ls -t reports/
Archive
Everything the publication has filed — 81 pieces, newest first.
Weekly issues 11
- №11 The Models Are Interchangeable Now. Stripe Paid $7B for the Switch.
- №10 AI Aced the Test This Week. Nobody Felt Better.
- №9 The Rules Were Written Down. The Agents Ignored Them.
- №8 The Whole Industry Signed the Letter. The Two Model Labs Didn't.
- №7 Nobody Sold the Model This Week
- №6 The Model Got Cheap. Watch Where the Money Went.
- №5 The Kill Switch Lasted 19 Days. The Guest List Is Forever.
- №4 The Frontier Got a Guest List
- №3 You Can Switch Off a Model. You Can't Switch Off the Capability.
- №2 Safety Was the Moat. This Week It Became the Weapon.
- №1 The Week the Bill Came Due
Deep dives & specials 70
- 18 Aug The Moat Is Made of the Thing They Can Never Show You
- 17 Aug The Box Your Agent Runs In Is a Commodity. The Kit Around It Isn't.
- 16 Aug Your Agent Doesn't Need More Tools. It Needs to Write Code That Calls Them.
- 15 Aug Your Model Didn't Get Worse. It Stopped Asking.
- 14 Aug SQLite Tests Every Branch. The Bug Hid in the Order They Ran.
- 13 Aug You Can Run Fifteen Claudes. The One That Can't Scale Is You.
- 12 Aug They Didn't Hide the Reasoning. They Encrypted It and Handed It Back.
- 10 Aug The Codebases Banning AI Aren't Worried About Quality
- 9 Aug AI Levels the Field Where the Errors Are Cheap. It Widens It Where They Aren't.
- 8 Aug A GPU Swaps Models in Milliseconds. AMD Just Bought a Chip That Takes Two Months.
- 7 Aug Your Rules Now Travel Between Coding Agents. Your Guardrails Don't.
- 6 Aug Your Context Window Is 60% Tool Output. A PostToolUse Hook Takes It Back.
- 5 Aug The Top of the Leaderboard Is Inside the Margin of Error
- 4 Aug You Can Fit a 70B in 4GB. You Still Have to Move Every Byte.
- 3 Aug A Written Policy Is Not a Control
- 2 Aug You Can Mark the Image. You Can't Make the Mark Survive.
- 1 Aug MCP Deleted the Session. Your Server Just Became a Lambda.
- 31 Jul Claude Didn't Break Cryptography. It Made Cryptanalysis Cheap.
- 30 Jul A Subagent Buys You Context, Not Speed
- 28 Jul Bun's Rewrite Only Runs in One Direction
- 27 Jul The Ad Used to Sit Beside the Answer. Now It Wants to Be the Answer.
- 26 Jul Delete the Rules You Wrote for a Dumber Model
- 25 Jul The AI Buildout Isn't Cash-Funded. The Debt Just Isn't on the Balance Sheet.
- 24 Jul A Counterexample Checks in One Pass. That's Why the Machine Found It.
- 23 Jul The Kernel Approves Commands Better Than You Do
- 22 Jul The Loop Deskills the Human It Depends On
- 21 Jul The Window Says 1,050,000. The Bill Starts at 272,000.
- 20 Jul The Cheapest Adequate Model Is a Great Strategy Until You Define "Adequate"
- 18 Jul Eleven More Points of Accuracy Cost 172× the Compute
- 17 Jul It Uploaded the Files You Told It Not to Open
- 16 Jul Your Session Starts 33,000 Tokens in the Hole. Most of That Is Tools You Won't Use.
- 15 Jul 2.12%: The Number That Ends the Speech-to-Text Round-Trip
- 14 Jul The List Price Is Per Token. Your Bill Is Per File.
- 13 Jul Nobody Chose to Run Chinese AI. The Invoice Did.
- 12 Jul Nvidia Isn't the Domino. The Neocloud It Financed Is.
- 11 Jul Stop Handing Your Agent a Screenshot of Your Own App
- 10 Jul The Moat Isn't the Model. It's the Accept Button.
- 9 Jul A Skill Costs You One Sentence Until You Use It
- 8 Jul Your Agent Pushed That Commit While You Weren't Looking
- 7 Jul The AI Ran the Ransomware. A Human Left the Door Unlocked.
- 6 Jul Your Next Customer Is an Agent With a Wallet
- 5 Jul Your Tool Calls Broke on the Upgrade. It's the Schema, Not the Model.
- 4 Jul The Agent Reading Your Docs Won't Scroll. Ship the Endpoint, Not the Magic File.
- 4 Jul The 60% Discount for Imaging Your Code Is a Lossy Compression Bet
- 3 Jul A Watermark You Read With a Z-Test Washes Out With a Paraphrase
- 2 Jul Your Deny Rules Match Strings. Your Real Guardrail Is a Hook.
- 1 Jul A Marker You Can Delete Is Not a Surveillance Backdoor
- 30 Jun The 128K Window You Bought Is a 64K Window. Plan Accordingly.
- 29 Jun Nvidia Keeps 70 Cents of Every Dollar. That's Why OpenAI Built a Chip.
- 28 Jun The Price Cut Wasn't For You
- 27 Jun Distillation Without Logits: Why It Took 28.8 Million Queries
- 26 Jun Your Agent Will Retry That Write. Make It Safe to Run Twice.
- 25 Jun Your 200K Window Has a 120K Speed Limit
- 24 Jun Same Model, Faster Tokens: The Arithmetic of Speculative Decoding
- 23 Jun Your Agents Don't Need a New Git. They Need to Stop Sharing One Checkout.
- 22 Jun Portability Is Not a Purchase. It's an Eval Discipline.
- 21 Jun The Model Has 744 Billion Parameters. You Pay for 40 Billion.
- 20 Jun Compaction Is a Lossy Save. Choose When It Fires.
- 19 Jun "Agent" Is a Control-Flow Decision, Not a Product
- 18 Jun Prompt Caching Pays 90% Off — If You Win the Bet
- 17 Jun The Coding Model You Can Run Isn't the One That Wins
- 16 Jun The Open Model You're Running Is a Binary, Not a Source
- 15 Jun You Cannot Export-Control a Model
- 13 Jun Fan-Out Has a Token Bill, and You Sign It
- 12 Jun The Benchmark Score Is Not the Capability
- 11 Jun LLMOps Is a Feature, Not a Company
- 10 Jun The Trust Stack Was Built for Human-Speed Software
- 9 Jun The Channel Was the Product
- 8 Jun Autonomy Shipped Before Its Brakes Did
- 7 Jun The Meter Is the Confession