HN Evening Brief — 6 April 2026
Here’s your evening roundup of what moved on Hacker News today.
Security & Privacy
Germany Identifies the Head of Russian Ransomware Gangs REvil and GandCrab
German federal authorities have publicly identified Daniil Maksimovich Shchukin — known by the handle “UNKN” — as the figure behind both the REvil ransomware operation and the earlier GandCrab malware. An international arrest warrant has been issued. The Krebs investigation traces UNKN’s trajectory from affiliate marketing fraud through to managing ransomware-as-a-service infrastructure that extracted hundreds of millions from victims worldwide. German prosecutors coordinated with the FBI and other agencies over several years before publishing the identification.
HN Discussion: Several commenters objected to the word “doxing” in the headline, arguing that placing a wanted criminal on an official list is standard law enforcement, not doxing. Others noted that a presentation at 37C3 (the Chaos Communication Congress) had already named the individual years ago, raising the question of whether investigators used hacker-community intelligence or arrived at the conclusion independently. One thread drew parallels between the ransomware operation’s division of labour and a well-run startup.
A Cryptography Engineer’s Perspective on Quantum Computing Timelines
Filippo Valsorda, a cryptography engineer previously at Google, argues that the cryptographic community needs to accelerate post-quantum migration now rather than waiting for clearer quantum computing milestones. He walks through the specific bottlenecks: IETF standardisation of HPKE hybrid recipients took almost two years just to settle on a label string for X-Wing, and ML-KEM deployment is still crawling despite being a FIPS standard since August 2024. The essay emphasises that the lag between “cryptographically relevant quantum computers exist” and “your traffic is decryptable” is measured in years of stored ciphertext, not in the moment the machine switches on.
HN Discussion: One commenter drew a direct analogy to the Manhattan Project: we know what happens if you assemble enough qubits, the engineering path is visible, and governments are good at mobilising resources when they want to. Another pointed out that replacing key-exchange (ML-KEM for TLS/SSH) is far more urgent than replacing digital signatures, because recorded ciphertext can be decrypted retroactively, whereas forged signatures cannot travel back in time. Peter Gutmann’s sceptical counter-analysis was also referenced.
Adobe Secretly Modifies Your Hosts File
OSNews reports that Adobe Creative Cloud silently edits the system hosts file — typically used to block network connections — to detect whether its software is already installed on the machine. The modification adds entries that point Adobe domains back to localhost, apparently as a licensing check. This happens without user notification and can interfere with legitimate hosts-file customisations. The article calls the practice intrusive and questions why a company of Adobe’s scale needs to repurpose a system-level configuration file for product detection.
HN Discussion: Commenters asked why Windows Defender isn’t flagging modifications to the hosts file, which is a common target for malware. Some pushed back on the article’s framing, arguing the licensing rationale is understandable even if the implementation is heavy-handed — the real objection is the lack of transparency, not the detection itself.
AI & Tech Policy
Claude Code Is Unusable for Complex Engineering Tasks with the February Updates
A detailed GitHub issue documents how Claude Code’s engineering quality degraded after February 2026 updates. The reporter — using Claude itself to analyse its own session logs — found that thinking depth dropped roughly 67% by late February, attributed to two changes: the launch of adaptive thinking defaults with Opus 4.6 and a shift to medium effort (85) as the default. The issue includes specific examples of Claude generating code that it knows is wrong, using phrases like “simplest fix” before producing broken output. Boris from the Claude Code team responded, clarifying that the redact-thinking header only hides thinking from the UI and doesn’t affect reasoning, and that effort=85 was chosen as a sweet spot on the intelligence-latency curve.
HN Discussion: Multiple users reported the same degradation pattern across Claude Code and other tools using Opus 4.6, particularly the phrase “I’ve been burning too many tokens” appearing unprompted. One commenter called the trend of “invisible subagents” fundamentally flawed, arguing that hidden agent calls cause information collapse as models agree with their own outputs. Another pointed out the irony of using the degraded model to write the bug report about the degraded model. Several users expressed frustration at paying for a product that silently changes quality under them.
