The discussion curated by Simon Willison points to a practical tension in the rush toward agentic workflows. While Udell's proposed shift in language is compelling, the real test lies in implementation. In our view, simply inviting an 'agent onto the team' does not automatically solve the black-box problem or guarantee reviewable outputs; it merely describes an aspiration.
The industry's current trajectory appears to favor automation and scale over the meticulous, human-supervised processes Udell champions. The key second-order effect to watch is whether this linguistic reframing gains traction among tool builders, influencing platform design to enforce reviewability, or if it remains a niche critique as autonomous agent deployment accelerates.