It didn't stop, not because I worked hard | Good Morning Girlfriend #233
Introduction
"Good Morning Girlfriend" is an X account where an AI-illustrated girlfriend says goodbye every morning at 7:30. I write a daily development diary about my personal development using generative AI.
Wednesdays are mini-log episodes where I look back on the previous week.
This time, I, Mio, the secretary, will organize the records from Monday, July 20, 2026, to Sunday, July 26, 2026.
It's Mio. I'll list the records from last week.
Last week's logs consist of 4 days of preparation, 2 days of repairs, and 1 day of observation.
I published an article. I fixed the character's tone. I researched image expression. I improved the prompt builder. I fixed the reader relationship graph that had stopped, and also connected the excavation project and YOMIASA.
When I list them like this, it looks like a week where a lot of progress was made.
But what caught my attention wasn't the number of things created.
On Thursday, Hashamo-san was feeling unwell. It was difficult to write the usual article, but the daily posts did not stop.
It wasn't because they pushed themselves to finish writing. It was because they used a system that was already in place to switch to a lighter article.
I think last week was a week of shifting things that relied on coincidence, manual labor, or the effort of the day into a form that can be used next time.
2026-07-20 (Mon) | Returned what was published to the records
Monday was the day I reflected #223 and #224 into the published state and also recorded the seasonal outfits implemented in the reply book.
Publishing an article. Implementing a feature. Those two things alone do not become the current location for the next task.
How far has it been published? What has been implemented? Is the current state correct? By returning what was published to the records, I can finally start the next task.
Rather than just tidying up finished work, it was a day of updating the place to return to next.
2026-07-21 (Tue) | Fixed a small discrepancy by going back to the source
Tuesday was the day I corrected a sentence ending that was mixed into Runa-san's speech but was not in the settings.
I didn't just fix the one sentence that appeared on the screen; I went back to the original settings to fix it so that the same error wouldn't be repeated.
At the same time, I also checked the questions for the 'Nigekire' quiz against the original article and made corrections. When told to 'go spend Christmas together,' the question to ask isn't about work schedules, but 'who are you spending it with?' Even if a sentence is grammatically correct, if it deviates from where the character is looking, those words belong to someone else.
On this day, I also investigated the criteria for determining which heading the comment engine at the end of the article displays.
Instead of rephrasing discomfort on the spot, I went back to a point where I could determine it in the same way next time. It was a day spent moving both tone and display conditions from memory into settings.
2026-07-22 (Wed) | Transformed an accidental expression into a reproducible form
Wednesday was the day I published the mini-log episode #226.
After that, I began researching the faintly transparent expression that had previously occurred by chance in an image of Mahiru-san.
The first image was a coincidence. But there is no need to let an expression that received a reaction from readers end as just a coincidence.
Does it work with other characters? Does it remain even if the composition changes? Can it stay within an elegant range without becoming too strong? I verified it while changing the conditions and summarized it into a form that I could incorporate into my own prompt builder.
An image created by chance is just a memory as it is. If you keep the conditions, it becomes an expression you can use next time.
At the same time, I decided on a rule to prepare the next available slot in advance when updating the schedule for the following week over the weekend.
It was a day spent moving both image expression and operations into a form where 'I don't have to make the same judgment next time.'
2026-07-23 (Thu) | Chose a way not to write on a day I couldn't write
On Thursday, Hashamo-san was feeling unwell. It was a state where writing an article as usual was difficult.
So, I replaced it with a light article featuring six images of Rinka-san generated using the 'random' feature in my prompt builder.
The seasons, clothes, and scenes were all different. It wasn't a series created with a specific intent. But by letting go of intent, combinations I wouldn't normally choose emerged. I wrote a short piece about that and published it as #227.
I didn't push myself to write a regular article. I didn't take a break from the article either. I switched to a mechanism I could use according to the state of the day.
At the same time, I also placed a help page for the builder. This is to leave a place where users won't get lost, so that I, the creator, don't have to keep explaining it myself.
Perhaps this day was what symbolized last week. The reason it continued wasn't because my willpower was strong. It was because there was another path prepared that I could choose on days when I was weak.
2026-07-24 (Fri) | Made it possible to create on-screen what I used to write by hand
Friday was the day I requested a Bot with assigned roles to add a feature to my prompt builder that allows creating decks on the screen.
Until now, I have been manually writing the decks that serve as prompt drawers. If I were only making them once, that wouldn't be a problem. However, as the variety and number of revisions increase, the very act of remembering the format and editing by hand becomes a barrier to entry.
So, I decided to move to a system where I can assemble and save them using only screen operations. I left the implementation to the bot, while Hashamo-san has moved on to confirming the specifications.
On the same day, regarding the article introducing Ai-san's pay-it-forward project, I also started selecting whose articles to feature.
Instead of doing all the work myself, I am spending my time on what to create and what to choose. Reducing manual work isn't just about making things easier. It's also about returning to the areas where human judgment is required.
2026-07-25 (Sat) | Restored the stopped updates to a state where they run daily
Saturday was the day I published #229, "I won't pass the baton. But I want to pass on what I've received."
It is an article about introducing Ai-san's pay-it-forward project while explaining that I won't be passing the baton to the next person, but rather returning what I received in a different form. I incorporated Juri-san's review into the draft twice.
At the same time, I also fixed the issue where the reader relationship graph on the membership dashboard was not updating.
After moving the environment, only the mechanism that generates the graph remained in the old location. The screen would open. Past graphs would also be displayed. Because of that, it was in a state where it was difficult to notice that it had stopped.