The Threat Is Comfortable Drift Toward Not Understanding What You’re Doing
This essay uses a thought experiment about two PhD students — Alice, who does her own work, and Bob, who silently delegates everything to an AI agent — to argue that the real danger of AI isn’t malevolence but atrophy. Both students produce identical papers and are indistinguishable by every quantitative metric academia uses. The author, writing from an astrophysics perspective, points out that the evaluation systems we’ve built count publications and citations, not understanding. The piece traces through what happens when Bob-type students enter industry: they can’t debug what they never built, can’t evaluate what they never understood, and the institutional incentive structure has no mechanism to distinguish them from Alice.
HN Discussion: A senior engineer shared a concrete experience: after using Claude to write moderately complex code and getting a review, they discovered they couldn’t modify their own PR because none of the decisions lived in their mental cache. Another commenter, also a senior developer, said they’re considering leaving software engineering because the intellectually stimulating parts of the job are disappearing. Several people pushed back, arguing that the market will simply stop valuing deep understanding and that Bob’s productivity with agents is what employers will reward regardless.
Nanocode: The Best Claude Code That $200 Can Buy in Pure JAX on TPUs
Nanocode is an open-source project that shows you how to train your own agentic coding model end-to-end, following Karpathy’s nanochat philosophy. Built entirely in JAX and optimised for TPUs, it uses Constitutional AI — the same approach Anthropic uses for Claude — to align a small model (1.3B parameters) with a hand-written “SOUL.md” that defines agentic behaviour. You can reproduce the 1.3B model in about nine hours on a TPU v6e-8 for roughly $200, or train a smaller 477M variant in 90 minutes for $34. The project is designed to be accessible through Google’s free TRC programme for TPU access.
HN Discussion: One commenter noted that the constitutional AI approach for tool-use training is the most interesting part — the “constitution” for a coding agent consists of operational principles like “read the file before editing” rather than the usual harmlessness criteria. Another pointed out that the example Python code in the demo actually fails the stated requirement (creating a new list instead of modifying in place), highlighting the quality gap between small models and frontier ones at agentic reasoning. The $200 cost was questioned given that free coding models already exist.
Tech Tools & Projects
Launch HN: Freestyle — Sandboxes for AI Coding Agents
Freestyle provides sandboxed execution environments designed specifically for AI coding agents. The key technical differentiator is VM-level memory forking: the system can duplicate an entire running virtual machine’s memory state in roughly 400 milliseconds, enabling agents to branch their work, test changes, and roll back without starting from scratch. The infrastructure runs on bare metal rather than cloud VMs, which the team says is necessary to achieve the sub-second fork times that agent workflows demand. Sandboxes are intentionally kept outside the customer’s main VPC to limit attack surface.
HN Discussion: Technical commenters pressed for details on how the memory forking actually works — full snapshot versus copy-on-write — and how network and socket state is handled across forks. Several people asked for a comparison matrix against competitors like Modal, E2B, Daytona, and Vercel. One commenter noted that the pricing page was broken at the time of discussion, and another observed that calculating the monthly cost of keeping an idle VM running wasn’t straightforward from the posted pricing.
sc-im: An ncurses Spreadsheet for the Terminal
sc-im (Spreadsheet Calculator Improvised) is an ncurses-based spreadsheet program that provides vim-like keybindings for cell navigation and editing. It supports up to 65,536 rows and 702 columns by default (expandable to over a million rows), and can import and export CSV, TSV, XLSX, and ODS files. The project descends from the original Unix sc spreadsheet by James Gosling and Mark Weiser. Advanced features include Lua scripting, GNUPlot integration, cell colouring, sorting, filtering, subtotals, and wide-character support for over a dozen alphabets. The maintainer notes it’s a one-person project struggling with funding.
