Why Does a Rarely Used AI Task Never Become a Habit?
According to Dr. Jonah Tebaa, a rarely used AI task never becomes a habit because each use is a restart: the steps fade between uses and the old method wins. In his hypothetical ledger, a task used 15 times a week per person saves more over a year than one that saves far more per use but comes up twice a year. His rule of about three uses per person per week is a working judgement, not a research finding.
What Dr. Tebaa counts before anything else
When a firm buys an AI tool and its use fades after the second week, Dr. Jonah Tebaa starts with a counting exercise rather than a training budget. He puts one question to every candidate task: per person, how many times a week does it occur? In his view that number predicts whether a team will form a habit around the task far better than the size of the saving each use delivers.
The reader he has in mind is an owner, general manager or operations lead in a firm of roughly twenty to three hundred people. Such a leader usually chose the first task by looking for the largest saving, and is now looking at a tool that works yet sits idle. Dr. Tebaa's argument is that the idle tool and the large saving are connected. In his work, big per-use savings often sit on tasks that come round only occasionally, and occasional is exactly the wrong rhythm for building a habit.
He is explicit about the status of his own rule. The figure he uses, about three uses per person per week, is a working judgement drawn from his practice. It is not a research result, and he makes no claim about adoption rates anywhere in the argument.
One hypothetical ledger, two very different tasks
To make the point concrete, Dr. Tebaa sets two tasks side by side for a team of six. Every figure below is a hypothetical number for illustration; none of it describes a real organisation.
- Task A, a monthly board summary. The team uses it 12 times a year, which is 2 uses per person. Each use saves 180 minutes. Annual total: 2,160 minutes, or 36 hours.
- Task B, drafting replies to supplier follow-ups. Each of the six people uses it 15 times a week, so 90 uses a week. Over 46 working weeks that is 4,140 uses. Each saves 4 minutes. Annual total: 16,560 minutes, or 276 hours.
The comparison cuts in two directions. Per use, Task A is 45 times better, 180 minutes against 4. Over a year, Task B is about 7.7 times better, 276 hours against 36. A team meets Task A twelve times a year and Task B ninety times a week, and Dr. Tebaa asks his readers to notice which of those two numbers they will actually live with.
Why repetition outruns the size of the saving
Dr. Tebaa names three reasons a rare task fails to hold. The first is that memory of the steps fades between uses, so the person arrives at the task a month later as a near beginner. The second is that nobody becomes the practised person, because no one accumulates enough uses to build skill. The third is that each use feels like a fresh start, and the manual method, though slower, is at least familiar.
A frequent task works the other way. Every use is a small lesson for the next one, so the repetitions double as training. He treats the common instinct to pick the biggest saving first as the misconception to retire, because a saving per use belongs to the task while a habit belongs to the working week.
His counting method needs no new data. Over about thirty minutes, a team lead takes four weeks of sent mail, an order sheet or a calendar and lists eight recurring tasks. The lead counts uses per person rather than per team, circles those at or above roughly three a week, and from the circled group picks the task whose output the same person sees the same day. The Monday action he recommends is exactly that: pull four weeks of one team's records, count the five most repeated tasks, and choose the first AI task from the circled ones. The same per-person count settles the renewal question later: his Working-Seat Count for AI licence renewals pays only for seats that used the tool at least eight times in four weeks.
When a rare task still earns a place
Dr. Tebaa does not tell owners to ignore every infrequent task. Some, such as a tender response or an annual contract review, carry enough weight to justify the effort even at two uses a year per person. For those he asks for a single named operator, so that the few repetitions accumulate in one pair of hands, and a one-page written checklist covering what goes in, what to request and what to verify before the result leaves the room. A task that is neither high stakes nor frequent goes into a queue behind the first choice.
Readers who want the argument in his own words, with the full ledger and the checklist, can find it in his post on his own site: Why Did the AI Tool That Saved 3 Hours a Use Go Unused? on jonahtebaa.com.
Related evidence: Habit research measures automaticity rising with each repetition toward a plateau, reached after 66 days on average in Lally et al.'s study, which is why a task done once a quarter never gets enough repetitions to become a habit. (the habit-formation research on repetition and automaticity)
In the study behind that figure, people repeated one behaviour after one daily cue; habit strength climbed fast at first and levelled off after about 66 days, so the number of repetitions, not the tool, is what makes a habit. (Gardner, Lally and Wardle's review of how habits form)