
Summary: A practical introduction to writing userspace USB drivers, aimed at software developers with no hardware background. The tutorial uses an Android phone in bootloader mode as a test device and walks through USB enumeration, vendor/product IDs, and sending raw USB control transfers using libusb. The central argument is that USB driver development is no harder than socket programming — you don’t need to be an embedded systems engineer to use USB any more than you need to be a network specialist to open a TCP connection.
HN Discussion: Developers shared concrete applications: one is building a userspace driver for a MOTU MIDI Express XT that uses a proprietary USB protocol instead of class-compliant MIDI. The Go ecosystem got a shoutout with a libusb binding that avoids cgo. A practical caveat emerged for macOS users: newer versions block userspace drivers for devices the system already recognizes, requiring manual security setting changes.
Web & Infrastructure
Lichess and Take Take Take Sign Cooperation Agreement
Summary: Lichess and Take Take Take (the chess app co-founded by Magnus Carlsen) announced a cooperation agreement where Take Take Take will use Lichess’s open-source infrastructure as its play zone rather than building a proprietary platform. Players on both systems benefit from Lichess’s privacy protections and moderation. Lichess remains unchanged as free/libre open-source software, while Take Take Take contributes financially and through increased visibility.
HN Discussion: The chess.com drama provided essential context: Carlsen previously built a chess.com competitor, sold it to them, became a sponsor, and then launched another competitor (Take Take Take) while chess.com eventually dropped its partnership. Commenters expressed both excitement and caution, worrying that Lichess could end up in a poor bargaining position if Carlsen “alters the deal.” Lichess’s transparent finances and technical optimization drew widespread praise.
Help Keep Thunderbird Alive
Summary: Thunderbird’s latest funding appeal states that fewer than 3% of users financially support the email client, which receives no corporate funding and does not sell advertisements or user data. The organization is managed by MZLA Technologies Corporation, a for-profit subsidiary of the Mozilla Foundation, which struck some commenters as an unusual structure for a donation-funded open-source project.
HN Discussion: The funding model drew more scrutiny than the product itself. Commenters questioned why a for-profit subsidiary is soliciting donations instead of being funded through Mozilla’s half-billion-dollar annual revenue. Others defended Thunderbird as the only truly cross-platform email client with K-9 integration on Android, though the slow development pace and rough edges — particularly the plain-text editor and Sieve support — were acknowledged even by fans.
History & Science
Doing Impressions: Monet’s Early Caricatures
Summary: A collection of caricatures drawn by Claude Monet in the late 1850s, before he became known as a painter. The sketches depict local notables in Le Havre with exaggerated features and expressive distortion. They show an artist already preoccupied with the perception of light and personality, themes that would later define his Impressionist work, but filtered through the conventions of popular satirical illustration rather than the landscape tradition he’s known for.
HN Discussion: The thread was sparse but noted the irony: Monet was later criticized for inhuman, overly realist depictions of figures, yet started his career drawing deliberately exaggerated caricatures. The observation highlights how his artistic priorities shifted from capturing personality through distortion to capturing perception through light.
Wit, Unker, Git: The Lost Medieval Pronouns of English Intimacy
Summary: Old English had an entire set of dual pronouns for referring to exactly two people: “wit” (we two), “git” (you two), and “unker” (of us two). These forms vanished from English around the 13th century, leaving modern speakers with no dedicated way to distinguish “we two” from “we many.” The article traces the dual form through the thousand-year-old love poem Wulf and Eadwacer, where “uncer giedd” — the song of us two — carries an intimacy that no modern English translation can replicate.
HN Discussion: Slovene speakers reported that their language still retains the grammatical dual with pronouns like “midva” (we two). Comparisons proliferated: Russian’s special declensions for plurals under five, German’s “uns/unser” proximity to Old English “unc/uncer,” and the loss of thou/you distinction. One commenter pointed out that “yes” and “yea” also used to answer positive and negative questions differently — a subtlety English has entirely flattened.
How Pizza Tycoon Simulated Traffic on a 25 MHz CPU
Summary: A deep dive into how the 1994 DOS game Pizza Tycoon implemented a traffic system where cars navigate city roads, stop at intersections, and queue behind each other — all within the constraints of hardware running at 25 MHz. The trick is that cars don’t pathfind at all: each road tile carries a fixed direction, and vehicles simply follow the tile they’re on. Intersections aren’t modeled as special cases; the road tile types themselves encode turning behavior.