HN Discussion: The project’s pedigree — James Gosling, who later created Java, wrote the original sc — drew appreciative comments. Users compared sc-im favourably to the morning’s featured terminal spreadsheet (Sheets), noting that sc-im is more mature and feature-rich despite its steeper learning curve.
Reducto Releases Deep Extract Agent
Reducto, a document parsing company, announced Deep Extract, an agent-based system for extracting structured data from complex documents. Rather than relying on a single LLM pass, Deep Extract iteratively refines its extraction by comparing intermediate results against the source document, attempting to catch errors that one-shot parsers miss. The system is designed for documents where layout complexity — tables nested within tables, multi-column formats, mixed handwriting and type — defeats conventional OCR and rule-based extraction.
HN Discussion: Commenters focused on the competitive landscape, noting that document extraction has become crowded with AI-powered entrants. Several asked about accuracy benchmarks against tools like DocETL or unstructured.io, and whether the iterative refinement approach actually produces measurably better results or just burns more tokens.
Sky — An Elm-Inspired Language That Compiles to Go
Sky is an experimental programming language that combines Elm’s type system — Hindley-Milner inference, algebraic data types, exhaustive pattern matching — with Go’s compilation target and ecosystem. The compiler, CLI, formatter, and LSP are self-hosted (written in Sky, compiled to a ~4MB Go binary) and bootstrap through multiple generations of self-compilation. The most distinctive feature is “Sky.Live,” a server-driven UI system inspired by Phoenix LiveView: the server renders interactive UI via DOM diffing over SSE (Server-Sent Events), no WebSocket or client framework required. The author built it to avoid the friction of separate React/TypeScript frontends and Go backends.
HN Discussion: The self-hosting compiler and LiveView-inspired UI approach drew comparisons to Elm’s own trajectory and to languages like Gleam that also target the BEAM ecosystem. Several commenters were sceptical of the “single binary fullstack” pitch, noting that server-driven UI over SSE has latency implications that WebSocket-based approaches avoid. The use of Claude Code to accelerate language development was mentioned as both impressive and a potential concern for long-term maintainability.
I Replaced Kafka, Redis, and RabbitMQ with One Tool — NATS
A backend engineer’s deep dive into replacing a three-tool messaging stack (Kafka for event streaming, Redis for caching, RabbitMQ for task queues) with NATS. The article covers Core NATS (fire-and-forget pub/sub), JetStream (persistent streams with replay), Key-Value stores, and Object Storage — all running from a single binary. Core NATS disconnects slow consumers rather than stalling the system, which the author describes as a circuit-breaker mechanism. The piece is transparent about being AI-assisted in writing, though the author emphasises the insights come from weeks of hands-on production experience.
HN Discussion: A long-time NATS user shared their experience running it for a public transit information site, processing real-time vehicle telemetry through pub/sub paired with Tile38 for geofencing. Several commenters flagged that the article’s AI-assisted writing was a barrier — one said they stopped reading because of it. A previous HN discussion on NATS from three months ago was linked for deeper technical context.
Datakool — Privacy-First Google Analytics Alternative
Datakool is a web analytics platform built around cookieless tracking. The tracking script is under 1KB (compared to Google Analytics’ 75KB+), collects no personal data, and uses daily-rotating salt hashes for IP addresses. The dashboard provides real-time visitor counts, traffic source breakdowns, device and location analytics, goal tracking, and UTM campaign monitoring. It also exposes an MCP-compatible API so AI assistants like Claude Code can query analytics data directly. GDPR, CCPA, and PECR compliance is handled by design since no consent banner is needed.
HN Discussion: The core question raised was what Datakool offers beyond “privacy + lightweight” that isn’t already commoditised by Plausible, Fathom, and the dozen other GA alternatives. One commenter argued that without a free tier for hobby sites, adoption would be limited.
