The commit message convention — write why, not what — is a good idea that fails in practice for structural reasons. Articulating reasoning after the fact is hard. The thinking happened during implementation; the message is written after, when context has partially dissolved and the pressure to ship has fully arrived. Even disciplined developers produce commit messages that describe the diff rather than explain the decision. The convention asks humans to do something humans are not good at: reconstruct and articulate reasoning retroactively, consistently, under time pressure, for an audience they cannot anticipate. This is a tooling failure, not a discipline failure.
(Code without reasoning is archeology · Agent sessions make reasoning explicit)