After three weeks with WorkBuddy, I deleted all those official document writing tutorials from my Taobao favorites

Conclusion first: Spending money on material libraries and writing tutorials is worse than getting WorkBuddy's processing workflow running smoothly. The remaining bits that truly require brainpower are things neither tutorials nor AI can help you with.

Recently, my leader assigned a quarterly summary. As usual, I opened Taobao wanting to stock up on materials. Searching "summary official documents," the Top 100 sold everything imaginable: Official Document Writing Masterclass Volumes 1 & 2, High-Level Material Writer Collection, Government Agency Official Document Examples Encyclopedia, AI Prompt Template Handbook, and even paid communities like "Veteran Insider Guide to Writing Hand-in-Hand." I almost placed an order again. Then I thought about it. I know this trick. The material pack I bought last time contained mostly content stitched together from online sources; very few items were actually usable.

Coincidentally, around those days, I saw news about the RAISE Summit. A bunch of Silicon Valley big shots were chatting about the future of AI at the Louvre in Paris—Llama publishers, AMD CTOs, Databricks founders, and Replit presidents. Seeing them discuss on stage how AI agents reshape workflows, then looking down at my Taobao cart full of "Official Document Writing Masterclasses," I suddenly felt like someone still using an abacus while the outside world was discussing quantum computers. So I cleared my cart and decided to seriously master WorkBuddy instead of continuing to hoard dusty tutorials.

After using it for over a week, I first threw in the semi-annual summary materials my unit needed to report. Previously, manually organizing fragmented summaries reported by various departments took at least two nights of staying up late. Now the process runs like this: Throw a dozen Word and PDF files into WorkBuddy, let it classify by line first, then automatically extract core practices and numbers from each document, generating a summary table. The first run was generally reliable in recognition. Two PDF scans had poor quality; it recognized them but classified them incorrectly, requiring manual dragging. The key was one-click draft generation. It could blend materials from all departments into a structurally complete summary, laying out boilerplate sections like "Existing Problems" and "Next Steps" for you.

But there's a pitfall here I must clarify. WorkBuddy isn't ready-to-use out of the box. Its default action is "generate a complete summary document," and it really takes liberties by rewriting the wording in your original files. The first time I ran it, I found it changed a department's written "Focus on promoting" to "Vigorously promote." Such subtle differences are major issues in government agency materials. If the original department staff saw it, they'd surely get angry. So afterwards, I added a rule in folder monitoring: Any file entering the folder gets backed up as an original copy first. WorkBuddy can only read copies, not touch originals. This trick came from accidentally stepping into a pit. Now I always confirm monitoring is running before processing.

How to use it well: Don't treat it as an omnipotent player that "can write anything." Treat it as a quick-handed intern. Give it sufficiently clear instructions and boundaries, and it saves you lots of repetitive labor. If you vaguely toss in a line like "Help me write a summary," it returns a bland, stable document completely lacking your unit's characteristics. Later, I set its role as "Government Secretarial Assistant," requiring strict use of our unit's customary expressions, no creative brainstorming by default, and absolute fidelity to original files regarding numbers. Only then did it truly run smoothly.

To be honest, WorkBuddy's strongest part remains batch processing. Last week, my leader temporarily requested a rural revitalization report PPT. From notification to report, there was only one afternoon. I collected materials from five villages, threw them into WorkBuddy to first extract core data for each village, then generate an outline, and finally apply our unit's reporting template for direct layout. About forty minutes later, a draft PPT of twenty to thirty pages was ready. I only needed to change a few numbers and reorder some paragraphs. Previously, this kind of work took me at least one night, relying on image editing software to screenshot everywhere. This week, I compared Gamma for the same PPT. Gamma's aesthetics are indeed online, with smooth chart animations. But the problem is it doesn't understand the logic of government agency materials. The output looks good, but showing it to leaders would result in rejection—because it fails to grasp the "three-part" structure and hierarchical relationships. WorkBuddy's layout is practical-oriented. It understands that government materials need conclusions on page one, background on page two, and specific practices afterwards. This matters much more than looking good.

But frankly, WorkBuddy still feels like that quick-handed intern who needs supervision. Three weeks ago when I first touched it, it easily misunderstood instructions like "Generate a weekly report based on document content." Now it has evolved significantly, able to distinguish between "summarizing the past" and "looking forward to the future." However, it still lacks finesse in understanding subtext and subtle expressions in government materials. For instance, phrases like "leveraging external forces" versus "developing by borrowing strength"—one is neutral, one is positive—it can't quite tell them apart.

A colleague asked me, "Will writing get worse using WorkBuddy?" My thought is: At most, it's a porter, moving materials from various departments and integrating them into a framework. The real flesh and blood must be filled in by yourself. Those three measures in "Next Steps," I still have to figure out and write myself, because only I know where next year's work priorities lie. The tool saves you mechanical organization time, giving you more energy to think about methods and strategies. This is probably the correct relationship between AI and writing materials.

By the way, recently attending the RAISE Summit wasn't just Silicon Valley big shots; even Nasdaq presidents and NVIDIA European VP attended. I only skimmed some reports. The core topics were AI computing power and energy, indicating generative AI tools are no longer rare. Everyone is starting to compete on more fundamental resources and infrastructure. After seeing this, I felt relieved. Tools will only become more widespread. Getting the tasks at hand running smoothly sooner is better than anything else. The only regret is it can't write the phrase "I'll briefly say a couple of sentences" in "Leader Speech Drafts"—because behind that sentence often follows ten thousand characters. That indeed requires real skill. But helping me accumulate the preceding nine thousand characters is pretty good too.