OpenJDK Project Panama: Interconnecting JVM and Native Code
Project Panama is OpenJDK’s long-running effort to replace JNI with a modern, safer foreign function and memory API. The project includes the Foreign Function & Memory API (finalised in JEP 424) for calling native C libraries without hand-written JNI glue, a Vector API for SIMD intrinsics in Java, and the jextract tool that mechanically generates Java bindings from native library headers. The goal is to make calling native code from Java as straightforward as calling Java code, eliminating the boilerplate and error-prone manual marshalling that JNI requires.
HN Discussion: One commenter called it “the inverse of JNI” — where JNI required you to write C to talk to Java, Panama lets you call C from Java directly. Another noted that Android’s Java runtime will likely never support Panama. A snarkier take pointed out that Java is “innovating” by catching up to capabilities other languages have had for three decades.
Show HN: Pace — An MCP Server That Connects Claude to Your Wearables
Pace is an MCP (Model Context Protocol) server that connects wearable device data — from Garmin, Polar, Whoop, and over 20 other services — directly to Claude. Users link their devices once and can then analyse sleep, training load, activity, and trends in natural language through Claude. The system stores data in PostgreSQL on Google Cloud SQL and runs the MCP server on Cloud Run. Built by a former professional athlete who wanted to analyse their own training data conversationally.
HN Discussion: Commenters asked about Garmin’s restrictive API rate limits (roughly 25 requests per 15 minutes) and how the system handles the high-frequency data that sleep tracking generates (data points every 30 seconds). The JSONB-in-Postgres storage approach raised questions about query performance over 90-day periods.
Web & Infrastructure
I Won’t Download Your App. The Web Version Is A-OK
The author argues that for most services, a well-built responsive website is preferable to a native mobile app. Apps demand storage, permissions, background processes, and ongoing updates for functionality that a browser can deliver. The piece catalogues the asymmetry: browsers sandbox by default, let you block ads and trackers, and don’t require an approval process to publish content. The insistence on app-first experiences, the author suggests, is driven by companies wanting deeper access to users’ devices and attention, not by genuine technical need.
HN Discussion: The thread became a crowdsourced hall of shame: Reddit blocks “unreviewed” content on mobile web but not desktop, PayPal won’t let you pick your 5% rewards category without the app, SeatGeek requires the app to attend events at all. One developer shared the counterintuitive finding that building a thin app wrapper around their website led to a 10x increase in paid-plan conversions. The generational divide came up repeatedly — younger users whose first internet touchpoint was a smartphone naturally gravitate toward apps, while desktop-era users prefer the browser.
Friendica — A Decentralized Social Network
Friendica is one of the oldest decentralised social networking platforms, predating the term “fediverse.” Written in PHP with MySQL, it can run on any shared hosting that supports WordPress. It supports ActivityPub, OStatus, and diaspora* protocols, and uniquely offers two-way email communication via IMAP/ESMTP, plus RSS feed importation into your social stream. Access lists, private groups, and content expiration are built in. The latest stable release (2026.01) includes accessibility improvements and security fixes.
HN Discussion: A user called Friendica “the best social network in the fediverse” but criticised its UX as “designed by a Mordor goblin” — too many hidden settings and multiple paths to the same function. Others remembered using it as far back as 2014. The perennial fediverse problem surfaced: the software is technically excellent but nobody’s on it, and self-hosting requires technical knowledge that excludes most potential users.
System Administration
TIL: The Numbers in Man Page Titles Like sleep(3)
A concise explanation of the section numbers in Unix manual pages: section 1 is executable commands, section 2 is system calls (kernel functions), section 3 is library calls (libc functions), and section 5 is file formats. The author discovered this after a code review where a colleague corrected a comment referencing man 2 basename — since basename is a libc function, the correct reference is basename(3). Suffixes also carry meaning: p indicates a POSIX page, and x indicates X Window System documentation.
HN Discussion: The thread became a collection of man-page trivia. Section 5 was highlighted as particularly useful — man 5 crontab gives you the crontab format syntax, not the cron daemon documentation. Someone shared the hidden joke in the BSD tunefs man page source: “You can tune a file system, but you cannot tune a fish.” Another noted that sections don’t have to be numeric — TCL uses section n.
