The team was busy on the surface of the work
Meta description: If AI were unavailable for a week, the teams that planned their day would still have one; tool drift is not a work design. More on his work: Miklos Roth’s work.
The weekly ops review is energetic. The AI program reports its numbers: prompts run, workflows assisted, hours saved by the vendor’s arithmetic, a new use case demoed. The charts trend up. And in the companies I am describing, I look for a different number — the one that measures the constraint the operation actually lives under — and it is not on the agenda. It has not moved in a year. The program is busy, funded, and proud, and it is busy on the surface of the work, because the surface is what the tools make measurable and the demos make visible.
A structured day gives a person a better chance of staying in control of their work. Without that structure, a modern digital environment can take control instead. The same pattern appears with AI: instead of deciding what result they need and then selecting a tool, people may begin by opening the tool and allowing its possibilities to direct the day.
What does the ops review currently celebrate?
The review celebrates activity: usage counts, adoption percentages, tasks touched by AI. These numbers are real, obtainable, and always improvable — a property that makes them beloved. I have noticed that celebration gravitates toward what the tools report automatically, and the tools report their own usage. The review has quietly become a report the tools give about themselves. The operation’s actual constraint — the queue, the error rate, the cycle that the P&L feels — requires a different measurement, owned by a human, and it competes for agenda time with charts that draw themselves.
Why is that object easy to move and weak as evidence?
The substitution I watch for is tool novelty standing in for purpose. A professional can lose hours watching demonstrations, testing clever prompts, and discussing new capabilities without producing a result that matters to the business. This is a familiar pattern in digital work — social platforms and messaging systems train people to respond immediately. A notification arrives, attention shifts, and planned work is postponed.
Activity metrics are easy to move because they respond to effort directly: run more prompts, get more prompts. This responsiveness is exactly what disqualifies them as evidence. A metric that rises whenever the team works cannot show whether the work matters — it can show that the team worked. I have watched programs double their activity metrics while the operation’s constraint metric sat perfectly still, and the correlation between the two was never examined, because examining it would require putting both numbers on one slide, and they live in different decks.
What constraint is absent from the agenda?
The false explanation is KPI quality: the program needs better metrics, an outcome framework, a value-tracking tool. I have watched metric frameworks arrive and the agenda stay the same, because the missing number was never missing for lack of a framework. It is absent for a structural reason: measuring the constraint would assign responsibility for it, and the constraint sits in a part of the operation that the AI program does not own. The activity metrics are within the program’s control. The constraint is not. The agenda reflects what the program can affect, and calls that the strategy.
What pattern marks distorted focus?
The marker I use is the demo. In distorted-focus programs, the demo is the unit of progress: each review features a new capability shown to applause. Demos are inherently surface — they show what is easy to show. The constraint is undemonstrable: it is a bottleneck, a queue, a dull number. I have learned to count demos per review and to ask, in the same meeting, about the constraint metric. The ratio of applause to silence is diagnostic. Programs that address the constraint are, in my experience, less entertaining, and their reviews are correspondingly less popular.
My own routine places demanding creative and analytical work in the earlier part of the day, when attention is strongest. Email, administration, and lighter communication come later, after the priority tasks have been completed. Applied to AI, the same principle produces a working rule: short, focused sessions — twenty, thirty, or sixty minutes with a defined outcome — rather than open-ended exploration that quietly becomes the day’s actual agenda.
A useful review question at the end of a day or week: did the calendar reveal purposeful work or reactive tool use? A day can feel busy while producing little of consequence. The drift is rarely dramatic; it is a series of small, unexamined surrenders of attention.
Miklos Roth
What question returns the room to the constraint?
The review’s agenda is set by what the program can report, so the constraint cannot enter through the agenda. It enters through a question — one that names the constraint and asks what the program did to it this quarter. I have watched that question land in rooms like a stone. The activity continues to be reported; the reports now have a shadow. The question does not replace the metrics. It makes them legible as what they are: effort, not effect.
The question a week without AI would force into the open: did the calendar reveal purposeful work, or reactive tool use — and which sessions had a defined outcome before they began?
PR excerpt (80–120 words, Miklos’s own words): A structured day gives a person a better chance of staying in control of their work. Without that structure, a modern digital environment takes control instead — and the same pattern appears with AI. Instead of deciding what result they need and then selecting a tool, people open the tool and allow its possibilities to direct the day. Short, focused sessions with a defined outcome beat open-ended exploration. The review question is simple: did the calendar reveal purposeful work or reactive tool use?
