After a Month with WorkBuddy, I Shifted from 'Essay Ghostwriting' to an 'Evidence-Based Pipeline'

Title: Used WorkBuddy for a month, changed it from "Essay Ghostwriter" to "Evidence Pipeline"

The most useful thing about WorkBuddy for elementary school teachers is fixing the most annoying repetitive actions in essay critiques: collecting files, finding original text, listing follow-up questions, exporting tables. If it can only be asked "What score is this essay?", it's an expensive but unreliable grader; only if it can spread out evidence item by item according to my workflow does it look like an office assistant that saves time. WorkBuddy serves as a traceable grading pipeline here.

Friday afternoons are the worst for organizing after grading. Our class submits essays on Thursdays, and critique sessions are on Friday afternoons. The original process was fragmented: collecting student drafts from WeChat (some images, some Word or WPS docs); using image-to-text or opening in Word, reading one by one, writing comments, organizing common errors, making a slide deck, and replying to parent messages. My tests show it takes about 2.5 to 3 hours before a critique session.

Previously, I used Wenxin Yiyan, DeepSeek, and Kimi for similar tasks. They give comments quickly, but the problems are obvious: conclusions are too smooth, evidence is unstable. Sometimes it says "Description isn't specific enough" but doesn't point out which sentence. When students bring comments to ask me, I have to flip through the original text again. Time saved shifts to re-verification.

I also stepped on pits when starting with WorkBuddy. Initially, it couldn't read essay drafts in my folder; task context and file permissions weren't aligned. I thought it "promised but couldn't find things." Later, I changed to putting essay drafts into WorkBuddy's workspace first, naming files uniformly as "Student ID + Name + Title," then letting it process via template. This step is clumsy but stable.

Now my WorkBuddy flow is fixed: First, create an "Essay Critique" workspace, putting essay drafts, scoring rubrics, and student self-evaluations inside. Then let it batch extract three types of info from each essay: an original sentence, a follow-up question, and a suggested revision direction. Output as a table with fixed fields: Name, Original Evidence, Question Type, Student Needs to Supplement, Teacher Review. It only gives the teacher version, not directly generating student version conclusions. After students get the table, they must first write "Do I agree with this evidence?" then add a sentence in their own words.

The key in this prompt is "Extract only, do not judge." For example, if a student writes "He was very sad," WorkBuddy outputs original evidence "He was very sad," follow-up "Can you add an action, expression, or environment?", suggestion direction "Change from result word to process description," and won't directly say "Emotion is thin." This makes my review fast because I know it's speaking based on the original text; evidence lands on the original text.

Running this, initial screening tables for about ten essays come out in 8 to 15 minutes. Not always correct; individual student names might be missed, and spaces/punctuation get messy after image-to-text. But it shortens the most time-consuming part: "finding evidence in essays." Originally I had to flip through one by one; now I just pick errors.

A more realistic effect is it helps me leave traces. Last week during critique, a student asked why I said his writing was empty. I opened the table exported by WorkBuddy; the original sentence was right there in that row. Seeing the four characters "He was very sad," he laughed himself, saying it seemed he indeed only wrote the result. In that moment, I felt this tool had some educational value; it laid out the evidence students try to evade.

Recently I scrolled through an AI office agent comparison article. The reasons given for WorkBuddy's ranking were roughly balanced ecosystem connectivity, autonomous task execution, and office delivery, but points cost transparency is a weakness. I basically agree with this judgment. However, from our teaching group's usage, WorkBuddy has another issue: the local file entry is too unintuitive. You need to know how to hand files to it.

Some comparisons say WorkBuddy's strength is the thickest ecosystem and good task linkage, while weaknesses are points and cost transparency. I basically agree, but elementary teachers ask one more question: Can it leave evidence traces?

I compared it with Qianwen Office and QoderWork. I used Qianwen Office beta for about two weeks; it produces neat notifications, courseware frameworks, and scoring rubric finals, like a "slow craftsman," suitable when you need a document ready to send. WorkBuddy's advantage is linking files, tables, follow-ups, and reviews. QoderWork leans more towards professional plugins and role-specific workbenches, heavier engineering vibe, suitable for those who like tinkering with processes. For ordinary elementary teachers, I don't need it to write code; I need it not to mess up parent notification formats.

So if comparing only "who generates prettier lesson plans," I don't think WorkBuddy definitely beats Qianwen Office. But if comparing "can it go from collecting drafts to critique tables on Friday afternoon," WorkBuddy suits me better. It's like an assembly line.

Where it truly saves time is spreading out evidence. This is also why I remain cautious about AI grading. If students submit homework with comments generated directly by WorkBuddy, that's dangerous because they might not even have looked at the original text. Our current requirement is: AI handles initial screening, students must answer with evidence, teachers make the final judgment. I've told parents in groups: AI tools help teachers do less repetitive work, saving time for questioning.

My attitude toward WorkBuddy is clearer now than when I first started using it. Teachers willing to break essay critiques into steps, needing batch processing of Word tables/notification materials, and treating AI output as first drafts will find it smooth to use. Those wanting one-click comment generation sent directly to students, or unwilling to review original evidence, will find it troublesome.

I hope it adds a few small features. Teacher review mode forcing every row to cite original text; no citation, no table. Student follow-up mode generating only questions, not answers. Local file permission checks telling me directly which step failed to mount, don't make me guess. In the teaching group, we've also organized materials using Spark, Wukong, and the National Smart Education Platform, but the reason WorkBuddy stayed is it can connect these scattered actions into one task.

Used WorkBuddy for a month; I didn't hand over essay grading, but rather dared to use it more. Because it forced me to break "what is a good essay" into checkable evidence. The remaining time, I spend arguing with students about whether that one sentence was actually specific.