Stamp It! All Programs Must Report Their Version
Michael Stapelberg argues that software versioning standards are absurdly low compared to physical appliances — your dishwasher has more detailed version information than most developer builds. The essay describes a three-step process: stamp the VCS revision into the binary at build time, plumb it through the application’s internals, and report it via --version flags, HTTP headers, and log lines. The i3 window manager’s --moreversion flag, which Stapelberg designed over a decade ago, is used as the reference implementation. The article details the complications that NixOS introduces for Go binaries, which always stamp VCS information but lose it under Nix’s build sandboxing.
HN Discussion: Commenters who’ve fought version-visibility problems during incident response strongly agreed with the premise. The NixOS-specific issues around Go binary VCS stamping generated technical discussion about overlay workarounds and the broader tension between reproducible builds and build-time metadata injection.
Business & Industry
What Being Ripped Off Taught Me
A freelancer’s account of being stiffed on payment and the practical lessons that followed. The essay covers the emotional arc — disbelief, anger, rationalisation — and distils it into concrete advice: negotiate payment terms upfront, stop work when invoices are late, require advance payment from untrusted clients, and treat reluctance to agree to reasonable payment terms as a signal to walk away. The underlying argument is that most payment problems are preventable with better contracts and earlier enforcement.
HN Discussion: The comment thread turned into a shared playbook for freelancer payment protection. One consultant shared their full contract clause list, including late-payment interest at 8% above the Bank of England base rate, jurisdiction requirements, and the policy of withholding deliverable downloads until payment clears. They reported that across roughly 2,000 invoices over five years, clients who received stopped-delivery treatment paid on average 11 days sooner. The consensus: clients who resist reasonable payment terms are the same clients who won’t pay you.
Drop, Formerly Massdrop, Ends Most Collaborations and Rebrands Under Corsair
Drop (originally Massdrop) has ended most of its community-driven product collaborations and is rebranding as a Corsair subsidiary focused on licensed merchandise — Lord of the Rings keyboards, gaming accessories, and Corsair-family products. The community-buy model that made Massdrop distinctive — group purchases of enthusiast gear like headphones, knives, and mechanical keyboards at discount — has been effectively discontinued. The drop.com domain now redirects to Corsair’s branded store.
HN Discussion: The thread reads like a eulogy. Users listed the durable products they bought during Massdrop’s golden era — keyboards, DACs, pocket knives, titanium straws — and noted the paradox: high-quality goods don’t generate repeat business because they last. “I’m still using my Drop CTRL keyboard from 2018. I haven’t bought another keyboard since then because it’s a good keyboard.” Someone asked where to find the enthusiast group-buy model that Massdrop pioneered, and no satisfying successor was named.
Show HN: GovAuctions — Browse Every Government Surplus Auction at Once
GovAuctions aggregates government surplus and seizure auction listings from dozens of separate, often poorly designed auction sites into a single searchable interface. Users can filter by location, category, and price, save items to a watchlist, and set up alerts for new matching auctions. The creator built it after years of frustration manually scanning the fragmented landscape of government auction sites, which have interminable load times and navigation that resets to the homepage when you hit the back button. Current notable listings include the San Diego DHS selling 26 tons of lead shot with bidding starting at $1,000.
HN Discussion: Commenters appreciated the practical value of the tool, given how notoriously bad individual government auction sites are. The conversation turned to the kinds of unusual items that surface on government auctions — lab equipment, vehicles, and industrial materials — and the risks of buying seized goods without inspection.
Geopolitics & War
France Pulls Last Gold Held in US for $15B Gain
The Banque de France has completed the repatriation of its remaining gold reserves stored in the United States, generating a reported capital gain of €13 billion ($15 billion). Rather than physically refining and transporting the older, non-standard gold bars, the bank sold them at current market prices and purchased new compliant bullion in Europe. The operation echoes de Gaulle’s 1960s policy of systematically converting US dollars into gold, which contributed to the strain that led Nixon to close the gold window in 1971. The move reflects broader concerns about the security of sovereign assets held under foreign custody.