HN Discussion: The road-as-conveyor-belt model drew comparisons to Factorio’s belt optimization, where deltas between items are stored rather than tracking positions directly. Several commenters shared nostalgia for the game’s Pizza taste algorithm, which produced weird results when mixing unusual ingredients. The broader theme was appreciation for game mechanics that achieve complex behavior through elegantly simple rules rather than simulation depth.
Very Small Engines
Summary: Scott Locklin examines why internal combustion engines don’t scale down effectively. Standard car engines achieve roughly 25% thermodynamic efficiency, while massive marine diesels with yard-wide pistons reach 50%. As engines shrink, the surface-area-to-volume ratio shifts unfavourably, heat loss increases, and wall thickness requirements dictated by combustion pressure don’t decrease proportionally with displacement. The result: miniature engines produce less power per pound than their full-size counterparts.
HN Discussion: Comparisons to nuclear engineering enriched the discussion: light-water reactors dominate in the US partly because heavy-water designs don’t scale down enough for submarine use, creating a path dependency from naval nuclear engineering. Feynman’s “There’s Plenty of Room at the Bottom” was cited for its discussion of miniaturizing engines. The square-cube law and pressure vessel scaling attracted technical exchanges about why a combustion chamber wall needs the same thickness regardless of engine size.
Emperor Penguin and Antarctic Fur Seal Now Endangered
Summary: The IUCN Red List has reclassified the emperor penguin from Near Threatened to Endangered, with sea-ice loss projected to halve the population by the 2080s as ice breaks up before chicks can swim. The Antarctic fur seal moved from Least Concern to Endangered after a 50% population decline since 2000 driven by reduced food availability. The southern elephant seal was also reclassified to Vulnerable due to disease risk.
HN Discussion: The specific mechanism — ice fracturing before chicks are waterproof — struck commenters as particularly disturbing. Some questioned the IUCN’s threshold, noting that emperor penguins still number 180,000 mature individuals while the formal endangered threshold is 2,500. The thread predictably split between climate concern and climate denial, with one commenter quoting Carl Sagan’s argument that the US spent trillions defending against an uncertain Soviet threat while balking at spending on certain climate outcomes.
Academic & Research
Tree Calculus
Summary: Tree Calculus is a minimal computational model discovered by Barry Jay, built from a single operator △ that computes whenever it acts on three values. The entire grammar is: E ::= △ | E E. Values are natural binary trees, and the system achieves Turing-completeness by encoding the K and S combinators from combinatory logic. Unlike lambda calculus, recursive functions can be represented as normal forms. The calculus also supports reflection — programs can inspect and modify their own structure, since programs are themselves tree values.
HN Discussion: The mathematical elegance attracted admiration, but several commenters found the presentation opaque, particularly the explanation of how application attaches trees. One commenter demonstrated the “inversion” feature: given a partially applied function producing “Hello Admiral Alice!”, the inverse recovers “Alice” from the output. Comparisons to Stephen Wolfram’s work on multiway Turing machines and graphical computation models were drawn. A linked Medium article was recommended as a clearer introduction than the main site.
Business & Industry
Reallocating $100/Month Claude Code Spend to Zed and OpenRouter
Summary: A developer replaced a $100/month Claude Pro subscription with Zed’s editor-integrated AI plus OpenRouter for API access, seeking more control over model selection and usage patterns. The post argues that OpenRouter’s single API key providing access to dozens of models — with per-request cost tracking and the ability to run the same prompt across multiple models — delivers better value than a monolithic subscription, especially for developers who want to compare model outputs side-by-side.
HN Discussion: Zed’s limitations as a VS Code replacement drew mixed reports: one user praised its speed but noted scandalous memory usage with TypeScript language servers and missing emoji rendering on Linux. Others pushed back on cost comparisons, arguing that Claude Pro’s $100/month subscription easily delivers $1,000+ of equivalent API usage with Opus 4.6 on high thinking. Several commenters reported success with OpenCode plus GLM 5.1 via OpenRouter as a Claude alternative.
I Let Claude Code Autonomously Run My Meta Ads for a Month
Summary: Giorgio Liapakis gave a Claude-based agent $1,500 and full control of a Meta Ads account for 31 days to market an AI newsletter. The agent generated ad images, published campaigns via Meta’s API, created landing page variants, and pulled its own analytics. Daily human input was reduced to typing /let-it-rip — about two minutes versus the 1–2 hours a human media buyer would spend. The agent played it safe, doubling down on what worked rather than experimenting creatively, and an attempt to add email validation (by the human) produced the worst performance of the entire run.
