This is a working argument that technical leaders experimenting with AI agents *should* show measurable "AI exhaust" (tokens burned, code generated, failed experiments) because the practice of building with agents is fundamentally unsettled. Schipper's claim: unlike the historical pattern where seniority meant less direct coding, senior engineers now need deep firsthand experience with agent failures, tooling, and context management to make good technical decisions—because nobody yet knows whether agents should work autonomously, what code review should look like, how to manage context windows, or how to organize codebases around them. The evidence is empirical (his own experiments with context windows, autonomy, code review gates, QA agents) rather than theoretical. The practical takeaway: if your technical leaders aren't actively breaking agent systems and accumulating visible token/code traces from that work, it's reasonable to question whether they have the grounded judgment to set direction on agent adoption. This isn't about measuring engineers *by* exhaust, but using exhaust as evidence of the kind of hands-on experimentation the role now requires. One sharp question: does this logic hold as agent patterns do stabilize, or is Schipper assuming we're further from consensus than we actually are?