Be Digital · Field notes

Two meters for your AI coding tools

Today I looked at two simple tools side by side. Think of the first as a utility meter — it shows what your AI coding assistant is really costing. Think of the second as a weather board — it shows whether the services you rely on are up or having a bad day. Between them they answer the only two questions that matter: what is this costing me, and is it even working right now. Here's what each one does, in plain terms, and how the pair applies to two tools everybody's using — Copilot and Claude.

Quick disclosure, because it's the honest thing to do: a lot of the thinking here comes from Vinny Carpenter, a former colleague of mine from my Northwestern Mutual days. I've since moved on from NM, but I still read his notes — and the two tools below are ones he's written about and built. I'm adding my own spin: how a smaller shop would actually use them to keep an eye on Copilot and Claude.

The cost meter, in plain terms

The first tool is called ccusage. It's a small, free program you run on your own computer. Your AI coding assistant already keeps a log of what it did; ccusage reads that log and adds up what all that usage would cost if you were paying the full, pay-as-you-go price. Nothing leaves your machine — it just tallies what's already there and shows you a simple daily table.

The interesting part isn't the grand total — it's what's hiding underneath. In a piece called "The Subsidy, Itemized," Vinny ran this over 26 days of his own work. On paper it added up to $2,556 — but his actual bill was a $200 monthly subscription. In other words, the company is currently covering most of the real cost to get people comfortable using it. That's normal for new software, but it's risky to bet your whole way of working on a discount that won't last forever.

Two smaller lessons made the tool worth running. First, most of that "usage" wasn't the assistant doing brand-new thinking — it was re-reading things it had already seen, which is far cheaper (a cached copy instead of retyping the page). That's why a scary-looking number cost so little. Second, for a few days he switched to the newest, most powerful — and most expensive — model, and you could see the bill jump for work that wasn't always better. Then that model suddenly disappeared: the government restricted it and the company had to switch it off for everyone, overnight. His work had to keep going without it.

So the number that actually matters isn't "how much did we use." It's "did we get something worth keeping, and could we still do it if our favorite tool went away." The meter doesn't answer that for you — but you can't manage what you can't see, and this makes the spending visible.

The weather board, in plain terms

The second tool answers the question the meter can't: is the service even up? It's called Barometer, and it works like a weather board for the internet. Every few minutes it checks the official "are we healthy?" page of nine big services — Amazon, Microsoft, Google, Cloudflare, GitHub, OpenAI, Anthropic (the maker of Claude), Vercel, and DigitalOcean — tidies all their different answers into one easy-to-read format, remembers the history, and shows a live dashboard. If something breaks, it emails you — but only when the status actually changes, so you're not buried in alerts. And if it can't get a clear answer, it honestly says "unknown" instead of pretending everything is fine.

That's the other half of the picture. The cost meter tells you what your usage cost; the weather board tells you whether the thing you depend on was actually reachable when you needed it.

How this applies to Copilot and Claude

Put the two together and you have a simple way to keep an eye on any AI coding tool: watch the bill, and watch the uptime. Two quick examples.

Claude

• Is it up? Claude's maker, Anthropic, is already on the weather board. You'd get a heads-up the moment it starts having problems.
• What's it costing? Point the cost meter at Claude's logs to see the daily spend — and how much was cheap re-reading versus expensive new work.
• What if it disappears? That real "the model got switched off overnight" story is the reminder to always have a backup you can fall back to, so a bad Friday doesn't stop your Monday.

GitHub Copilot

• Is it up? Copilot runs on GitHub, and GitHub is already on the weather board — so you're covered without doing anything extra.
• What's it costing? The same cost meter can read Copilot's logs; on a company plan, you compare that with the official billing so the two agree. This is the direct fix for the "we used up half our credits in two days" panic — now you can actually see where it went.
• Can you switch if you need to? Keep the important rules inside the project itself, so the way your team works keeps working even if you swap the underlying AI for a different one later.

The short checklist behind both gauges

Neither tool decides for you. They just keep you honest before you talk yourself into a story. Five plain questions worth asking:

Read both gauges. One tells you what the AI cost; the other tells you whether it was there when you reached for it. What matters in the end isn't how much you used — it's whether you got something worth keeping, and whether you could still get it if your favorite tool vanished tomorrow.