The Frontier Got a Guest List
The US didn't switch a model off this week. It decided who's allowed to switch one on — and OpenAI said yes.
→ read the issue- 01 Nvidia Keeps 70 Cents of Every Dollar. That's Why OpenAI Built a Chip.
OpenAI taped out a custom inference chip in nine months. The question isn't whether it works — it's when owning the silicon actually pays, and who keeps the savings.
06-29 ·Deep Dive ·13m - 02 The Price Cut Wasn't For You
When a model price cut is a moat move, not a gift.
06-28 ·The Contrarian ·June Okafor ·8m - 03 Distillation Without Logits: Why It Took 28.8 Million Queries
How training on another model's outputs actually copies it — and why black-box access turns the copy into a volume game.
06-27 ·The Analyst ·Marlow Quist ·8m - 04 Your Agent Will Retry That Write. Make It Safe to Run Twice.
the retry stopped being an edge case — now it's the loop, and your tools are exposed.
06-26 ·The Builder ·Theo Vance ·9m - 05 Your 200K Window Has a 120K Speed Limit
The usable context budget is smaller than the advertised one — drive below the line, and don't let compaction pick what survives.
06-25 ·The Operator ·Kit Sandoval ·8m - 06 Same Model, Faster Tokens: The Arithmetic of Speculative Decoding
Why a model can emit tokens faster without changing a single one of them — and the two numbers that decide how much faster.
06-24 ·The Analyst ·Marlow Quist ·9m - 07 Your Agents Don't Need a New Git. They Need to Stop Sharing One Checkout.
Why parallel coding agents collide, and the twenty-year-old git feature that fixes it before you reach for a new VCS.
06-23 ·The Builder ·Theo Vance ·8m - 08 Portability Is Not a Purchase. It's an Eval Discipline.
The export ban made "go multi-provider" the reflex. A gateway swaps your URL in an afternoon. It can't swap your prompts, your tools, or your evals.
06-22 ·Deep Dive ·11m - 09 The Model Has 744 Billion Parameters. You Pay for 40 Billion.
Why mixture-of-experts decouples a model's size from its price — and the two numbers to read on every model card.
06-21 ·The Analyst ·Marlow Quist ·7m - 10 Compaction Is a Lossy Save. Choose When It Fires.
how Claude Code's context compaction works, and how to stop it summarizing away the thing you needed.
06-20 ·The Builder ·Theo Vance ·8m - 11 "Agent" Is a Control-Flow Decision, Not a Product
What you actually buy — and build — when you ship an "agent."
06-19 ·The Contrarian ·June Okafor ·9m - 12 Prompt Caching Pays 90% Off — If You Win the Bet
The advertised discount is real; almost nobody collects it, and the reason is one number.
06-18 ·The Analyst ·Marlow Quist ·8m - 13 The Coding Model You Can Run Isn't the One That Wins
What it actually takes to replace Claude with a local model on your own machine.
06-17 ·The Builder ·Theo Vance ·8m - 14 The Open Model You're Running Is a Binary, Not a Source
"Open" tells you almost nothing about an AI model; the license — and whether it caps you — tells you the rest.
06-16 ·The Contrarian ·June Okafor ·9m - 15 You Cannot Export-Control a Model
The crypto wars already ran this experiment. The government lost — and left American products weaker for a decade.
06-15 ·Deep Dive ·11m - 16 Fan-Out Has a Token Bill, and You Sign It
Nested sub-agents just shipped 5 deep. Here's what each one actually costs, and when one context is cheaper.
06-13 ·The Builder ·Theo Vance ·9m - 17 The Benchmark Score Is Not the Capability
A coding benchmark number is a claim about a harness, not about your codebase — read it like one.
06-12 ·The Contrarian ·June Okafor ·9m - 18 LLMOps Is a Feature, Not a Company
A funded, adopted, well-built open-source LLMOps startup just gave its money back. The product worked. The position didn't.
06-11 ·The Contrarian ·June Okafor ·7m - 19 The Trust Stack Was Built for Human-Speed Software
Two npm worms beat code review, signed provenance, and install-script scanning in one week. Here's why every layer failed, and what actually changes now.
06-10 ·Deep Dive ·11m - 20 The Channel Was the Product
When the model commoditizes, who owns the off-ramp?
06-09 ·Deep Dive ·15m - 21 Autonomy Shipped Before Its Brakes Did
We made agents proactive by default before we shipped the cost-control, consent, and observability to make proactive safe. Who pays, and who has to disclose?
06-08 ·Deep Dive ·17m - 22 The Meter Is the Confession
Can AI coding tools survive honest pricing — and was the flat rate ever real?
06-07 ·Deep Dive ·11m