The strongest case for workplace AI rarely begins with a dramatic breakthrough. It begins with ordinary time savings: a report drafted faster, a plan assembled in minutes or a first pass produced without the usual blank-page delay. Those gains matter. Yet speed can become a dangerous proxy for value when leaders stop asking what happened to the judgment behind the work.
Architecture offers a useful warning. A recent RIBA survey reported that AI adoption among practices had risen sharply and that many users saw productivity improvements. At the same time, far fewer respondents believed AI had improved design quality, while many worried that early-career professionals would struggle to develop essential skills. That gap between faster production and stronger professional capability reaches far beyond architecture. It applies to HR teams screening candidates, lawyers reviewing contracts, analysts preparing forecasts, designers developing concepts and managers writing recommendations.
The wrong scoreboard
Most organisations still measure AI through activity. They count licences, active users, AI-assisted tasks, hours saved and output volume. Those numbers can show whether a tool is present. They cannot show whether it improves the work.
A team may produce more documents while increasing the correction burden on colleagues. A manager may save an hour drafting a proposal while spending two hours verifying weak assumptions. A junior employee may submit polished work without learning how to identify the underlying risks. A professional may become quicker at accepting a plausible answer and slower at forming an independent view.
These effects often remain invisible because the people who absorb the costs sit downstream. The employee who creates an AI-assisted analysis records a productivity gain. The colleague who has to reconstruct the evidence, challenge the logic or repair the tone experiences the loss. Both events belong to the same workflow, but most organisations measure only the first.
That creates a misleading business case. Leaders see local efficiency while the organisation accumulates rework, review fatigue, brittle skills and growing dependence on a system that cannot carry professional responsibility.
Make judgment visible
A better approach starts by defining the decisions that require human judgment before choosing where AI belongs. In a design process, that may include understanding client intent, interpreting constraints and recognising when an apparently elegant solution creates a practical problem. In HR, it may involve distinguishing relevant experience from superficial similarity. In finance, it may mean deciding which assumptions deserve challenge rather than simply extending a model.
Once those judgment points are explicit, teams can assign AI a bounded role. It can generate alternatives, retrieve information, identify inconsistencies or produce a preliminary draft. The professional remains responsible for deciding what matters, what evidence supports the conclusion and what should happen next.
This distinction sounds obvious, but it changes how managers review work. “Did a human check it?” becomes too weak a question. A quick glance at a polished output may amount to little more than approval by appearance. Leaders need to know what the reviewer tested, what evidence they examined, what they changed and what uncertainty remains.
That does not require a bureaucratic audit trail for every email. It requires proportionate review. Low-risk internal drafts may need a simple sense check. Decisions affecting safety, employment, customers, money or reputation need a visible chain of reasoning. The more consequential the decision, the more clearly the organisation should separate machine contribution from human accountability.
Protect the learning loop
The harder problem involves development. Professional judgment grows through repeated exposure to imperfect information, feedback and consequences. Early-career employees learn by making a first attempt, seeing what they missed and understanding why a more experienced colleague reached a different conclusion.
AI can strengthen that learning loop when it provides examples, challenges assumptions or helps compare alternatives. It can also short-circuit the loop when employees use it to jump directly to a polished answer. The work looks more advanced while the worker’s underlying capability remains shallow.
Managers should therefore design different AI rules for learning and production. A junior employee might first outline a recommendation independently, then use AI to test it. A team could compare its original reasoning with an AI-generated alternative and discuss the differences. Reviewers could ask employees to explain which parts of the output they accepted, rejected or changed. These practices keep the cognitive work visible without pretending people should avoid useful tools.
Organisations also need to measure capability over time. Can employees explain their decisions without the tool? Can they notice when an output conflicts with context? Can they challenge a confident but weak recommendation? Can they transfer what they learned to a new problem? These questions reveal whether AI is extending expertise or quietly replacing the practice needed to build it.
Productivity deserves a place on the scorecard, but it should sit beside quality, rework, review burden, error detection and skill development. Otherwise leaders may reward teams for producing more while weakening the very judgment that makes their work trustworthy.
The central workplace question is no longer whether professionals will use AI. They already are. The question is whether organisations will shape that use around durable human capability. The employers that get this right will gain speed without sacrificing the expertise needed to know when speed is taking them in the wrong direction.
