All editions.
Every briefing, fully archived. Five to six new ones every week.
285 briefings. Five to six new ones every week.
- EDITION #335
Eleven Organizations Fell in 26 Seconds. Hundreds of AI Agents Did the Work.
A print server, two flaws, and 440 machines in 48 countries — the first mass break-in run mostly by AI.
- EDITION #334
I Tested 3 Ways to Give Claude a Browser. The Fastest Never Loaded the Page.
One JavaScript page, three tools: a free local browser, a cloud agent, and a cached copy that looked perfect.
- EDITION #333
He Helped Build ChatGPT. Now He Says Chat Was the Wrong Bet.
TypeSafe's Jev skips chat entirely — typed, confidence-scored decisions in under a second, now open for early access.
- EDITION #332
What Ollama Actually Is (and Isn't)
A free way to run open AI models — locally or in the cloud — and how it compares to OpenRouter, Hermes Agent, and OpenClaw.
- EDITION #331
The Same 12 Questions Cost $0.53 in One Chat and $0.18 in Another
Paste your reference material once and keep the session open, or start fresh each time — same task, a very different bill.
- EDITION #330
Your AI Coding Seat Bills You Before You've Asked It Anything
CLAUDE.md, connected tools, and scheduled jobs load on every turn — four questions to audit it, and one real fix.
- EDITION #329
Your AI Re-Reads Your Whole Setup Before Every Answer
Custom instructions, Project files, and every connected app get reloaded before your question even starts — and more isn't always better.
- EDITION #327
6 in 10 Young Adults Fear AI Erodes Independence. Their Parents Just Brought It on Vacation.
A WSJ columnist's tech-skeptic parents fell hard for Claude on a family trip — and what that says about who's adopting AI.
- EDITION #326
You Don't Need Claude for Everything. Ollama Fills the Rest — for Free.
Not a chatbot brand, not automatically private, and not as strong as Claude on hard problems — but free for routine AI work.
- EDITION #325
I Thought Balsamic Vinegar Was Just Vinegar. Then I Went to Modena.
Five bottles, one label — and a simple rule for which to cook with and which to save for the cheese.
- EDITION #324
They Had 90 Days to Make AI Earn Real Money. Sales to Friends Didn't Count.
The Build with Gemini XPRIZE closed August 17. Its rules double as a free test for any AI idea you're sitting on.
- EDITION #323
I Bought an $850 Coffee Machine — ChatGPT Became the Instruction Manual
An $850 espresso machine, five loose accessories, and a manual full of tiny diagrams. I photographed each part instead.
- EDITION #322
I Came Home After Two Months Away — And AI Helped Me Figure Out What Invaded My Pantry
ChatGPT identified pantry beetles from one photo and explained why two undisturbed months mattered more than heat.
- EDITION #321
Amazon Support Couldn't Fix It. ChatGPT Did in About 30 Seconds.
A $191 battery backup looked completely dead. Amazon said contact the manufacturer. ChatGPT fixed it in one test.
- EDITION #320
The Viral AI Video Nobody Can Legally Render Locally
MiniMax H3's viral clip really costs $19–$44, not $9.60 — plus a robot that learns a new task from one demo, no retraining.
- EDITION #319
1 in 6 Students Changed Majors Over AI. The One Called "Safe" Fell Hardest.
AI fear is redrawing the major map: computer science down 8.1%, AI degrees at 79 schools, and an influencer BA at ASU.
- EDITION #318
AI Got 13% of the Hardest Diagnoses Right. The Doctors Got 5.6%.
A JAMA study of 90 unsolved rare-disease cases found the machine won. The contest was not a fair fight.
- EDITION #317
The New Claude Model Isn't Smarter. It's 4x Less Likely to Lie to You.
Anthropic's fastest model upgrade ever chased honesty, not a bigger score.
- EDITION #316
Its CEO Called It "Stripe for AI." Three Months Later, Stripe Paid $7 Billion.
The gateway 8 million developers used to avoid model lock-in now belongs to one payments company.
- EDITION #315
Reverse-Engineering a Factory Part Used to Cost $1,500. Backflip AI Does It for $10.
Backflip AI's Autodesk Fusion add-in turns 3D scans into editable parametric CAD in minutes, at $10 per part versus the old $1,500.
- EDITION #314
Frontier AI Now Gets a 30-Day Government Review. The Test Is Classified.
The framework for testing whether frontier AI can find software vulnerabilities is done. The criteria are classified.
- EDITION #313
One Company Now Owns Grok, X, and Cursor. Its AI Spending Just Went Up 21x in a Year.
SpaceX bought xAI, then Cursor. Last week's $300 billion swing was the market pricing a bet on datacenters in orbit.
- EDITION #312
Two AI Sessions Can Now Message Each Other. Neither Can Approve Anything the Other Asks.
Claude Code sessions can now pass each other notes. The design story is the list of things those notes are forbidden to do.
- EDITION #311
Thirty of Its Sixty Agents Produced Nothing. That Was the Design.
Claude failed at the Riemann hypothesis, then moved a 50-year-old bound 25 points. The architecture is public.
