Top Stories

Ask HN: Add flag for AI-generated articles

721 points · news.ycombinator.com

An Ask HN thread proposing that HN add an explicit flag for AI-generated articles rocketed to the top of the front page, a sign of how raw the community’s nerves are about slop flooding the feed. The debate isn’t just about labeling — it’s about how you’d even detect machine-written content reliably, and whether flagging punishes thoughtful human-AI collaboration alongside low-effort spam. It’s a nicely meta moment: the site’s readers openly arguing about the same content-authenticity problem the rest of the web is quietly drowning in.

Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k

604 points · systima.ai

A wire-level teardown comparing coding agents found Claude Code shipping roughly 33,000 tokens of system prompt and tool scaffolding before it even looks at your actual request, versus about 7,000 for OpenCode. For anyone paying per token or watching latency, that overhead is real money and real milliseconds. The HN crowd loves this kind of receipts-on-the-table analysis, and the thread quickly turned into a broader argument about how much invisible context these agent frameworks quietly stuff into every call.

What xAI’s Grok build CLI sends to xAI: A wire-level analysis

479 points · gist.github.com

Continuing the week’s theme of pointing packet sniffers at AI dev tools, this analysis dissects exactly what xAI’s Grok build CLI transmits back home. Findings about telemetry, code snippets, or environment details being phoned upstream always light up HN, where developers are increasingly wary of handing their private repos to a coding assistant that reports back to the mothership. It’s part of a growing genre: independent researchers auditing what your AI tools actually do on the network.

Zig Creator Calls Spade a Spade, Anthropic Blows Smoke

322 points · raymyers.org

A pointed commentary piece wading into a public spat between prominent developers and Anthropic over messaging around AI coding tools. The post frames one camp as being blunt about tradeoffs while accusing Anthropic of marketing spin. Whatever your read on the merits, it captures the current tension between the builders of editors and languages and the AI labs whose models increasingly live inside those tools — and HN, home turf for exactly these people, showed up to argue.

Vint Cerf, “father of the Internet,” is retiring

322 points · techcrunch.com

Vint Cerf, co-designer of TCP/IP and one of the handful of people who can legitimately be called an architect of the internet, is stepping back. For HN it’s a moment of genuine reverence — Cerf’s protocols are the plumbing under everything the community builds. The thread filled with anecdotes, appreciation, and the occasional reflection on how the open, interoperable internet he championed has fared against the walled gardens of today.

GhostLock: a stack-UAF that has existed in all Linux distributions for 15 years

290 points · nebusec.ai

Security researchers disclosed GhostLock, a stack use-after-free bug they say has quietly lived in every Linux distribution for about fifteen years. Long-dormant, deeply-buried vulnerabilities are catnip for HN, both for the technical writeup of how it went unnoticed and for the uncomfortable reminder of how much foundational code runs on trust and inertia. Expect the comments to dig into exploitability, the disclosure timeline, and how something this old slipped past everyone.

LARP – Revenue infrastructure for serious founders

254 points · larp.website

A cheekily-named startup pitching “revenue infrastructure” for founders drew a crowd, partly for the product and partly for the branding audacity of calling itself LARP. Whenever a new tool promises to handle billing, revenue tracking, and the unglamorous plumbing of getting paid, HN’s founders show up to compare it against Stripe, ChartMogul, and whatever they’ve duct-taped together themselves. The thread is equal parts product critique and startup-naming comedy.

Migrating a production AI agent to GPT-5.6: 2.2x faster, 27% cheaper

213 points · ploy.ai

Fresh on the heels of OpenAI’s GPT-5.6 release, a team published concrete numbers from moving a production agent over: 2.2x faster and 27% cheaper. Real migration writeups with actual latency and cost deltas are far more useful than launch-day benchmarks, and HN rewards them accordingly. The discussion gets into where the gains actually come from, what broke during the switch, and whether the savings hold up under real traffic.

Why write code in 2026

170 points · softwaredoug.com

A reflective essay on why a developer would still hand-write code in an era of capable AI assistants — arguing that the act of writing remains where understanding, design, and judgment actually get formed. It’s the kind of thoughtful counterweight to “just let the model do it” that resonates with HN’s craft-minded contingent. The comments split predictably between those who see coding as increasingly a supervisory role and those who insist the keyboard time is the whole point.