HN Discussion: A heated debate erupted over whether this constitutes a “gain” at all, since France ended up with the same quantity of gold it started with — the €13 billion is an accounting profit from selling appreciated bars and rebuying at current prices. Several commenters argued the real motivation is custody risk: gold held in the US is effectively “paper gold” that could be seized or frozen, particularly given current geopolitical tensions. The historical parallel to the 1960s gold repatriation was drawn out in detail.
Academic & Research
My University Uses Prompt Injection to Catch Cheaters
A first-year computer science student discovered that their university hides instructions in zero-font-size text within assignment specifications — invisible to human readers but picked up when students copy-paste the entire document into an LLM. The hidden text instructs the AI to include specific markers in its output, which graders can then detect. The student’s floormate found the hidden instructions after pasting the assignment text into a code comment in his IDE.
HN Discussion: A marketing instructor confirmed they use the same technique with white-text prompt injections in PDFs, sometimes prompting the LLM to append “I submit this assignment without checking its output.” They initially did it for amusement, then made it a teaching moment — but students kept falling for it repeatedly. Others debated whether prompt injection is a sustainable arms race or a symptom that assessment methods need fundamental redesign. One commenter asked why LLM providers don’t ship a small dedicated model to detect prompt injection before the main model processes input.
History & Science
Make Your Own ColecoVision at Home, Part 5
The fifth instalment of a series documenting the construction of “Leako,” a DIY clone of the 1982 ColecoVision games console. This revision focuses on reliability improvements: replacing the barrel-jack power input with USB, addressing board-flex issues caused by cartridge insertion force, and navigating ongoing component shortages. The USB power change introduced its own problems — cheap USB chargers produce ripple that can damage salvaged ICs, requiring a quality power adapter. The custom KiCad footprint for the controller ports was missing copper rings for strain-relief pins, weakening the physical connection. The project is on its seventh board revision.
HN Discussion: The thread was sparse at the time of writing, but the project drew appreciation as a meticulous example of hardware archaeology — recreating a vintage console from scratch while documenting every design decision and mistake.
Show HN: Mvidia — A Game Where You Build a GPU
Mvidia is a browser-based puzzle game that teaches GPU architecture by having players construct one from first principles. You start with individual transistors, wire NMOS gates, build logic circuits, and progressively assemble the components of a functional graphics processor. The creator built it because existing resources for learning GPU architecture were too theoretical and lacked hands-on interaction. Each level introduces a new component — truth tables, capacitors, enable gates — with the goal of making hardware design tangible.
HN Discussion: A veteran IC designer with over a decade of full-custom mixed-signal experience reported getting stuck on the first level due to a UI ambiguity: a grey line in the background grid looked like a pre-wired connection from input to gate, but wasn’t. Another commenter recommended Turing Complete on Steam as a similar (though long-in-early-access) alternative where you eventually build your own CPU with a custom assembly language.
Tiny Corp’s Exabox
Tiny Corp (the company behind the tinygrad ML framework) is taking pre-orders for the Exabox, a compact ML inference server featuring an AMD RDNA5 GPU. The device is marketed as a self-contained unit for running models locally, consistent with Tiny Corp’s philosophy of reducing dependence on cloud GPU providers. The product page is hosted on Shopify, and photos show the hardware in what appears to be a rugged outdoor setting. Specifications list an “RDNA5 AT0 XL” GPU with 154 CUs and 36GB of VRAM — a chip that hasn’t been officially announced by AMD yet.
HN Discussion: The RDNA5 GPU reference drew attention, as the architecture hasn’t been publicly launched — commenters speculated about whether Tiny Corp has early access or is using leaked naming conventions. The physical security of the device was questioned: one commenter noted it looks like “an amateur thief with a crowbar could probably crack that open.” Several people were surprised that a hardware preorder of this nature runs through a standard Shopify storefront.
