This week's two pieces ended up forming a pair, by accident more than design. One is about a change to how Claude is billed, the other is about a habit worth picking up before the change lands. Together they're really about the same thing: where your AI quota actually goes, and what to do about it.
The first piece, Anthropic metered the agentic bit, and it mostly helps, is about a change that goes live on 15 June 2026. Anthropic is splitting agentic usage off from the main subscription rate limits and onto a separate metered credit. Pro gets 20 a month, Max 5x gets 100, Max 20x gets $200. When that credit runs out you either stop or pay overage at API rates. It's a sensible change in most respects, but it makes the cost of headless agentic work newly visible. If you run a lot of Claude Code, GitHub Actions, or third-party agent tooling, the number on that page next month will be a useful one to look at.
The second piece, A kaizen review of your AI conversations, is the habit I'd pick up between now and 15 June. Claude Code keeps a full record of your conversations on your own machine, in ~/.claude. You can point Claude at it and ask it to look back through what you've actually been doing, picking out the patterns that are costing you. Some of what comes back will be context you keep pasting in. Other patterns will be the same request asked four different ways, or tasks the assistant rediscovered last week because nobody wrote them down. Each one is a candidate for a small bit of infrastructure (a script, a CLAUDE.md entry, a slash command) so the cost is paid once rather than every time. Find the rule, don't run the rule.
If you read just one this week, the kaizen piece is the more useful one for everyone. The metering piece is the why-now.
I'll be back in the inbox next Saturday. Until then.