I tracked every AI time-saving claim I made for a month. The real number was two hours a week

Everyone quotes big numbers about AI saving time, so I logged my own for a month: every task where I used an AI tool instead of doing it the old way, with my honest estimate of minutes saved or lost. Not a productivity system. Just a tally, four weeks, no cheating.

The wins were predictable but smaller than advertised. Email drafting: about 15 minutes a day, mostly on the routine half. Meeting notes: 40 minutes a week I genuinely don't miss. Research and summarizing: another hour, lumpier — some days it saved nothing, one day it saved two hours when I needed a regulatory change explained with sources. Total: about eight hours over the month. Two hours a week. The ads say twenty. The log says two.

The losses went unreported everywhere, so I logged those too. One evening debugging a YAML block an AI wrote for my automation — negative 90 minutes. An hour reformatting a generated document whose structure broke in the export. Maybe twenty minutes a week of re-reading outputs carefully, which is the insurance premium nobody counts.

Net: call it 90 minutes a week, reliably. Here's the thing though — two hours a week is a hundred hours a year, and that's real. The compounding part isn't the volume, it's which hours they are. They're uniformly the worst ones: the admin tail of every task, the formatting, the first-draft dread. Nobody gets their best hours back. They get their worst ones, and that's precisely why it doesn't feel like twenty.

The practical takeaway from the log: I stopped using AI for tasks where it saved under five minutes, because the checking overhead ate the gain, and doubled down on the three workflows where it saved 20+. Two big wins beat fifteen small ones. Your mileage will differ, but the method won't — the log took me two minutes a day, and it's the only AI productivity advice I actually trust.