After Copying Ops Report Automation, I Used WorkBuddy to Semi-Automate Government Ledger Summaries

Conclusion first: WorkBuddy isn't suitable for writing materials for me, but it's great for semi-automating repetitive daily tasks in government offices, like summarizing ledgers, generating submission lists, and organizing to-dos. It mainly saves effort on moving data around; material judgment still has to be done by me.

Yesterday I saw an article about automating e-commerce ops daily reports, saying CodeBuddy writes scripts while WorkBuddy runs scheduled collection, generates PPTs, and sends emails. Originally took an hour, now reportedly five minutes. I didn't feel ops became easier; instead, I felt a bit envious. Our office has its own "daily report": departments submit progress, collect ledgers, chase materials, summarize issues. Every morning, a pile of Word docs, PDFs, and spreadsheet screenshots sit in the shared drive. I've used WorkBuddy for nearly a month. A while back I wrote about it messing up formatting when processing official red-header documents. This time, I decided to bring over the ops mindset, keeping only WorkBuddy and not touching CodeBuddy, because I can't write scripts here, nor is it appropriate to push things toward code.

My approach was creating an Automation task in the WorkBuddy desktop client. Finding this entry point was confusing at first—it's not next to the regular chat box but in the left-side function bar. Clicking in, select New Task, name it Daily Submission Ledger. There's a term called Working Directory; this tells WorkBuddy which folder it can access, preventing it from messing with my computer. I set the directory to Submission Ledger within our unit's shared drive, not the entire drive. Then in Scheduling Rules, I set Every day at 08:30. This time is when my department's submissions are most complete; earlier means nobody has reported, later means leadership is chasing. I wrote the task prompt strictly: read Word, PDF, and spreadsheets in 01_To_Summarize, extract department, item, deadline, existing problems, generate a Daily_Report_Draft_Date.md, save to 03_Ready_For_Review, move unreadable or conflicting files to 02_Needs_Manual_Confirmation, and do not overwrite originals.

The most critical part of this step is folder layering. I didn't let WorkBuddy lump all files together; instead, I pre-created four directories: 01_To_Summarize for colleagues to drop raw materials, 02_Needs_Manual_Confirmation for scans, old versions, and unclear filenames, 03_Ready_For_Review for drafts generated by WorkBuddy, and 04_Historical_Archive for versions I've confirmed. I learned a hard lesson before: the first time I let it summarize all files directly, it mixed draft-for-comment versions with final versions, resulting in two different progress statuses for the same item in the table. Later, I added a line to the prompt: only process files containing 【Final Version】 in the filename; everything else goes to 02_Needs_Manual_Confirmation. This restriction seems dumb, but in government offices, the biggest fear is AI being overly clever.

I also made simple permission divisions. In the shared drive, colleagues can only upload to 01_To_Summarize, not modify 03_Ready_For_Review or 04_Historical_Archive. On my local computer, WorkBuddy runs scheduled tasks, with the account authorized only for the Submission Ledger directory. This way, even if I'm on a business trip and a task fails, it won't mess up others' files. Previously, AI reviews directly overwrote originals, making audit trails difficult, so now I require WorkBuddy to only generate new files, not edit original ones directly. After the draft comes out, I manually apply our unit's template. I've just started trying Word templating these past few days and am not proficient yet, so WorkBuddy outputs md or docx first, and I handle the final formatting and wording.

Daily maintenance isn't actually complex. Every morning when I arrive, I check WorkBuddy's Execution Records. If status is failed, I open the logs. Most issues are because a certain PDF is a scan and text recognition fails. When encountering this, I don't force it; I simply move the file to 02_Needs_Manual_Confirmation and take a manual look. Every Friday, I compress and backup 04_Historical_Archive once; at month-end, I clear backlog files in 02. When year-end materials pile up, I temporarily change the schedule from 08:30 to 10:30, or switch to twice daily, because submission completion times vary.

After running this, my judgment is that the value of tools like WorkBuddy is fixing the pipeline of "collecting files, categorizing, extracting fields, generating drafts, archiving," not producing beautiful summaries. Government office file formats are complex; AI gets confused as soon as it touches red-header documents—that's true. But as long as boundaries are tightened, letting it do only mechanical moving and initial screening, it's steadier than blindly stacking tools. I don't have an ops team's four sets of monitoring systems, nor code scripts, just a pile of submission forms. WorkBuddy compressing this pile from thirty minutes in the morning to ten-plus minutes already counts as successful slacking off.

WorkBuddy can't save my material quality, but it gives back the thirty minutes I spent repeating organization every day.