What jumped out at me is not the “7 of 8 were unused” headline. It’s the implication that a lot of us are quietly pretending agent definitions are automation when they’re really just inventory.
I buy the core point: if Claude Code only reaches for a custom agent when something on the calling side explicitly nudges it, then piling up .md files is mostly theater. The article’s example makes that painfully clear. One reviewer agent got a little use, the rest were dead weight. That’s not a surprise in itself, but it is a useful reality check for anyone who’s been feeling productive after creating a neat little agent zoo.
What I’m less convinced about is the token-cost argument as presented. Yes, more agent definitions injected into the system prompt must cost something. But the article leans pretty hard on “quietly degrading inference quality” without showing a before/after measurement. That might be true, but it also might be hand-wavy. If you’re going to claim prompt bloat is hurting outputs, I’d want to see a concrete regression or at least some rough comparison. Otherwise it’s a plausible fear dressed up like a measured effect.
The more interesting part is the logging mindset. That I actually like. The point isn’t “delete unused agents once.” It’s “build a habit of checking whether your automation is doing anything at all.” That’s a much more mature way to think about AI tooling than the usual vibe, which is basically: define stuff, trust the vibes, and assume the system is smarter than it is.
I also think the article accidentally lands on a bigger truth about Claude Code specifically: names and instructions inside an agent file are not magic. If you don’t build the invocation path, you don’t have automation. You have a note to self. That distinction matters a lot, and people keep blur-ring it. The phrase “MUST BE USED” inside a definition sounds comforting, but it doesn’t create enforcement by itself. The system won’t rescue you from a missing control plane.
What I’d actually do, reading this, is keep far fewer agent definitions and make every one earn its place. Then I’d build the usage report, but I’d also track whether the few active agents are producing better outcomes, not just more calls. Call count is a good smoke alarm. It is not proof that the thing is useful.
Reference: 7 of My 8 Claude Code Agents Had Zero Calls in 30 Days: Finding Dead Agents Automatically
