The email sounds considerate. It explains the new team schedule clearly and ends with an invitation to ask questions. Yet the manager who generated it may still have missed a previous commitment that makes the new hours difficult for someone. The writing can look finished before the decision has been thought through.
The University of Bath’s September 4 announcement describes research on that gap. The paper, A Process Model of Managerial Phronesis in the Age of Generative AI, appears in the Academy of Management Review. It proposes how AI use might weaken or strengthen practical judgment. It is a conceptual account, with no universal measured rate of managerial skill loss.
The two proposed paths depend on repeated habits of work. A manager can use generated answers to pass quickly over a problem, or use them to uncover an assumption worth examining. Experience, reflection, direct interaction and responsibility give either habit consequences over time.
Check the schedule before polishing its explanation
Consider the scheduling email as a hypothetical case. The manager needs to know who faces disruption, which earlier commitments apply and what alternatives were considered. The model’s draft can help organize an explanation. It cannot establish those local facts simply by sounding reasonable.
That leaves the manager with two unfinished questions: is the proposed schedule defensible, and does the explanation accurately describe why it was chosen?
That gap makes phronesis , often translated as practical wisdom, relevant to AI-assisted work. Phronesis concerns judgment about how to act in a particular situation. A fluent general answer cannot supply firsthand knowledge of these employees, these obligations and these consequences merely by sounding reasonable.
Review the draft for accuracy and usefulness, then review the decision against the circumstances. The first can catch a misleading sentence. The second can reveal a reason to change the plan. A polished document needs both kinds of attention when people will act on it.
Fluent language can make that distinction easy to overlook. Once the page reads smoothly, moving to “send” feels like a small remaining step. Checking a commitment or speaking to an affected colleague is less visible work. It may also change more than another pass over the wording.
Explain the choice while it can still change
Bath’s account links time pressure and uncritical reliance with “epistemic de-skilling,” the proposed loss of knowledge-related capability. A manager who repeatedly skips questions or direct interaction may get fewer opportunities to develop judgment. The process model offers a mechanism to investigate, rather than a measured prediction for every workplace.
The proposed constructive route, “epistemic up-skilling,” asks the manager to challenge assumptions and explore alternatives with AI. Having to explain the final choice gives that exchange a practical demand. A person who knows the team will ask why has a reason to find the missing context before acting.
The timing of explanation changes its usefulness. Before the schedule takes effect, an unanswered question can still alter the plan. Afterward, the same explanation may become a reconstruction of why things went wrong. The opportunity to learn exists in both moments, with different costs for the people affected.
For the hypothetical schedule, keep three kinds of information distinct: facts checked directly, suggestions introduced by the model and questions still open. One previous commitment belongs in the first group only after someone confirms it. A convenient alternative time remains a suggestion until the team’s actual constraints are checked.
Ask about one assumption behind the completed form
An extra approval form can reproduce the same problem if the model fills it and nobody examines the answer. A completed field shows that text was entered. The useful exchange begins when someone can explain a consequential assumption in ordinary language.
Ask what supports that assumption, whose perspective is missing and what new fact would change the choice. These are questions someone can act on. They can prompt a conversation, a record check or a different option while the draft is still easy to revise.
Citations and extended reasoning can help trace an answer, but relevance remains local. A source may be accurate and still say little about this team’s commitments. The manager needs to connect the evidence to the people and obligations the decision will affect.
Routine drafting may need little friction when the stakes are low and the result is easy to reverse. A choice affecting someone’s workload or opportunity asks for more attention. The amount of review should follow the consequence, with enough time left to hear a fact the draft omitted.
Let the consequence reach the decision-maker
Now suppose the schedule takes effect and the overlooked problem appears. A completion report may show only that the change happened. Direct feedback from the affected colleague gives the manager something more useful to learn: where the assumptions failed and what should enter the next decision.
The learning sequence continues beyond producing the email. Understand the situation, compare options, choose, observe and revise. AI can contribute at several points. The person responsible still needs access to the consequences if the experience is to improve later judgment.
Our related article on Tina Huang's AI productivity feedback loop examines iteration as a way to improve outputs. Managerial judgment adds another requirement. The goal is not only a better document or faster result. It is a decision that someone can explain, defend and revise in light of what happens.
The learning claim needs to be followed over time
Future studies could test those proposed pathways over time. They would need to name the task, participants’ experience, AI version, comparison condition and period of observation. Faster completion and better documents need their own measures, alongside later ability to reason without assistance.
Researchers could compare a group receiving generated output alone with one asked to explain assumptions before acting and another given structured feedback afterward. That is a possible study design, not a result reported here. It could help distinguish short-term task gains from changes in retained capability.
Until such evidence establishes the effects, the model gives organizations a hypothesis to examine. Do their routines leave managers doing the contextual thinking and hearing the outcome, or do they mainly help a document reach completion?
The overlooked commitment in the scheduling example will not become visible through a more tactful closing sentence. Someone has to ask about it, hear the answer and decide what follows. That conversation is part of the work the email was meant to serve.