Gleb Tsipursky, PhD is a behavioural scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook
August 12, 2026
Productivity gains from workplace AI mean little if they erase professional judgment
by Dr Gleb Tsipursky • AI, Architecture, Comment, Workplace design
Architecture offers a useful warning. A recent RIBA survey reported that AI adoption among practices had risen sharply and that many users saw productivity improvements. At the same time, far fewer respondents believed AI had improved design quality, while many worried that early-career professionals would struggle to develop essential skills. That gap between faster production and stronger professional capability reaches far beyond architecture. It applies to HR teams screening candidates, lawyers reviewing contracts, analysts preparing forecasts, designers developing concepts and managers writing recommendations.
The wrong scoreboard
Most organisations still measure AI through activity. They count licences, active users, AI-assisted tasks, hours saved and output volume. Those numbers can show whether a tool is present. They cannot show whether it improves the work.
A team may produce more documents while increasing the correction burden on colleagues. A manager may save an hour drafting a proposal while spending two hours verifying weak assumptions. A junior employee may submit polished work without learning how to identify the underlying risks. A professional may become quicker at accepting a plausible answer and slower at forming an independent view.
These effects often remain invisible because the people who absorb the costs sit downstream. The employee who creates an AI-assisted analysis records a productivity gain. The colleague who has to reconstruct the evidence, challenge the logic or repair the tone experiences the loss. Both events belong to the same workflow, but most organisations measure only the first.
That creates a misleading business case. Leaders see local efficiency while the organisation accumulates rework, review fatigue, brittle skills and growing dependence on a system that cannot carry professional responsibility.
Make judgment visible
A better approach starts by defining the decisions that require human judgment before choosing where AI belongs. In a design process, that may include understanding client intent, interpreting constraints and recognising when an apparently elegant solution creates a practical problem. In HR, it may involve distinguishing relevant experience from superficial similarity. In finance, it may mean deciding which assumptions deserve challenge rather than simply extending a model.
Once those judgment points are explicit, teams can assign AI a bounded role. It can generate alternatives, retrieve information, identify inconsistencies or produce a preliminary draft. The professional remains responsible for deciding what matters, what evidence supports the conclusion and what should happen next.
This distinction sounds obvious, but it changes how managers review work. “Did a human check it?” becomes too weak a question. A quick glance at a polished output may amount to little more than approval by appearance. Leaders need to know what the reviewer tested, what evidence they examined, what they changed and what uncertainty remains.
That does not require a bureaucratic audit trail for every email. It requires proportionate review. Low-risk internal drafts may need a simple sense check. Decisions affecting safety, employment, customers, money or reputation need a visible chain of reasoning. The more consequential the decision, the more clearly the organisation should separate machine contribution from human accountability.
Protect the learning loop
The harder problem involves development. Professional judgment grows through repeated exposure to imperfect information, feedback and consequences. Early-career employees learn by making a first attempt, seeing what they missed and understanding why a more experienced colleague reached a different conclusion.
AI can strengthen that learning loop when it provides examples, challenges assumptions or helps compare alternatives. It can also short-circuit the loop when employees use it to jump directly to a polished answer. The work looks more advanced while the worker’s underlying capability remains shallow.
Managers should therefore design different AI rules for learning and production. A junior employee might first outline a recommendation independently, then use AI to test it. A team could compare its original reasoning with an AI-generated alternative and discuss the differences. Reviewers could ask employees to explain which parts of the output they accepted, rejected or changed. These practices keep the cognitive work visible without pretending people should avoid useful tools.
Organisations also need to measure capability over time. Can employees explain their decisions without the tool? Can they notice when an output conflicts with context? Can they challenge a confident but weak recommendation? Can they transfer what they learned to a new problem? These questions reveal whether AI is extending expertise or quietly replacing the practice needed to build it.
Productivity deserves a place on the scorecard, but it should sit beside quality, rework, review burden, error detection and skill development. Otherwise leaders may reward teams for producing more while weakening the very judgment that makes their work trustworthy.
The central workplace question is no longer whether professionals will use AI. They already are. The question is whether organisations will shape that use around durable human capability. The employers that get this right will gain speed without sacrificing the expertise needed to know when speed is taking them in the wrong direction.
Gleb Tsipursky, PhD is a behavioural scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook