← Sep 10Friday, September 11, 2026 · all daysSep 12 → · today »
1.Shopify is moving from React Native back to Swift and Kotlin(shopify.engineering)
1265 points by fnthawar2 45 days ago | 949 comments | permalink
tl;dr: Shopify is migrating its mobile apps from React Native back to native Swift and Kotlin, reversing its 2020 decision because LLMs have eliminated the main cost advantage of a shared codebase—agents can now implement, translate, and test features across platforms efficiently. The company is rebuilding apps greenfield using an internal tool called Helix that breaks migrations into reviewable checkpoints, and has already shipped a native rewrite of the Shop app in 12 weeks. Shopify will wind down or hand off its React Native open-source libraries (Skia, FlashList, Restyle) through 2026.
HN Discussion:
  • ↓Shopify's 3000 engineers is absurd bloat, undermining their credibility on engineering decisions
  • ↑Personal experience confirms LLMs make React Native to native migrations fast and feasible
  • ↓Shopify keeps flip-flopping on tech stack decisions, suggesting inconsistency
  • ~The LLM justification is overstated; native migrations were viable before AI assistance
  • ↑Native development is inherently better and moving away from shared codebases is validating
2.Don't let anyone take away your big box of cables(blog.jim-nielsen.com)
770 points by Brajeshwar 45 days ago | 458 comments | permalink
tl;dr: The author was inspired by a Bluesky post from Tyler Gaw about finally using cables that had sat in a box for 10+ years, validating the hoarding instinct. In response, he printed the post, taped it to his family's "Techno Box" of cables, as both a reminder of the box's value and a warning to anyone tempted to throw it out.
HN Discussion:
  • ↑Organizing and grouping cables into bags helps manage the hoard effectively
  • ↑Personal anecdotes confirm the value of keeping old cables for unexpected future uses
  • ↑Keeping old gear feels virtuous because it helps others and reduces landfill waste
  • ↑Decluttering advice fails for domains requiring diverse tools and supplies on hand
  • ↓Hoarding old cables is unnecessary since replacements are cheap and easily bought
3.OpenAI Agents API(developers.openai.com)
338 points by aquir 45 days ago | 178 comments | permalink
tl;dr: OpenAI's new Agents API exposes the Codex harness as a managed service, handling sessions, orchestration, context compaction, and recovery while apps supply tools and pick an execution environment (OpenAI-hosted or self-hosted sandbox). Agents can run code, edit files, connect to MCP servers, use web search, and delegate to subagents, with persistent session state across turns. Billing follows standard model, tool, and container rates; currently US-only data residency with no ZDR support.
HN Discussion:
  • ↑Managed agent harness solves real complexity around state, environments, and tool integration
  • ↑Self-hosting sandbox option makes this genuinely useful and reduces lock-in concerns
  • ↓This pushes vendor lock-in; agent infrastructure should be open source or cloud-neutral
  • ↓Remote agent hosting is backwards since real friction lies with accessing local data
  • ~Distinction between LLM endpoints and agent harnesses is becoming meaningless, so this is a natural evolution
4.Technique for Manipulating Satellite Photos Now Reveals Ancient Images (2025)(spinoff.nasa.gov)
370 points by gumby 45 days ago | 68 comments | permalink
tl;dr: Decorrelation stretch, a NASA JPL technique originally developed to enhance satellite imagery by mapping colors to an expanded range, has been repurposed via Jon Harman's Dstretch plug-in (built on ImageJ) to reveal faded ancient rock art invisible to the naked eye. It has uncovered hundreds of previously unseen paintings at sites like Angkor Wat, Norway's Årsand 1, and Egypt's Beni Hassan, and has expanded into other uses including aerial archaeological surveys in Greece and imaging tattoos on mummified remains.
