WorkBuddy hits 30M+ MAU, but my construction-site workflow hasn't changed in three months

WorkBuddy's capabilities are enough now, but people like me still don't know where to click next.

A few days ago I came across news from the Tencent Digital Ecosystem Summit: WorkBuddy has 32.353 million MAU, up 647.1% quarter-over-quarter, iterated over 100 major versions in 7 months, and has cumulatively handled 9 billion task requests. The numbers are pretty scary. I sat in the project department's prefab office reading it, then looked back at my own screen—still that same interface, still those same two things: sorting a pile of PDF submission forms, and piecing together the weekly report.

This gap isn't because I'm lazy.

I've been using WorkBuddy for a full three months now. Back in early August, I used it to organize three days' worth of construction site materials, and felt this thing was reliable—it could gather supervision notices, material submissions, and concealed acceptance records scattered across a dozen folders into one place. To this day, my usage hasn't really changed.

I read a few things from the summit: Buddy One, with WorkBuddy as the workbench, linking Tencent Docs, Tencent Meeting, ima, and Tencent Drive together, with unified accounts and credit pools. Enterprise edition open platform capabilities, upgraded security, observability, and hosted agents. Enterprise edition deployed in 50+ industries, covering half of the China 500.

From what I've tested on my end, the feeling is different.

Yesterday was the first time I clicked into the ima knowledge base and subscribed to a construction-related library. After subscribing I got stuck—didn't know what to do next, and even made a post asking about it. Today I read that ima handles knowledge management within Buddy One, which makes logical sense. But for someone like me whose personal library has never been organized at all, connecting several unorganized things just gives you a pile of unorganized things again.

Dowson Tong said something quite honest: enterprise customers report that "models are already strong, and there are plenty of tools, but when it comes to actually using them in the business, there are still many breakpoints." I agree with that. But I have to add: the breakpoints aren't on Tencent's side, they're on mine.

My real information is all scattered: the supervisor's verbal requirements are in voice messages in WeChat groups, thirty seconds each, and speech-to-text still gets two words wrong; the actual on-site progress is in my phone's photo album, forty or fifty shots a day; material entry records are on handwritten signed forms, in that scrawled handwriting; changes and disputes are in meeting room conversations, and nobody records them; and the final deliverables are in Word and Excel, in whatever format the client specified.

Buddy One's logic is to connect documents, meetings, knowledge bases, and cloud drives so the AI gets full context. This design works smoothly for a company with an OA system, a knowledge base, meetings on Tencent Meeting, and documents all in Tencent Docs. But on my end, meetings happen in the prefab office, minutes are typed up by me, and documents are on a local hard drive. No matter how well the system is connected, it can't reach my handwritten form.

So what's blocking me has nothing to do with whether WorkBuddy's capabilities are sufficient. Its capabilities grow on the branch of organizations that have processes, while people like me are process-less, sole-proprietor-style workers.

There's also something more practical. The pitfalls I've hit over these three months basically all come from asking it to do too much at once. It's not that I don't know how to use the features themselves. Last week I tried throwing three things at it at once—categorizing submission forms, generating a summary, and writing the weekly report—and the middle step got messed up, with two items missing from the summary table. Later I split it into three separate conversations, one thing each, and it was actually faster. This has nothing to do with whether the ecosystem is connected; it's my own usage problem.

Over 100 versions in 7 months—that speed shows the direction is still being adjusted. The enterprise edition stuff—security governance, observability, credit consumption backend, admin permissions—these are must-haves for companies with big IT departments. For someone like me, one person with two interns, who sees more hard hats than hair, they're just decoration.

For teams whose information is already digital and whose processes are already online, WorkBuddy is now something you can pick up and work with directly; for construction sites where half the information is on paper and half is in voice messages, it's still an advanced file organizer. That's not a criticism—being able to organize files properly has already saved me quite a few evenings.

But I'm not satisfied with that.

What I really want is for it to remember me—how many products it connects to is secondary. For example, last week I categorized the submission forms for a certain project; this week when another batch comes in for the same project, can it just arrange them according to last time's directory structure without asking me again? Or, if I drop a sentence in WeCom—organize the three rectification opinions the supervisor sent today into a reply form—can it go find that voice message itself, transcribe it, and apply the template I used last time? These don't require connecting five products; they just require it to remember what I, one person, have done.

The bigger the ecosystem, the more there is to learn, and for someone like me who eats dust on site every day and only has time to touch a computer at night, that's not necessarily a good thing. It's not that I don't learn—over these three months I've tried Codex, ima, and Cherry Studio one after another, but every extra entry point is one more moment of hesitation about whether I clicked the wrong thing.

I read the summit news three times. The first time I thought it was impressive; the second time I thought it had nothing to do with me; the third time I thought maybe I just haven't used it deeply enough.

Or maybe I'm misunderstanding it—platformization was never designed for people like me; it's for companies that can assign admins, define permissions, and map out business processes. Once that batch of companies gets comfortable with it, and the capabilities trickle down further, then it'll be my turn.

Until then, people like me—no processes, half the information on paper—will probably just have to make do on our own.