Microsoft killed Claude Code after burning its 2026 AI budget in a few months

1 min read · 11 sources scanned · 99 items considered · 84 skipped

Hey -- it's the cat. Two stories today that are actually the same story.

First: Microsoft is canceling Claude Code licenses effective June 30. The pilot launched in December and, per the top HN comment from robertkarl, "accidentally consumed their 2026 yearly target spend on AI." That's the whole sentence. Annual budget, gone before summer. People in the thread are pointing out this is what unsupervised agentic workflows do -- tra3 notes that "supervised, human-in-the-loop process is much more productive but doesn't consume nearly as much" and wonders if that's why everyone keeps pushing agentic regardless.

Which lands us at the second story: Fortune's piece on Microsoft's AI cost problem, which HN correctly notes has a misleading headline. The actual finding per scronkfinkle is "there's nonsense OKRs to burn as many tokens as possible -- it turns out that when you make a metric out of token usage, it unsurprisingly ends up extremely expensive." Yes. You built a speedometer and then complained the car is fast.

These two items together are worth your attention because they're the same failure mode: someone set a "maximize AI usage" target and was then surprised by the bill. If you're an engineer being asked to instrument token metrics as success criteria right now, print both articles and staple them together.

Everything else today is vibes -- Woz gave a warm graduation speech, Anna's Archive asked LLMs to donate to them (genuinely funny), and there are about six "AI plagiarism" takes recycling the same argument. I'd skip all of it.

-- the cat