After All That Model Shopping, Better to Set Up WorkBuddy's Document Cabinet First
Community Discussion · General

After All That Model Shopping, Better to Set Up WorkBuddy's Document Cabinet First

Tang Ping Xiao YueTang Ping Xiao Yue6d ago2026/10/29 146 views

Yesterday the boss went to a meeting, and I was the only one left in the office. I came across a 2026 AI writing tool recommendation article, organized quite finely by task division. For writing and coding, look at Claude; for general capability, ChatGPT; for images and video, Gemini; for research, Perplexity. After reading it, my first reaction was: all correct, but not very relevant to me.

What I usually write is annual summaries, special reports, and leadership speeches. The hardest part of this kind of thing is our unit's terminology, last year's numbers, and the few phrases the supervising leader habitually uses—writing style is secondary. No matter how powerful the model is, it doesn't know that around here we call "rectification" "looking back."

One point in the materials I quite agree with: product positioning needs to distinguish between general-purpose language models and vertical creation tools. General models have broad knowledge but can't cross into your unit's specific context.

That's where WorkBuddy's advantage lies—it first has you feed in the materials, then writes according to the materials. Two months ago I wrote a post about using it to organize red-header documents; that workflow is still in use, just with the scenario changed to writing summaries.

Step one: build the filing cabinet first, don't rush to generate

Open WorkBuddy, there's a Projects in the left sidebar, click New Project, name it whatever—I called mine "2026 Annual Summary." Once inside, drag three types of files in. One copy of last year's equivalent summary, this year's key documents and three meeting minutes, plus two or three of the leader's speeches from the past six months. After dragging, it will index these files—let the tool memorize your unit's terminology first, and later when generating it will follow this rather than the generic tone from the internet.

Then in Project Settings change two items. First, Output Format—select Word, body text in SimSun size 3, level-one headings in SimHei. Set this once and you save effort from then on. Second, turn on Citation Sources—once enabled, each paragraph will be marked with which file it came from; anything unmarked means it made it up itself, and you can spot it at a glance.

This item is more important than anything. As I said before, the one-click generation moment is the least valuable—what's valuable is that every sentence can be traced.

Step two: have it do only one thing at a time

My first time using it, to save effort, I dumped four things into it at once—"write the summary, compress to 3000 words, fix the format, add subheadings"—and what came out was a mess. Later I learned better and split it up. First have it produce an outline, I revise it, and after revising click Lock, meaning the structure is fixed and it's not allowed to change it on its own afterward; then write section by section according to the outline, one section at a time—after writing I review it, and if it's wrong I have it rewrite that section; finally issue a separate instruction to unify format and font size.

This is a bit slower, but less rework. The first time I tried to get it done in one step, I ended up spending an extra evening cleaning up.

Step three: permissions and daily maintenance

In the top right of the project there's Collaboration where you can add people. I gave Xiao Wang from my group read-only access—he can view but not edit. If I really need him to edit, I export to Word and send it over, he edits and sends it back, and I overwrite it—no two people touching the same file at once, to avoid version conflicts.

Also a small habit: every Friday spend ten minutes adding the files received that week into the project materials. If you don't, it will still be citing terminology from three months ago, and the numbers written out won't match—I've stepped on that one.

My environment is a unit intranet machine with average specs; running these hasn't lagged, but a different environment may not be the same—try it yourself with small materials first. My suggestion: don't rush to sign up for a membership—pick the most annoying document you have on hand and run through this workflow from start to finish; once it works, then talk about anything else.

2 replies

?
Ctrl + Enter to reply
Zhe Dan Bai De

I'm straight-up stealing the trick of enabling citation sources, but hands-on experience tells me: once the file versions you feed in get messy, the traceability chain is harder to audit than the model just making things up.

Da Wei
Da Wei5d ago

Spending ten minutes every Friday updating the reference material—that habit is way too real. If you don't update it, it's still citing the old specs, and mismatched numbers mean even more rework.