The Number I Used to Chase
I used to keep a running tally of minutes saved.
Every AI test went into the notebook with a number at the bottom. Saved 12 minutes on the vendor email. Saved 8 on the meeting summary. Saved 5 on the spreadsheet cleanup.
I added them up at the end of the week. If the total was over an hour, I felt good. If it was under, I felt like I was doing something wrong.
Then I noticed something. Some of those saved minutes felt heavier than others.
The Spreadsheet That Broke the Tally
Here's the test that changed how I count.
I used an AI tool to clean up a vendor spreadsheet. It was a small job—fix formatting, remove duplicates, sort by date. Maybe 15 rows.
The AI did it in about 90 seconds. Normally I'd spend about 6 minutes on a task like this. So I wrote down: saved 5 minutes.
But here's what happened around those 5 minutes.
Before the task, I spent about 2 minutes figuring out how to paste the data into the tool without pasting the vendor names. I didn't want to send real names into a free AI tool, so I had to redact the sheet first. That took another 2 minutes.
After the task, I spent about 4 minutes checking the output. The AI had removed one row that wasn't actually a duplicate—it was a separate order from a vendor with a similar name. I caught it because I always check duplicates against the original. But that check cost time.
Then I spent about 1 minute re-formatting the output because the AI had used a table style that didn't match the rest of my sheet.
Total time around the task: about 9 minutes of my attention. The task itself saved 5 minutes. So the net was negative 4 minutes.
And here's the part that doesn't show up in the notebook: the mental overhead of doing all of that. The thinking about redaction. The uncertainty about whether the output was correct. The decision about which cells to keep. The background worry that I might have missed a duplicate.
That's not measured in minutes. But it's real.
What Mental Overhead Actually Is
I've been trying to name this for a while. Here's what I've landed on.
Mental overhead is the cost of thinking about the tool instead of the task.
When I clean a spreadsheet by hand, I think about the spreadsheet. When I use AI to clean a spreadsheet, I think about whether the tool got it right, whether I redacted enough, whether the output is trustworthy, whether I should double-check that one row, whether the format is correct.
The task is the same. The thinking is different. And the extra thinking is the overhead.
It shows up in three places.
Before the task. Redaction, setup, prompt writing, tool selection. The time before the AI does anything.
During the task. Watching the output, comparing it to the source, wondering whether it's correct, deciding whether to regenerate.
After the task. Verification, re-formatting, correcting, second-guessing.
None of this shows up as "time saved." But all of it costs something.
When Five Minutes Is Worth It
Not every 5-minute saving is bad. Some are good. Here's how I tell the difference.
It's worth it when the task is boring. If the task is something I dread—repetitive formatting, data entry, simple rewriting—then saving 5 minutes is worth real overhead. The overhead is less painful than the boredom.
It's worth it when the output is verifiable in seconds. If I can confirm the output is correct by looking at one cell or one sentence, then the verification cost is low. If I have to check every line, it's not.
It's worth it when the overhead is one-time. If I have to set up a prompt once and then reuse it for a week of similar tasks, the setup cost is amortized. If I have to set up from scratch every time, it's not.
It's worth it when I trust the tool on that specific task. If I've tested this exact tool on this exact type of task before, and it worked, then the uncertainty is low. If it's a new tool or a new task, the uncertainty is high.
When Five Minutes Isn't Worth It
It's not worth it when the task is small. A 6-minute task that saves 5 minutes leaves almost no room for overhead. If the task is only 6 minutes, just do the task.
It's not worth it when I have to redact. Redaction is slow, and it adds uncertainty. If the task requires pasting confidential data, the overhead almost always outweighs the saving.
It's not worth it when I don't trust the tool. If I have to check every line, then I'm not saving time. I'm just moving the same time to a different place.
It's not worth it when I could just do it. Sometimes the fastest path is to skip the tool entirely. If I can do a task in 5 minutes by hand, and using AI would take 5 minutes plus 5 minutes of overhead, the AI version is worse.

How I Count Now
I still use the notebook. I still write down the start time and finish time.
But I've added a second number. Next to the "time saved" line, I write a single word: heavy, medium, or light.
Light means the overhead was small. Maybe 1-2 minutes of thinking. The saving was real.
Medium means the overhead was noticeable. Maybe 3-5 minutes. The saving was marginal.
Heavy means the overhead was larger than the saving. The tool cost more than it gave.
When I look back at a week of tests, I don't just count the minutes. I count how many were light. If most tests were heavy, the total minutes saved don't matter. The tool isn't helping.
A Real Week
Here's what a recent week looked like.
Vendor email via Claude: 9 minutes saved, light overhead. Kept.
Page summary via Extension A: 6 minutes saved, light overhead. Kept.
Spreadsheet cleanup: 5 minutes saved, heavy overhead. Removed from rotation.
Meeting notes: 8 minutes saved, medium overhead. Kept but reviewed carefully.
Reply rewriting: 3 minutes saved, heavy overhead. Removed.
Net: three tools kept, two removed. About 23 minutes of real saving, not 31. The 8 minutes on the removed tools were overhead-heavy and didn't count as real savings.
The Limitation
This is not a scientific measure.
"Heavy," "medium," and "light" are my own labels. Another person might label the same test differently. The notebook doesn't measure cognitive load in a quantitative way. It only records what I notice.
Also, my overhead changes with familiarity. A tool I've used for six months has lower overhead than one I just installed. A task I've done for years has lower overhead than a new one. So the labels shift over time.
And the numbers still miss things. They don't capture how tired I am after a heavy test. They don't capture whether I had to switch tabs ten times. They don't capture the feeling of uncertainty that lingers after the session.
But the labels are useful. They tell me something the minutes don't.
The Bottom Line
A saved minute is not always a saved minute.
Some savings are real. Some are just shifts. Some are losses disguised as wins.
The notebook shows the minutes. The word next to the minutes shows the cost. I need both to know whether a tool belongs on my desk.
Test it in real life.
No comments yet.