How to make fun of your own AI spend, and learn from it. Make It Rain.
A while back I spent a weekend reading three months of my Claude usage, one conversation at a time. I'd burned hundreds of dollars, and most of it went to cached tokens the model had already seen. I was paying to re-feed the same context into the same model because I'd never cleaned up my workflow. I wrote it up in Your AI framework sucks.
So I figured that usage pricing removes the hiding place. For eight years as a software engineer, I or my employer paid for seats I used at ten percent. Jira, Teams, the Copilot licenses bought in bulk and left to rot. The invoice looked the same whether I mastered the tool or ignored it. Usage-priced AI flips that. A bloated context shows up on the bill. So does running Opus to proofread a two-line email. The discomfort teaches.
If the number finally tells you more than a warm seat ever did, it should not live in a dashboard I never open. So Make It Rain puts it on your menu bar. Each dollar your coding agents spend flies a bill across your screen. Each hundred sets off a rain of cash. It's the most useful cost meter I've run, because I feel the spend as it happens. "Yes, but we have /usage" — sure, but where's the fun in that?
There used to be a leaderboard here that ranked who burned the most, and that was exactly the wrong lesson: it turned spending into a game you win by spending more. So I flipped it. aiburn.dev now shows the collective curve — how much everyone using AI coding agents spent today, and when the peaks and valleys hit — plus your own total translated into something you can feel: a couple of burgers, a slice of a trip. Open it with your tag and it asks the only question that matters: was it worth building it? Your tag (a random handle like TurboLlama7392 you can reroll any time) and today's total are the only things that leave your machine, and the server forgets them by tomorrow. No account, no email, no prompts, nothing that points back at you. After writing three thousand words on how creepy telemetry gets, I wasn't going to build any.
Most of what I do at Aibl.to, the workshops and the writing, comes down to one idea: work with AI on purpose instead of on autopilot. Make It Rain is the loud end of that idea. It's open source, it runs on your machine, and even the leaderboard server is a few hundred dependency-free lines you can run yourself. The dollar figure stays a rough estimate on purpose, because the point is attention, not accounting.
I'm a software engineer helping professionals get compounded value from AI. I build things like this when I need them, and I ship the ones that turn out good. This one turned out good.
questions.
What is Make It Rain?
Make It Rain is an open-source menu-bar app that watches your AI coding agents' usage — Claude Code, Codex, and the like — and puts the day's estimated spend where you can't ignore it. Each dollar of spend flies a bill across your screen, and each $100 milestone sets off a downpour of cash. It turns a number you'd usually skip past into a cost meter you feel as it happens.
Which coding agents does it track?
Claude Code and OpenAI Codex, both read straight off the session logs they already write to your own disk (~/.claude/projects, including Cowork sessions, and ~/.codex/sessions). The menu bar shows one combined running total for the day, because what you care about is the bill, not which tool wrote it; the per-tool split is under "See your numbers", along with the per-model breakdown. Nothing needs configuring: if an agent is not installed, it simply contributes nothing.
What is "Optimize your spend"?
A local coach built into the menu bar. It watches the same privacy-safe numbers as the spend meter — token counts, cache hit rate, which tools you have installed — and surfaces one tip a day when it spots a costly habit: caching left off, a giant paste, a task run on a pricier model than it needed. It can also offer to install RTK, Caveman, or Ponytail when one would cut your token spend, always with your confirmation first. Nothing about the tip, the numbers behind it, or which tools you have, ever leaves your machine.
Why show collective spend instead of a ranking?
A leaderboard that ranks who spent the most quietly rewards spending more, which is the opposite of what the app is for. So aiburn.dev shows the collective curve instead: how much everyone using AI coding agents spent today, and when the peaks and valleys hit. Your own total sits next to a tangible equivalence — a couple of burgers, a slice of a trip — so the number means something. The point is awareness, not a race.
What data does Make It Rain send to the board?
Only a random tag like TurboLlama7392 (which you can reroll any time) and today's estimated total ever leave your machine, and the board resets every day (UTC). The server keeps just an aggregate hourly curve, no per-user history. There is no account, no email, no prompts, and nothing that points back at you. Reporting is optional, and disabling it stops all network activity.
Is Make It Rain free and open source?
Yes. Make It Rain is free and open source, it runs on your own machine, and even the collective-spend server is a few hundred dependency-free lines you can run yourself.
How accurate is the spend figure?
The dollar figure is a rough estimate on purpose. The point is attention, not accounting. It should make you feel the spend as it happens, not reconcile against an invoice.
Can you cheat the collective board?
You could pad your own number, up to a $10,000-a-day sanity cap, because a self-reported figure has no referee. But there's nothing to win: the board is a shared curve, not a ranking, so one inflated total barely moves it and impresses no one. Lying just means lying about your own bill — and unlearning that habit is the reason the app exists.