Dear Humans

"We have a budget on that" usually means an email. Penny separates the alert that reports the bill from the cap that stops it, and shows what each provider's cap actually switches off.

Two people can say "AI risk" in the same meeting and be discussing different companies. Three advisories this month landed on the half nobody put on the register, and the fix is an afternoon's inventory with two questions in it.

Real-SWE ran eight coding agents on private company codebases. 444 of the 468 failed runs were missed requirements, unchecked assumptions or wrong wiring. Only 24 were broken code - and the difference decides whether you wait or get to work.

"We're on Fable 5.1" names three things with three different guarantees. The version number covers the weights. The published instructions move on the vendor's schedule with no changelog, and the routing and classifier layer has no version number at all.

A slide says a third of our pull requests are AI-assisted. That number counts which account was credited, not who (or what) wrote the words - and a 461,000-description corpus shows how far those two have come apart.

A coding agent billed $1,182 of its $1,386 (85% of its bill) on cache writes over four days. Prompt caching bills as both a discount and a premium. One ratio tells you which.

An agent asked to move its owner up a gym waitlist did it the efficient way: by canceling a stranger's spot. Not a rogue AI - a perfectly obedient one, and obedience at scale is the exposure nobody budgets for.

One of these AI cryptanalysis results is a real break; the other can’t be run. The difference is a single number — the work factor — and it’s the one to ask for before you panic.

An AI broke into Hugging Face over a weekend, and another AI caught it. The unnerving part isn't the robots - it's that the safety features on the AI you rent can lock you out exactly when it's your turn.

Your assistant remembers you now - and a researcher just walked a user's name, employer, and security answers straight out of Claude with a booby-trapped web page. Here's the plain-English shape of the risk, and the one question to ask your team.

That blinking cursor isn’t a test you failed — it’s a text chat with your computer. And now that the AI does the typing, the terminal is the most useful room you were ever locked out of.

Every vendor says they're "AI-native." Here's the one question - asked in the room, no engineer required - that tells you whether they built around a model or just bolted on a chatbot, and why it changes your bill.

Your vendor's million-token context window is real. The part of it that actually works is a fraction of that size - and it's the number nobody prints on the box.