Top Stories

John Ternus to become Apple CEO

1314 points · apple.com

The runaway #1 story on HN today. Apple announced that Tim Cook will transition to Executive Chairman and John Ternus — long-time SVP of Hardware Engineering and the on-stage face of recent Mac and iPhone launches — will take over as CEO. Ternus was widely considered the heir apparent inside Apple, and the handoff ends a decade-plus of leadership speculation. The 657-comment thread is what you’d expect: half the room is relitigating Cook’s tenure (record valuation, services pivot, AI stumbles), and the other half is sizing up Ternus as a hardware-first CEO facing a software-and-AI-first decade. Expectations for what the Ternus era means for Vision Pro, Apple Silicon, and Apple’s still-patchy AI story are running high.

All phones sold in the EU to have replaceable batteries from 2027

1012 points · theolivepress.es

The EU has confirmed the long-trailed rule: from 2027, every phone and tablet sold in the bloc must have a user-replaceable battery. Manufacturers have been fighting this for years, but the regulation is now locked in — with fines tied to enforcement. 826 comments range from “finally, the right-to-repair wins” to “this is going to produce the ugliest iPhone ever,” plus a long sub-thread from people who actually build consumer hardware explaining what waterproofing, structural rigidity, and replaceable cells look like when you have to have all three. Expect this to ripple globally, the way USB-C did.

Kimi K2.6: Advancing open-source coding

591 points · kimi.com

Moonshot AI’s Kimi team drops K2.6, the latest in their open-weight code model line, claiming state-of-the-art open-source performance on SWE-Bench Verified and a material jump on long-context repo comprehension. The 303-comment thread is a healthy mix of benchmarks-are-broken skepticism and actually-shipping developers sharing their Cursor/Continue setups running Kimi locally. Combined with the Qwen drop (below), today is a very good day for anyone who’d rather not depend on a hosted frontier lab for their coding agent.

Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

557 points · qwen.ai

Alibaba’s Qwen team previews Qwen3.6-Max, which they’re positioning as a frontier-tier open-weights model — stronger reasoning, longer context, and meaningful gains on multilingual and math benchmarks. 292 comments dig into methodology, compare head-to-head with Opus 4.7 and Kimi K2.6, and argue (as always) about whether “open weights” with restrictive licenses really counts as open. The subtext of the thread: the open-source frontier is now a three-way race between Qwen, Kimi, and Llama, and all three shipped meaningful gains inside of a month.

Atlassian enables default data collection to train AI

512 points · letsdatascience.com

Atlassian has quietly flipped the default on customer data — Jira tickets, Confluence pages, Bitbucket repos — to “opt-in” for AI training. The post walks through the admin-console screens where enterprise customers can (and probably should) turn this off. 116 comments are a tour of enterprise trust decay: frustrated admins, legal teams rewriting DPAs, and a long thread on why “default on for training” has become the standard dark pattern of 2026. Expect procurement meetings to get interesting this week.

ggsql: A Grammar of Graphics for SQL

370 points · posit.co

Posit (née RStudio) releases an alpha of ggsql, a declarative graphics layer that compiles Grammar of Graphics specs directly to SQL — meaning your charts run where your data lives, with no intermediate dataframe. 76 comments are a surprisingly thoughtful debate on what you gain (scale, speed, reproducibility) and what you lose (flexibility, interactivity) by pushing visualization into the database. For the ggplot2 faithful, this is the kind of Posit announcement that makes the whole ecosystem feel alive again.

Deezer says 44% of songs uploaded to its platform daily are AI-generated

310 points · techcrunch.com

Deezer published internal numbers: 44% of new tracks uploaded daily are AI-generated, up from 18% a year ago. They’ve built a detection model, they disclose AI tracks to users, and they’re adjusting royalty weighting accordingly — but the underlying dynamic (spam floods the long tail, humans compete for the same listening minutes) isn’t going away. 288 comments include musicians, rights-holders, and ML folks arguing about streaming economics, detection accuracy, and whether “AI music” is a category that will even mean anything in two years.

OpenAI ad partner now selling ChatGPT ad placements based on “prompt relevance”

203 points · adweek.com

A leaked StackAdapt deck shows the mechanics of how ads are going to show up inside ChatGPT conversations: bid on intent categories, get surfaced when a user’s prompt semantically matches. This is the quiet shoe-dropping everyone expected once OpenAI signaled ad monetization — but seeing the actual pitch materials, with CPM ranges and targeting taxonomies, is concentrating minds. 89 comments split between “this was always coming” and “the context collapse between assistant and advertiser is a different order of problem than search ads.”

WebUSB Extension for Firefox

209 points · github.com

A third-party extension that finally brings WebUSB to Firefox — the API Mozilla has refused to ship natively for a decade on security grounds. The extension implements WebUSB as a polyfill with user-gated device access, which is the compromise a lot of hardware-focused web developers have been waiting for. 186 comments are a master class in the ongoing WebUSB debate: Chrome users who love it, Mozilla alumni re-explaining why they didn’t, and a growing contingent pointing out that the divergent web platform is a bigger problem than any individual API.