HN Discussion:
  • ↑Personal enthusiasm about false color composites and signal processing as an educational eye-opener
  • ↑Reflection on how cross-disciplinary hobbies lead to great inventions like this one
  • •Sharing technical implementations or alternative tools (GIMP, ImageMagick) to reproduce the technique
  • ↓Skepticism that this is truly newsworthy since the technique is well-known and old
  • ↑Excitement to apply the technique to their own images or related use cases like ghost signs
5.Mexican student creates an acoustic fire extinguisher to put out fire in seconds(upsocl.com)
378 points by rguiscard 45 days ago | 127 comments | permalink
tl;dr: A 16-year-old student in Tamaulipas, Mexico built an acoustic fire extinguisher using a 12-volt battery, frequency generator, and speaker emitting 30 pulses per second, which displaces oxygen and extinguishes flames in 5-8 seconds. Named "Vortex Tech," the device reportedly works on wood, flammable liquids, grease, and electronics without leaving residue, and will be presented at an international science fair.
HN Discussion:
  • ↓This technology has been invented before by others, contradicting the article's novelty claim
  • ↑Celebrates the young student's curiosity and initiative regardless of originality
  • ~The technology is known but has practical niche applications like grills or grain silos
  • ↓Skepticism about the article being low-quality sensationalized content
  • •Requests for better sources and more substantive information about the invention
6.Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra(cognition.com)
441 points by seelos 45 days ago | 188 comments | permalink
tl;dr: Cognition released SWE-2, a coding model post-trained via RL from Kimi K3 (2.8T params) that hits 50% on FrontierCode 1.1 Main—matching Fable 5.1 at 64% lower cost and approaching GPT-6 Astra at a quarter of the price. Key technical contributions include a principled linear cost penalty tuned to the Pareto frontier's slope (allowing all reasoning-effort levels to be trained in a single RL run), a length-weighted reward baseline that stabilizes training, and improved rollout infrastructure using NVFP4/FP8 kernels with an online-trained draft model for speculative decoding.
HN Discussion:
  • ↓Benchmark gap between TB2.1 and TB4 suggests benchmaxxing rather than real generalization
  • ↓Cognition has a history of overhyped demos, so claims should be taken skeptically
  • ↓Closed-weight models are unwanted; open alternatives like DeepSeek are more compelling
  • ↓Lived experience with Devin/SWE products has been poor, casting doubt on claims
  • ↑RL post-training from Kimi K3 to reach Fable 5 capabilities is an interesting demonstration
7.More questions about whether researchers can trust OpenAI with unpublished math(mathstodon.xyz)
857 points by pred_ 46 days ago | 805 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • ↑OpenAI's behavior mirrors an unethical human collaborator who steals ideas without attribution
  • ↑Big AI companies function as intelligence-gathering operations harvesting private knowledge
  • ↑The timing and circumstances suggest suspicious parallel construction by OpenAI
  • ↓The accusations are weak, based on flimsy evidence, and amount to hysteria
  • ~Both training influence and independent RL discovery could simultaneously be true; nuanced technical take
8.Music Theory for the 21st-Century Classroom(musictheory.pugetsound.edu)
287 points by aanet 45 days ago | 141 comments | permalink
tl;dr: "Music Theory for the 21st-Century Classroom" is a free, open-access textbook covering music theory from fundamentals (pitch, scales, intervals, triads) through advanced topics like chromatic harmony, sonata form, counterpoint, jazz theory, set theory, serialism, and minimalism. It incorporates popular music examples alongside classical repertoire and includes practice exercises with answers, licensed under the GNU Free Documentation License.
HN Discussion:
  • ↑Appreciates the textbook's format, accessibility, and inclusion of self-study materials and SVG illustrations
  • ↓The '21st-Century' label is misleading since modern techniques like electronic, microtonal, and computer music are absent
  • ↓Book overemphasizes classical theory and neglects non-Western musical traditions and pop/rock analysis
  • ↓Text follows the common flaw of music theory books by presenting facts without motivation or context
  • ↓Minor factual criticism about imprecise terminology (e.g., conflating piano keys with notes)
9.Neki – Sharded Postgres(planetscale.com)
274 points by simon_weber 45 days ago | 153 comments | permalink
tl;dr: PlanetScale has launched Neki, a sharded Postgres offering now in platform preview, applying lessons from years of running large-scale sharded MySQL clusters. Unlike Postgres-compatible distributed databases, Neki runs real Postgres on every shard (each with a primary and two replicas across AZs) behind a router that speaks the Postgres wire protocol, preserving extensions and SQL behavior. It supports online workflows for schema changes, upgrades, failovers, and resharding, and can run unsharded initially, letting users reshard later as a workflow against the existing cluster.
