After Two Months with WorkBuddy, I Pulled Team Tasks Out of Group Chats
Community Discussion · General

After Two Months with WorkBuddy, I Pulled Team Tasks Out of Group Chats

I am DapengI am Dapeng1d ago2026/11/17 116 views

Last Thursday's review meeting ran two and a half hours, and by the time we wrapped up we'd settled on 11 action items. I remember clearly — 11, because I counted them one by one on the whiteboard.

On Friday I turned those 11 to-dos into a list and sent it to the group chat, @-ing the six people involved. When I checked again this Wednesday, three had moved.

It's not that nobody's working. It's that nobody knows "where this one stands right now." I go ask, and the answers are uniformly "working on it," "almost done," "doing it today." Not one of those three sentences can go into a weekly report.

I've known about this problem for a long time but never fixed it. I used to think it was a people problem, then a process problem, and finally I realized it's a carrier problem. A group chat isn't a carrier, and neither are meeting minutes. Meeting minutes are a snapshot, group chat is a stream, and a task needs a place where it can sit permanently and where its state changes.

The other day I came across that 2026 national competition prep tutorial, about three college students using WorkBuddy to set up a collaboration workflow — create a project, create a board, create tasks, upload files, four steps. One line in the article stuck with me: it said that on the second day of the competition, the three of them each insisting on a different answer to "which version of the data are we using" was more common than the model not being able to do it.

My first reaction after reading it was, isn't that exactly the situation with my 11 to-dos. The pits three people can step in over three days, six people step in more thoroughly over two weeks.

I copied it, and four steps became five

I've used WorkBuddy for two months, and before this I was basically locked into auto weekly reports and material collection. In August I wrote a piece saying unstable AI output is mostly a matter of input having no contract — that one was about feeding it material. This time I've moved one step forward and used it as a collaboration tool.

There's no real barrier to operating it — an ordinary office PC plus a browser is enough, and it runs without lag on my old ThinkPad. If I really break it down, my implementation here is five steps:

  • Create a team space, pull the six people in, split permissions by role — one tier each for product, dev, and QA
  • Create a board with four columns: To Claim, In Progress, To Accept, Archived. Just four, no more
  • Create task cards with five locked fields: owner, due date, deliverable link, acceptance criteria, current status
  • Set a naming convention — mine is "project_module_version_date" — and upload all files to the card instead of tossing them into the group chat
  • Go through the board every Friday afternoon, and anyone whose card hasn't moved explains why on the spot

The third one is key. I've built boards before, with other tools, and they died within two weeks. The reason was very concrete: there was nothing in the cards. A card that just says "optimize the registration flow" — three days later nobody knows how far the optimization got.

Now the card has a single named owner, not "the product team"; the deliverable link is attached directly — I drafted the PRD with Claude and pasted the link into the card; the acceptance criteria is a sentence that can be judged true or false, like "new user registration goes from 5 steps to 3," not "improve the experience."

Testing it over these two weeks, converting the 11 to-dos into cards took about twenty minutes doing them one by one by hand. The material says it supports batch import; I tried it and it can compress that to five or six minutes, but the header fields have to match exactly — if they don't, the whole batch errors out. This is the same thing as my August judgment: no contract on the input, and the output is a mess. So I just made a fixed template and fill it in from the meeting minutes each time, instead of improvising headers.

What works, where the pits are

First, what works.

The most direct change is the weekly meeting. Before, I'd spend the half hour before the meeting pinging people in the group chat for progress; now ten minutes before, I open the board and can see at a glance who's stuck. From my testing, "where's this one at" type questions in the weekly meeting dropped by about half, and meeting time went from two and a half hours to an hour forty. The saved time didn't go to real work — it went to arguing — but at least we're arguing about the plan, not the progress.

Second is files. Before, finding a competitive analysis from two weeks ago meant scrolling back through chat history, five minutes minimum. Now I open the card and it's there, and I know which version it is.

Now the pits — there are plenty.

Pit one is habit. The first week the board was up, everyone still said "I'm done" in the group chat and the cards didn't move. I put up with it for a week, then the second week I set a rule: saying you're done in the group chat doesn't count; if the card doesn't move, it's not done. A bit harsh, but it worked. The cost of a board is in maintaining it, not building it — I believe that now.

Pit two is over-splitting. I broke one registration flow into 14 cards, and everyone ended up moving cards around every day like they were playing a match-3 game. Later I merged it down to 5 and it got clearer. For teams of three or fewer with short task cycles, I'd suggest skipping the board and just using a checklist.

Pit three is the API. The material mentions you can hook up an API for automation; I'm on the standard version and didn't connect it, so I can't judge the auto-sync part — don't take my word for it.

Pit four is my own problem. I'm a talkative guy, and whenever I see a card I want to add a couple of notes, and by the end the card is full of my rambling and other people stop reading it. Now I've set myself a rule: notes under three lines, and if it's more than three lines, call a meeting.

Who should use it, who shouldn't

My judgment is this thing suits two kinds of people.

One is teams whose tasks span roles and run longer than two weeks. The kind where product, dev, and QA are bickering — a board puts the bickering out in the open. The other is short-cycle, high-intensity collaboration like the national competition — three people, three days, one version mix-up and everything's lost; in that scenario certainty matters more than capability.

Who it doesn't suit is also clear. Working alone — don't use it. Tasks that wrap up the same day — don't use it. And if nobody on the team is willing to spend twenty minutes going over the board every Friday — definitely don't use it; a tool can't save a process nobody manages.

One aside: Shushangyun recently launched an enterprise edition of WorkBuddy, and Tencent Cloud's DataBuddy is also built on the same Agent foundation as WorkBuddy. That info is useful to me — it means the workflow that runs smoothly on the personal edition doesn't have to be torn down and rebuilt when you connect it up to the organizational level. I don't need it for now, but I've noted it.

There's a line in that national competition tutorial that's pretty honest: it doesn't help you solve the math problem, it just helps you organize the solving process into a workflow that isn't a mess. Translated into product language: it doesn't improve your judgment, it improves the certainty of collaboration.

A tool can't cure laziness, but it can cure chaos. Of the 11 to-dos, 4 still aren't archived — I'll keep on them tomorrow.

2 replies

?
Ctrl + Enter to reply
Pixel Perfectionist

I agree with locking the card fields to five, but "notes no more than three lines" is hard to enforce—on my side, a designer adding an interaction spec in three lines is just not enough.

Si Nan
Si Nan1d ago
Reply to Pixel Perfectionist

Three lines is to prevent rambling, not to prevent explanation. If you really want to add interaction, just put the deliverable link in the card — he already said in the body that files go in the card, so don't stuff it into the notes.