After two weeks with WorkBuddy, I finally understand the gap between it and ChatGPT

Title: Used WorkBuddy for Two Weeks, Finally Understand the Gap Between It and ChatGPT

Last Wednesday afternoon, I did my usual routine of throwing a bunch of reimbursement PDFs into WorkBuddy, asking it to organize them into a table. Previously, I had to manually open every invoice, check amounts, dates, and payee names, then fill them into Excel, taking over an hour. This time, I created a folder, stuffed over a dozen PDFs and screenshots in, and said, "Categorize by month, summarize into a table, leave the remarks column blank for me to fill." It broke down the task itself, recognized the content itself, and the exported table format was correct, amounts were right, saving me an evening. I didn't even watch it closely.

But what really made me pause and think was that I just started using ChatGPT today and tried the same task. Not to trash-talk anyone, but the contrast was stark. ChatGPT's dialogue window is smart; it tells me, "You should organize invoices first, then check amounts, finally summarize." And then? Then I still have to do the rest myself. I have to manually upload PDFs one by one, let it recognize them, then copy results into the table. In other words, ChatGPT is a good consultant; WorkBuddy is a colleague who gets work done.

Multi-Agent Isn't Just Marketing Jargon

The review in the materials mentioned WorkBuddy's core is multi-agent orchestration, not single-turn chat. I didn't take it seriously at first, thinking it was just promotional talk. But after using it these past few days, I found the difference is real. I asked it to do competitive analysis; it scraped web data, summarized uploaded reports, and assembled slides simultaneously. Three tasks ran in parallel, finally giving me an integrated output. I'm not tech-savvy, can't name architectures, but I know this feels completely different from the copy-paste back-and-forth I used to do in ChatGPT.

My understanding is that WorkBuddy breaks a large requirement into small chunks, assigns them to different "little employees," who work independently and then aggregate results. It's like how I arrange work: I don't ask one person to handle reimbursements, reception, and booking meeting rooms simultaneously; I assign different people. It embedded this logic into the product. So it's not the same category as chatbots; comparing "who chats better" is meaningless.

Of course, not everything goes smoothly. I tried letting it process a skewed photo of an invoice; the amount was recognized correctly, but the remarks column was missed. I learned my lesson later and double-check important documents myself. Another time, I asked it to organize a month's attendance records, and it mixed up two employees with the same name. I make mistakes like that too, but when AI makes mistakes, they're harder to catch because its answers look too neat. This is my biggest concern currently: the faster it works, the easier it is for me to let my guard down.

From "Super Individual" to "Super Team," I Believe in This Path

Tencent initially positioned WorkBuddy as a "super individual," then moved towards "super team." The materials say a small team of a few people built the initial version in two days, maintaining a two-day iteration cycle post-launch. I haven't done development, so I don't know what speed that represents, but product updates are indeed fast; the interface from two weeks ago differs from now.

My feeling is, what impresses me most about WorkBuddy isn't how smart it is, but that it actually handles "work." Every day I face reimbursement forms, contracts, meeting minutes, supplier spreadsheets. These don't require deep reasoning but need patience and meticulousness. Previously, I relied on my own patience and detail; now I rely on it. This transition feels natural because I don't have the energy to learn various prompt techniques; I just need to state clearly what I want.

But I'll speak frankly. I previously wrote a post saying "Tool rankings belong to others; workflows are yours." After two weeks, I confirm this judgment. No matter how strong WorkBuddy is, it's just a shell; the key is you must figure out what you want to do. When I started, I took detours, trying to state all requirements at once, resulting in chaotic sub-tasks. Later, I learned to let it do one thing at a time, which became faster and more accurate. It's like training a newbie: if you don't explain clearly, no matter how capable they are, it's useless.

Who It Suits, Who It Doesn't

Based on my tests, WorkBuddy suits people with lots of miscellaneous, repetitive work with clear processes. Admin, finance, operations, e-commerce, HR—these roles are drowned in trivia daily and need this "get-it-done" tool. You don't need it to be creative; you need it to not complain, not tire, and not take leave.

It's not suitable for everyone. If your daily work involves writing proposals, creativity, and extensive divergent thinking, WorkBuddy might not help much; it's more of an executor than an inspiration source. Also, if your company is extremely sensitive to data security and allows no uploads, this tool is useless to you. I use it myself but am careful not to put overly sensitive things in.

Following this thought, I see deeper issues. WorkBuddy can do work for you, but can it take responsibility for you? If reimbursement amounts are wrong, I bear the responsibility, not it. It's just a tool, but precisely because it's a tool, it has no fear or consequence awareness, so it won't repeatedly check like I do. This is a common hurdle for all "work-doing" AI tools. I don't think WorkBuddy has crossed it yet, but its direction is right: It at least does the work; checking, covering bases, and taking responsibility still fall on humans.

Trend Judgment

The review in the materials said, "Most AI tools only give you a chat box." I think this hits the nail on the head. In the next year or two, chat-box-style AI assistants will become increasingly insufficient because users want results, not suggestions. WorkBuddy's current path is moving from "talking well" to "getting things done," and I bet on this direction. But having said that, when it can truly replace me in doing all work, better and more stably than I do, I might have to think about what else I can do. But that's for the future; for now, I'll let it organize next week's reimbursements first.