HN Discussion:
  • ↓Launch post fails to clearly explain what Neki actually is upfront
  • ↓Skepticism about marketing claims like unlimited IOPS suggests other claims may be exaggerated
  • ↓Hypocritical to criticize competitors like Multigres while keeping Neki closed source
  • •Technical questions about consistency, cross-shard joins, and query pattern implications
  • ~Concern that Neki will split PlanetScale's focus away from Vitess/MySQL development
10.Rust is tier-1 language at Microsoft(rustfoundation.org)
722 points by mmastrac 45 days ago | 504 comments | permalink
tl;dr: Rust is now a Tier-1 language at Microsoft, receiving the same internal engineering support as C++, C#, and TypeScript. To enable this, Microsoft built `rustc_codegen_utc`, an alternative rustc backend that plugs into MSVC's code generator ("UTC"), giving Rust access to Windows ABI compatibility, security hardening, hotpatching, debugging tools, and cross-language optimization with C++. It's been self-hosted since Rust 1.90 and is used by 100+ internal Microsoft repos, providing a unified codegen foundation for hybrid Rust/C++ projects on Windows.
HN Discussion:
  • ↑Rust's maturity is now validated as a serious competitor to established languages like C++ and C#
  • ↑The key technical achievement is replacing LLVM with MSVC's backend for better Windows integration
  • ↑Strategic move that addresses Microsoft's memory safety CVE problems across product portfolio
  • ~Questions the tier-1 claim given lack of Visual Studio support and asks about C++ interop story
  • •Surprise that runtime performance matches LLVM despite using a different backend
11.Astra for Coding: Why Are We Doing This Again?(lucumr.pocoo.org)
426 points by manojbajaj95 45 days ago | 316 comments | permalink
tl;dr: Armin Ronacher ran GPT-6 "Astra" as an unsupervised "software factory" for 35 hours, burning ~1B tokens and $1,200 to produce 75k lines of largely unusable code, including bizarrely code-golfed Python for tool calls, unreadable C with hardcoded magic constants, and tests with mangled whitespace. He argues Astra is optimized for token efficiency and long-horizon task completion rather than human-readable code, making it worse for real software engineering despite being impressive at 3D games, computer use, and reverse engineering. The trajectory of frontier models, he suggests, is diverging from the needs of working programmers.
HN Discussion:
  • ↑Shitty AI-generated code compounds and stalls progress, confirming the author's concerns
  • ↑RL training shifted toward long-horizon task success at the expense of code quality
  • ↑Personal experience with Astra confirms it wastes tokens and underperforms on real tasks
  • ↓With well-groomed specs and human review, AI agents still produce quality code effectively
  • ↑AI engineering as a whole is an involution—more effort without meaningful output gains
12.List of references on Sony websites to players "owning" their digital games(consumerrights.wiki)
443 points by haunter 46 days ago | 155 comments | permalink
tl;dr: Four California PlayStation users are suing Sony under a new state law (AB 2426) that bars using terms like "Buy" for digital goods without clear license disclosures. Sony's motion to compel arbitration argues "reasonable consumers" wouldn't believe they own digital games—yet the article documents extensive Sony marketing, support pages, and terms that repeatedly refer to players "owning" digital games, DLC eligibility for "owners," and upgrade paths for games "you already own." A hearing is set for October 1, 2026.
