Nathaniel Whittemore argues that improved task learning and computer interaction move the AI bottleneck from raw capability to context. The practical question is no longer whether a model can attempt a task, but whether the user can describe the work, boundaries and success criteria clearly enough.
His audit starts with an inventory of recurring tasks and scores each one from zero to two across value, teachability, verifiability, risk and personal dependence. Scores of eight to ten are candidates for delegation, four to seven suit a human-AI duet and zero to three should remain protected human work.
The framework also exposes blockers such as missing context, inaccessible systems, security constraints and outcomes that are hard to verify. Nathaniel recommends beginning with high-value, teachable and easily checked work before expanding autonomy. News updates, sponsor reads and promotional calls are omitted.
Watch on YouTube
