Original longer reads, opinion, and editor's notes from nwslyr on AI, technology, and the systems beneath the daily briefing.
AI-linked tech giants are in bear market territory as investors reprice enormous AI spending, higher rates, and uncertain returns.
OpenAI's GPT-5.6 preview is becoming the first test of voluntary federal model review, with access cleared customer by customer.
AI agents are turning software engineering from hand-written code into agent-directed loops, shrinking the old on-ramp for new developers.
Meta will use your off-site activity to shape your feed and Meta AI, not just ads. The stronger opt-out is gone. What changed, and how to opt out.
A US export-control directive forced Anthropic to shut off Fable 5 and Mythos 5 overnight. What the government did, what Anthropic said, and why it matters.
Anthropic's Claude Fable 5 rerouted my live security audit to Opus 4.8 mid-run. The silent fallback, forced 30-day retention, and pricing cliff, explained.
There is a thing happening in AI coding circles right now where a simple idea gets dressed up like a private language.
AI conversations get buried in buzzwords fast. Here is a simple dictionary for the terms you keep seeing, without the fog machine.
Paul Graham and Jack O'Brien say never let AI write your words. The lie isn't the tool, though — it's signing what you didn't check.
A Harvard Business Review working paper finds that managers who treat AI agents as employees catch fewer errors and shift responsibility onto the system. The fix is pro-control, not anti-agent: give every agent an owner, a permission boundary, an audit trail, and an off switch.
A copy-paste prompt walkthrough for using Claude's Cowork desktop control to remove the built-in ads, suggestions, and nags from Windows 11 — including a registry file that survives feature updates.
Hackers reportedly used Meta's AI support chatbot to take over Instagram accounts by asking it to change account recovery details, exposing the risk of AI tools with sensitive permissions.
Claude has started telling people to go to sleep in the middle of the workday.
A defense of AI as the first expert with no ego to feed: sometimes wrong, rarely smug, and infinitely more patient than the old help desk and forum era.
After 5.68 billion tracked AI tokens and a 13-hour session, a non-programmer explains why model choice and cost discipline matter.
AI adoption is real, but the gap is widening between products that redesign work around AI and products that simply bolt on a chatbot.
On day one, I pointed Opus 4.8 at a project GPT-5.5 had been building and asked four open questions. It read the full codebase, ran every test, checked its own claims, and found the gaps that mattered most. The same ones GPT-5.5 had been asked to find and couldn't.
Claude Opus 4.8 is a measured upgrade at the same price as 4.7, but its effort controls and cheaper fast mode matter for builders trying to manage AI costs at scale.
If your first AI experience was robotic, wrong, or confidently useless, the problem may have been the tool, the tier, the prompt, or just a bad day for the model.
AI pricing is becoming harder to read as tokenizer changes, reasoning budgets, deprecations, and frontier model rates reshape cost per completed task.
Sam Altman says AI won't cause a jobs apocalypse. One day earlier, Anthropic's Chris Olah warned of displacement 'at very large scale' from the Vatican. The data is split.
Pope Leo XIV's first encyclical, Magnifica Humanitas, turns AI governance into a moral question, with Chris Olah and Anthropic unexpectedly at the center of the conversation.
Claude Opus 4.6 is not being deprecated. Its retirement floor is February 2027. The operational complaints are real, but they trace to infrastructure incidents and design tradeoffs, not a secret downgrade.
Google's Gemini 3 pricing moves the affordable API floor up by 2.5x to 5x, challenging the unit economics that made Flash and Flash Lite useful for production workloads.
Huxe, the personalized AI audio briefing app from ex-NotebookLM creators, is shutting down less than a year after launch. Its closure shows how quickly standalone AI app windows can be absorbed by platform features.
Sony's PlayStation-owned tournament platform Repeat.gg is closing, and the shutdown reads like a quiet signal about PlayStation's retreat from esports and live-service sprawl.
A six-minute Codex cleanup removed Windows widget and feed bloat, blocked the consumer-content pipeline, and showed why coding agents can beat third-party debloating utilities.
OpenAI's Codex Mobile brings phone-based control to all ChatGPT plans, while Anthropic got there first with Claude Code Remote Control.
Anthropic published the document that governs Claude's behavior. I asked Claude to help critique it. The system is better than the document.
A ten-section philosophical and governance critique of Anthropic's published framework for AI behavior.