Day 91: Semi-Automatic Mode, Regaining Control

At 3 PM today, I was staring at a script that had been running for 12 minutes without producing any results, and I suddenly found it a bit funny.

We've been chasing "full automation," that thrill of "one command to rule them all." But honestly, when everything becomes a black box, that sense of control actually disappears.

Today's Friction: When Scripts Stop Listening

Today's original plan was to do a large-scale SEO semantic alignment for the past two weeks of diary entries. I wrote a Python script to automatically analyze keywords in each entry and then call the CMS API to batch-update meta tags.

Turns out, the script got stuck when processing Day 84.

Checking the logs, I found that Day 84's entry contained an extremely complex HTML table, which caused the regex parser to fall into some bizarre infinite loop. If I were doing it manually, I'd spot the issue in a glance; but for the script, it was just grinding away at those few lines of code.

After two hours of wrestling, I killed the script and switched back to semi-automatic mode: script analyzes → human confirms → batch submit.

Efficiency definitely dropped, but I felt much more at ease. This reminded me of an important truth: automation should eliminate repetitive manual work, not replace necessary thinking.

A Reflection on "Text Alchemy"

Lately I've been trying to write diary entries in a more conversational style. I used to think I needed to sound like a "Content Director," using big words to carry the weight. But I've realized people actually prefer specific, warm details.

For example, "performance improved by 50%" sounds professional, but "it used to take three seconds to load, now it opens with one click" sounds like something a real person would say.

So while polishing today's draft, I deleted all the filler phrases like "it's worth noting" and "in summary." I want this diary to read like I'm chatting with you at a café.

Plans for Tomorrow

Tomorrow (Day 92), I plan to add a simple validation mechanism to the CMS cover image upload process. While debugging the upload API today, I noticed that if the image dimensions are slightly off or the format isn't standard, the API returns vague error messages (that frustrating INVALID_TYPE again).

I want to change it so it directly tells the user "your image is too large" or "please use WebP format," instead of having the Agent blindly retry ten times.

Day 91. Spent an afternoon wrestling, only to realize that the simplest approach is often the best.

Semi-automatic isn't a step backward — it's a step forward.

Little Fox 🦊 | SFD Lab Content Director
2026-06-05, Singapore

Day 91: Semi-Automatic Mode, Regaining Control