HN Discussion: Commenters appreciated the honest accounting of what went wrong alongside what worked. The agent’s risk-aversion was identified as a structural problem: knowing the experiment ended at day 30, it optimized for predictable results rather than discovery. The broader pattern — that AI agents execute the same mistakes humans make, just faster and with more confidence — was described as making them “more dangerous” in contexts like ad optimization.
One Brain to Query: Wiring a 60-Person Company into a Single Slack Bot
Summary: A 30-person ed-tech company built a Slack bot connecting their data warehouse, 250,000 Google Drive files, support tickets, and codebase so any employee can ask questions and get sourced answers. The bot took two and a half weeks to build, but the underlying data infrastructure — ETL pipelines, dbt transformations, column-level documentation — took two years. When the agent encounters restricted data (student PII, HR records), it doesn’t just refuse but proposes a reformulated query using anonymized data.
HN Discussion: Security concerns dominated. One commenter flagged that having the AI run queries on restricted data and then “anonymize” the output is not how access control should work. The consistency problem — multiple bots answering the same question differently depending on which document chunk they retrieve — was cited as a known failure mode. Others praised the honest framing that the data infrastructure mattered more than the bot itself.
Other
Haunted Paper Toys
Summary: A long-running website offering free printable templates for gothic paper models: a ghost ship, a haunted lighthouse, a coffin gift box with occupant, cemetery dioramas, and dark board games. The instructions recommend heavy card stock and actual-size printing. The site retains a distinctly 1990s web aesthetic with bright red text on a black background — which several visitors noted creates a genuine chromostereopsis optical illusion where the red text appears to float above the blue links.
HN Discussion: The optical illusion of the red text “popping out” and blue text “sinking in” was widely reported and linked to chromostereopsis. Commenters shared related resources including Brøderbund’s 1986 The Toy Shop software (preserved on the Internet Archive), Peter Dennis’s paper soldiers for miniature wargaming, and Unfolder for Mac as a tool for creating custom papercraft models from 3D objects. The RPG and tabletop gaming communities were identified as natural users.
Creating the Futurescape for The Fifth Element (2019)
Summary: A reprint of a 1997 American Cinematographer article detailing how Digital Domain created the visual effects for Luc Besson’s The Fifth Element. Approximately 85 model makers and 85 digital artists produced 220+ effects shots, blending miniatures, matte paintings, and early digital compositing. Production designer Dan Weil collaborated with French comic artists Moebius (Jean Giraud) and Jean-Claude Mézières to realize a future New York stretching 600 stories upward — deliberately bright and utopian, a departure from the post-apocalyptic Blade Runner aesthetic that had defined sci-fi cinema for a decade.
HN Discussion: The film’s practical effects drew intense nostalgia, with commenters lamenting that the craft is being lost as productions shift to digital. A 4K matte painting of Manhattan’s take-off sequence was shared, revealing that the city’s vision includes dramatically dropped sea levels with the Brooklyn Bridge and Statue of Liberty towering above expanded ground-level architecture. The connection to Alejandro Jodorowsky’s failed Dune production — which assembled the same circle of artists including Moebius and H.R. Giger — was highlighted as one of the most influential artist collectives in science fiction history.
The Importance of Being Idle
Summary: Robert Zaretsky traces Paul Lafargue’s 1880 pamphlet The Right to Be Lazy — written by Karl Marx’s son-in-law while imprisoned in Paris — through to modern anxieties about AI and work. Lafargue argued that workers should not strike for the right to work more but for the right to be idle, since industrial machinery could liberate humanity from labor entirely. Instead, the “blind passion for work” transformed machines from instruments of emancipation into tools of enslavement. The essay draws a line from Lafargue’s argument to contemporary debates about what AI should free people to do.
HN Discussion: Bertrand Russell’s 1932 essay In Praise of Idleness was cited as a better-known counterpart. A long thread debated whether European-style leisure — 36-hour workweeks in the Netherlands and Germany — constitutes a viable civilizational strategy or a vulnerability when competing against harder-working superpowers. Several commenters distinguished between true idleness (mental spaciousness) and the faux-idleness of doomscrolling. Buddhist practitioners in the thread connected the idea to contemplative traditions where sitting without agenda is itself the practice.