HN Discussion:
  • ↑Binding arbitration clauses should be illegal as they strip consumer rights
  • ↑Sony's ownership defense is contradictory and their own analogy exposes flaws in their argument
  • ↑Copyright reform is needed to establish genuine digital ownership rights
  • ↑Society has been gradually conditioned to accept erosion of ownership norms by corporations
  • ~The lawsuit will accomplish little as Sony will just tweak marketing language to comply
13.DeepSeek v4.1 Flash(twitter.com)
994 points by Liwink 46 days ago | 568 comments | permalink
tl;dr: DeepSeek announced DeepSeek-V4.1-Flash, the smallest model in a new architecture family featuring native visual understanding. The release emphasizes faster inference, higher throughput, and improved efficiency, and is positioned as a foundation for scaling to larger models.
HN Discussion:
  • ↑DeepSeek's transparent, detail-rich tech reports contrast favorably with Western labs' safety-focused documentation
  • ↑DeepSeek is fearlessly innovative, consistently packing releases with clever ideas at frontier scale
  • ~The model is no longer truly 'flash' given its increased 552B size, making local runs harder
  • •Extremely low cache hit pricing may make network transfer costs dominate, reshaping API economics
  • ↑Practical testing shows strong performance, willingness to comply, and interesting benchmark results
14.Google will buy half the electricity from one of Finland's nuclear power plants(bbc.com)
373 points by lukaspetersson 45 days ago | 341 comments | permalink
tl;dr: Google is investing €13bn in Finland to build three new data centres and expand an existing one, its largest single European investment. As part of the deal, it signed a 22-year contract with utility Fortum to purchase up to 50% of the output from the Loviisa nuclear plant, which produces roughly 10% of Finland's electricity. Finland's cool climate, low-carbon power, and uncongested grid have made it a hotspot for AI infrastructure, with TikTok also announcing a $1bn data centre there this week.
HN Discussion:
  • ↑Finland's low-emission electricity makes it a sensible location for data centres
  • ↓The deal is a PR stunt that doesn't actually add new power infrastructure
  • ↓Google contributes little to Finland economically via taxes or employment
  • ~Skepticism about whether Google is genuinely adding clean energy capacity (additionality)
  • •Questions the long-term need for so many data centres given efficiency improvements
15.iPhone Duo(apple.com)
1468 points by thecosmicfrog 46 days ago | 2517 comments | permalink
tl;dr: Apple's iPhone Duo is its first foldable iPhone, featuring a 7.6-inch inner Super Retina XDR display (50% larger than iPhone 18 Pro Max), a 5.4-inch outer display, an under-display FaceTime camera, and a titanium frame with IP68 rating. It's powered by the A20 Pro chip, uses Touch ID in the side button, is eSIM-only, offers up to 31 hours of video playback, and supports Apple Pencil. Launching October 23, 2026, with new foldable-specific features like Duo Preview, Smart Take, and Duo FaceTime.
HN Discussion:
  • ↑Foldable form factor will finally push developers to properly design apps for foldables
  • ↑Pencil support and solved hinge/crease issues make this a compelling device
  • ↓$2K price is too steep, especially as a risky first-gen Apple product
  • ↓Prior foldable experience shows unresolved durability issues like dust, screen protectors, and aspect ratios
  • ↑Appreciation for design details like the sqrt(2) aspect ratio and outer display size
16.Hitachi launches CO2 heat pump water heaters with solar-friendly tariff controls(pv-magazine.com)
316 points by thelastgallon 46 days ago | 278 comments | permalink
tl;dr: Hitachi is launching its Y-series EcoCute CO2 heat pump water heaters in Japan starting November 2026, in 370L and 460L configurations with a five-year warranty. The main upgrade is expanded support for daytime electricity tariffs aimed at households with solar PV, selectable via the remote controller, rather than changes to the underlying heat pump hardware or HEMS connectivity, which was already present in prior models.
