Excerpted from 55 and Out, AI Automation
The Automation I Killed Because It Was Faster by Hand
The automation was meant to handle quarterly client reports. Pull data from four sources, format, generate a PDF per client, send a personalized email with the PDF attached. About thirty clients. The manual version took me roughly four hours per quarter.
I built the automated version over three weekends. Custom data pulls, a templating engine, a PDF generator, an email scheduler. By the time I was done, the script was about 600 lines of code with a small admin UI for previewing each report before send.
I ran it. It worked. Each report took about 90 seconds to generate. Total runtime for all 30: about 45 minutes.
Then I noticed the problem.
The manual version took four hours, but those four hours included me looking at every report. I'd see when a client's metrics had drifted in a worrying direction. I'd remember a conversation we'd had and add a personal sentence. I'd notice that one client's numbers looked too good and double-check the data. I'd catch the occasional input error.
The automated version skipped all of that. Reports went out faster, looked cleaner, and were emptier. The data was right. The thinking was missing.
I killed the automation after one cycle. I went back to the manual version, but kept one piece of the automated work — the data pull and assembly into a draft PDF. That part was a clear improvement. The script now hands me 30 draft reports in 10 minutes, and I spend the next 3 hours doing the part that needed me — reading, noticing, personalizing, sending.
Net time savings: about an hour per quarter. Net quality: better than either pure version. Net cost: a humbling lesson about what automation is actually for.
What I'd do again: the data pull and assembly. That's pure mechanical work. Automate it.
What I wouldn't: automate the part of any process that's actually a thinking step disguised as a mechanical one. The give-away is when the manual version takes longer than it "should" — that's usually because something subtle and important is happening in those extra minutes. Automating that step doesn't save time; it deletes thought.
The deeper lesson: every job has a craft layer underneath the procedure layer. Automate the procedure. Protect the craft. If the craft layer disappears, so does the value of the work.
If you're about to automate something and the calculation is "this will save me X hours" — pause. Ask: what was happening in those hours? If the honest answer is thinking, then the savings are illusory.