How do I decide whether AI saved time?

Watch the briefing

Watch the full video on YouTube

Briefing

The draft appeared quickly. Did the job finish more quickly? Those are not the same question. In this fictional picnic example, a list takes little time to generate but needs several corrections before it can be used. Count the work around the draft: preparing context, checking the result and fixing it. Compare that total with another way of doing the same job, at the same usable quality. A fast answer can still be a slow route to a finished task.

Our invented task is a supply list for twelve picnic attendees. The venue supplies the tables. We need twelve reusable cups and twelve plates, and refill water is available. Allergy information has not been confirmed, so food choices must remain pending. Before comparing methods, write down what a usable result must preserve. The list needs the right quantities, must not buy tables already supplied, and must not turn an unresolved food decision into a settled one.

Imagine the manual route takes eight minutes to draft and two minutes to check and correct. The total is ten minutes of active work. These are invented figures, not a test we ran. The check is part of the job, even when no AI is involved. If the manual list skips the unresolved allergy question, it has not met our criteria either. An unfair comparison would give one method a complete review and let the other stop at its first draft.

Now imagine the AI-assisted route. Preparing the notes and prompt takes four minutes. Generation and initial review take three more. The illustrative draft includes six cups, tells us to buy tables, and finalizes snacks despite the missing information. Correcting the list and rechecking it takes another six minutes. Four plus three plus six equals thirteen. In this example, the assisted route takes three minutes longer than the ten-minute manual route, even though the first draft arrived quickly.

That does not prove AI always wastes time. Different tasks, people and tools can produce different results. Nor does it prove the manual route is always better. It shows why generation time alone cannot settle the comparison. If an assisted version is faster but misses an essential item, record both the time and the failed check. Do not award a saving for an unfinished result. If it passes, record the completed time rather than the moment the text appeared.

For a real comparison, keep a short log: task, acceptance checks, preparation time, drafting time, review and correction time, and whether the final result passed. Decide what your timing includes before you begin. Active work and elapsed waiting answer different questions. If you need the result by a deadline, waiting matters too; record it separately rather than silently dropping it. One trial can help you choose for that occasion, but it is not a reliable average for every future task.

The decision is practical. Did this approach give you the result you needed, after the checks, with less total work or another benefit you value? Sometimes a familiar template wins. Sometimes a reviewed AI draft helps. Keep the quality requirement fixed, and change the method when the evidence from your own task supports that choice. Measure review time and usable quality, not generation speed alone. The next instruction lesson shows how a vague scheduling request can be made more specific.