An obituary topped the day: Bob Cringely, who told the personal-computer story to a general audience, has died. The rest of the issue splits AI’s two faces — running large models on a home PC and reining in runaway bills — alongside resurrected old hardware, a war-time energy case, and arguments about trades work.

From a seven-year silence around the Snowden archive to a pocket radar for camera glasses, from long-horizon AI models to a rewritten promise of ad-free streaming, today’s stories ask how technology stays legible and accountable to ordinary people.

As tools get lighter — GET-only exfiltration, transparent image layers, and peer-to-peer model archives — the harder questions move to defaults for privacy, memory, and health, with hardware and math keeping score in the background.

From copyright on trial to passkeys that lock you out, from a coding app that quietly copied whole repos to fines and AI intel failures, the question is not just what technology can do, but whether we can verify, refuse, and own it.

DuckDB’s move to AWS tests the promise of open governance, a next-gen Qwen architecture tries to make long context cheaper, and stories about searchable executables, supply-chain concentration, license boundaries, visa pauses, and offline maps show how infrastructure choices ripple into everyday life.

A day of trade-offs between trust and convenience: marking words versus choosing them well, putting capable models on laptops while a central code forge wobbles, and redrawing the lines around pricing, consent, and attention.

Open models push coding and security forward while collaboration pain grows, Australia turns surplus solar into cheaper evenings with home batteries, and Google makes encrypted inference more practical.

Tailscale chased a 16-year SQLite race into the open, while AI kept rewriting how engineers and mathematicians work. License-plate history searches, rage-bait payouts, and a font that covers Manhattan filled in the public side.