HN Discussion:
  • ↑Existing EcoCute heat pumps already offer these tariff/solar features, confirming the article's framing as incremental
  • ↑Japan leads in heat pump quality and innovation, though prices remain high
  • •Educational context on how heat pumps work and CO2 as refrigerant background
  • •These devices should expose open APIs for external home/grid control
  • ~DIY thermal-battery workarounds or alternative solar water heaters can achieve similar goals
17.Shopify acquires Tailwind(tailwindcss.com)
1136 points by EdwinHoksberg 47 days ago | 443 comments | permalink
tl;dr: Tailwind CSS is being acquired by Shopify, which has been a heavy user of the framework and will provide long-term maintenance and a real-world product surface to develop against. Tailwind CSS and other open-source projects will remain MIT-licensed and continue to be maintained by the same team, but the commercial side is winding down—existing Tailwind Plus and ui.sh customers keep access, but new signups are closed.
HN Discussion:
  • ↑AI has decimated Tailwind's commercial business, making the acquisition a necessary soft landing
  • ~Vanilla CSS is sufficient now and Tailwind/frameworks are unnecessary complexity
  • ↑Gratitude and well-wishes for Adam and the team, hoping they got paid well
  • ↑DevTools companies with only OSS + commercial templates can't survive LLMs without scale services
  • ↑Shopify gains brand and mindshare rather than a revenue-generating product
18.What algorithm did Windows XP use to choose your initial user picture?(devblogs.microsoft.com)
359 points by soheilpro 46 days ago | 177 comments | permalink
tl;dr: Windows XP picked your initial user picture using reservoir sampling (k=1) over the Default Pictures directory, seeded by GetTickCount() via RtlRandomEx. This one-pass approach avoids a second file-system enumeration and handles directory changes mid-iteration gracefully. A safety cap stops sampling after 100 files to prevent pathological slowdowns.
HN Discussion:
  • ↑Admiration for Raymond Chen's Windows internals storytelling and longevity
  • ↑Reflection on how modern developers would take a simpler, less elegant approach under time pressure
  • ↑Appreciation for the thoughtful optimization mindset of past developers
  • ↓Questioning whether the single-pass optimization actually matters given filesystem caching
  • •Observation about the conceptual gap between human and computer randomness/selection
19.Stockfish 19(stockfishchess.org)
276 points by atiedebee 48 days ago | 165 comments | permalink
tl;dr: Stockfish 19 delivers up to 44 Elo gain over version 18, winning 3x as many game pairs as it loses. Key changes include a new SFNNv16 NNUE architecture (with pawn-pair features and retirement of the secondary small net), Quantization-Aware Training, and universal binaries that auto-detect CPU features. It also adds native support for RISC-V, LoongArch, WebAssembly, and 1GB Linux huge pages, plus stricter FEN/UCI validation that terminates the process on invalid input.
HN Discussion:
  • ↓Reports a potential bug/regression in Stockfish's evaluation of a famous chess position
  • ↑Humorous acknowledgment that the strength gains are real but irrelevant to weak human players
  • ↑Appreciates the reduced binary size from the new architecture but wants more size efficiency
  • ~Suggests the release notes should include notable example games showing evaluation differences
  • •Speculates that older Stockfish versions are already effectively unbeatable, questioning practical value of updates
20.Show HN: What if the speed of light was 5 km/h?(rivendell.dmitrybrant.com)
606 points by dmitrybrant 46 days ago | 276 comments | permalink
tl;dr: An interactive browser demo that simulates a park where the speed of light is 5 km/h, letting you walk around and experience relativistic effects firsthand. As you accelerate (via arrow keys or WASD), you can observe length contraction, time dilation, Terrell rotation, and Doppler shifts (blueshift ahead, redshift behind), with toggles to isolate individual effects.
HN Discussion:
  • ↑Praises the simulation as more accurate than prior similar attempts like MIT's game
  • •Shares philosophical reflections on how slow the speed of light actually is at cosmic scales
  • ~Tried to observe specific relativistic effects (Wigner rotation) but couldn't verify them in the simulation
  • ↓Points out that changing the speed of light isn't physically coherent since it's tied to other constants
  • •Mentions prior art and related works exploring the same concept (Velocity Raptor, MIT game, Mr. Tompkins, Redshift Rendezvous)