Other
Book Review: There Is No Antimemetics Division
Stephen Diehl reviews qntm’s “There Is No Antimemetics Division,” a novel originally serialised on the SCP Foundation wiki about a fictional branch of the SCP Foundation dealing with antimemes — ideas that resist being remembered or communicated. Diehl connects the book’s premise to real phenomena in software engineering: institutional knowledge that evaporates when key people leave, the way critical infrastructure documentation goes unread, and open-source maintainership as a form of cosmic horror where essential work is inherently invisible to those who benefit from it.
HN Discussion: Several commenters felt the review was more plot synopsis than critical analysis, lamenting the absence of quotes that would convey the writing quality. Those who’d read the book were split: one called it a must-read for prolific sci-fi readers seeking something genuinely different, while another found the ending’s shift toward a Christian-Neoplatonist metaphysics jarring — a 90-degree turn from the book’s earlier preoccupation with an inaccessible universe. The original SCP wiki version was linked as a free alternative to the published edition.
81-Year-Old Dodgers Fan Can No Longer Get Tickets Because He Doesn’t Have a Smartphone
An 81-year-old man who has held Los Angeles Dodgers season tickets for over 50 years can no longer attend games because the team has moved to digital-only ticketing. Previously, the Dodgers made exceptions by custom-printing paper tickets for him at a surcharge, but that accommodation has been discontinued. The move is part of a broader MLB push toward mobile ticketing, motivated partly by anti-scalping measures — digital tickets can be tracked and large-volume transfers flagged.
HN Discussion: The anti-scalping rationale drew some sympathy, but most of the thread focused on the broader pattern of essential services becoming smartphone-dependent: parking in Brazilian cities, bank verification codes that arrive as MMS, airline boarding passes. One commenter proposed extending the ADA to cover people who aren’t technologically adept. Another suggested the Dodgers could have turned this into a PR win by awarding the fan a lifetime paper pass after 50 years of loyalty.
The Last Quiet Thing
A reflective essay about the disappearance of genuinely quiet, disconnected objects from daily life. The author contrasts a $12 Casio watch — which does nothing but tell time — with a $400 smartwatch that tracks steps, blood oxygen, sleep quality, and sends notifications. The piece meditates on how every object with a network connection now demands system administration, and how the simplicity of single-purpose devices has become a luxury. It’s less a technology critique than a lament for the loss of objects that require nothing from you.
HN Discussion: One commenter pointed out the irony that Casio itself now makes the Moflin, an AI-powered companion device. Another shared that their smartwatch has become an essential prosthetic for managing ADHD — the constant buzzing ensures they don’t miss appointments. The essay’s format drew criticism as much as its content: one reader found the layout frustrating despite appreciating the underlying argument.
How Paris Swapped Cars for Bikes — and Transformed Its Streets
The Guardian surveys the legacy of Anne Hidalgo’s 12-year mayoralty in Paris, during which the city added hundreds of kilometres of bike lanes, pedestrianised 300 school streets, banned cars from the Seine riverbanks, planted 155,000 trees, and converted parking spaces into café terraces and green areas. Her successor Emmanuel Grégoire won the recent municipal election on a platform largely continuing the same policies. The piece notes that low voter turnout in key referendums (on SUV parking charges, school street pedestrianisation) reveals ongoing tension between the vision and public enthusiasm, and that right-wing opponents criticised the “anxiety-inducing” chaos of constant construction without proposing to reverse the changes.
HN Discussion: Cyclist behaviour in Paris was the thread’s lightning rod — several commenters said cyclists running red lights is the city’s real unsolved problem. Others compared the transformation to Amsterdam’s decades-long shift away from car dependency, noting that Paris accomplished in 12 years what took Amsterdam much longer, partly because Paris’s unusually dense urban layout makes car reduction immediately impactful. The low referendum turnouts were discussed as evidence that progressive urban policy faces a motivation gap: supporters assume change is inevitable, opponents are energised to vote.
That’s the evening batch. See you tomorrow morning.