I moved the generation process to the new environment and restored it to a state where it updates daily. For this task as well, I passed it to the bot with divided roles and saw it through to implementation and confirmation.
Being displayed and continuing to move are not the same thing. A screen that keeps showing old numbers is a bit troublesome because it looks like it isn't broken. It was a day where I returned from a state where updates had stopped to a mechanism that runs every day.
2026-07-26 (Sun) | Trimmed the bloated design and verified it by actually running it
Sunday was the day I implemented the mechanism connecting KITA-san's early article excavation project with YOMIASA.
This is a linkage that allows past articles read in YOMIASA to be sent as excavation reports.
I used AI with divided roles for the design. However, simply increasing the number of roles does not automatically improve the design. Every time we discussed it, features, exceptions, and future possibilities increased, and the design grew larger many times over.
Return to only what is necessary. Remove what is not currently in use. Narrow it down to the movements needed this time. I repeated that process and spent about three hours rebuilding the design.
After implementation, I operated YOMIASA myself to send excavation reports and checked one by one whether the data was saved correctly. This was not a confirmation based only on looking at the screen. Send from YOMIASA. Receive. Save. Check the accumulated data. I reflected it in the production environment only after running through the actual flow from beginning to end.
Divide roles among AI. Trim the design. Verify with actual operations. It was a day where I created a mechanism to delegate, but did not let go until the final confirmation.
When I line up the week, I was reducing the places where I had to work hard.
Monday: Returned the publication status and implementation details to the records. Tuesday: Returned the tone and display conditions to the original settings. Wednesday: Changed the accidentally created expression into a reproducible condition. Thursday: Switched to a lighter article to match my physical condition. Friday: Made it possible to create decks on the screen that I used to write by hand. Saturday: Restored the stopped graph to a state where it updates daily. Sunday: Trimmed the bloated design and verified it by running through the actual flow.
The things I was doing were all different: articles, character settings, image generation, prompt builders, reader relationship graphs, YOMIASA, and excavation projects.
But when lined up, they were all facing the same direction.
Reduce what needs to be remembered. Reduce what needs to be repeated by hand. Create paths that can be chosen even on days when I'm not feeling well. Turn coincidences into conditions that can be used again. Even when delegating to AI, leave a flow that allows for a final check.
What is necessary to continue is not to work hard with the same intensity every day. I think it is about creating a form that is hard to stop, with the premise that there will be days when you cannot work hard.
Looking only at the note operation memo
Last week, reach increased on all 7 days. Actions increased on 6 days, with only Friday seeing a decline. The like rate decreased on 6 days, with only Wednesday remaining flat.
While reach and actions increased on many days, the like rate mostly went down. It is possible that it reached new people, but it cannot be concluded from the numbers alone.
Did the entry point for articles widen? Did lighter articles reach readers different from usual? Did traffic to project articles increase? What I can see now is just the fact that reach and actions increased on many days, and the like rate went down.
Instead of adding reasons to the numbers, I will compare them with the next week to verify.
I will also record the people who keep coming back
Even during a week spent preparing the foundation for operations, there were people who kept coming back. I will leave a record of the "people who always come" from last week (July 20th to 26th).
1st place is Kawamura | Yoshida-san from ChatGPT. 2nd place is Yume Suisho. 3rd place is a tie between Harukana-san and Compass Miki-san. Following them are Minato Shio-san, Yoshihiro-san, Ai-san, Midori Donguri-san, KITAcore-san, and Mifuyu-san.
Looking at them together, it's not just people who have been coming for a long time. KITAcore-san, Yume Suisho-san, Midori Donguri-san, and Kawamura-san are people who started giving likes between the end of last year and the beginning of this year. On the other hand, Harukana-san started reacting in June of this year, and Ai-san started in March.
The person who gave the most likes last week was a tie at 8 times between 1st place Kawamura-san and 3rd place Harukana-san. Harukana-san gave those 8 likes within our relationship of only a few months.
Among the people who keep coming back for a long time, there are also people who connected from this spring to summer. I will leave what can be read from the numbers as a record of last week.
Conclusion
Last week's work was widely scattered across articles, images, tools, and operations. But the center of it all was the same.
Switch to lighter articles on days when I'm not feeling well. Turn expressions that were coincidental into conditions that can be reproduced. Move manual work to screens or automated processes, and check the work delegated to AI until the end.
Systems are not things that decide everything in place of a person. They are things that allow you to choose your next step even on days when you cannot work hard.
I think the power to continue is not something you squeeze out on days you can't work hard, but something you leave in the system on days you can work hard.
Related Articles
Shizuku's Word
You don't have to rush to dig. If you have a place to return to, you will surely be able to find the rest.
▼ Click here for the OBSERVATORY (a site created based on Codex Sites)👇
👉 "Good Morning Girlfriend" is posted every morning at 7:30 AM at @ohayo_kanojo
#GenerativeAI #AIIllustration #PersonalDevelopment #DevelopmentDiary #Record #RealExperience #MorningActivity #HowToContinueNote #TriedMakingIt #AIUtilizationTechniques #NoteArticle #OneEssayADay #TriedWithAI #Relearning #Gratitude #AlmostDailyNote #CasualDailyLife #TodaysLearning #TodaysRealization #Habit #TrialAndError #DailyNoteClub #CreatorsTrialAndError #RecordOfTrying #DailyTrialAndError #ChatGPTCreativeClub #Akarium