WorkBuddy Moves Into WeCom, and It's Not the Bot I Care About

Saw a post today saying WorkBuddy has integrated with WeCom. WeCom 5.0.12's sidebar has a new "Smart Robot" entry — click in and you can claim a WorkBuddy robot of your own, and once claimed it sits in your chat list and you can talk to it directly. The post also said that asking it to summarize a one-hour livestream took about fourteen minutes, and midway it suggested pausing on its own, saying continuing would repeatedly burn credits.

My first reaction after reading was: how are permissions split?

Whether an AI colleague is reliable mainly depends on how much it can see and how much it can change. I've stepped in this pit in my own department.

My setup is both WorkBuddy Personal and Enterprise installed — Personal for three weeks, Enterprise for four, plus one on my tablet. Different versions may have menus in different places, so the following may not apply to everyone. Let's start with the shift schedule.

Open WorkBuddy, and the gear in the bottom left is Settings. Click in, and there's a Workspace column on the left, and further down Permissions. There's a toggle here controlling file modification, and it's open by default. I didn't touch it at first, and as a result when it tidied the shift schedule it would fill in data on its own — who's off that day, who's on night shift that day, if it felt it flowed it just filled it in for me. Later I changed it to ask me before touching files each time, and every time it hits a number it pops up for me to confirm. Slower, but the ledger doesn't go wrong.

One level up is Skills. The word Skill is confusing the first time you see it — it's just a rule you write down, with higher priority than global settings. The path is Settings to Skills, plus sign in the top right, create a new one, and write the department's shift rules into it. Who doesn't do night shifts, who can't be scheduled two days in a row, who needs a standby on weekends — write it all clearly. After writing it, you don't need to attach it manually; it takes effect automatically at the same level. I wrote a post about this priority thing — configure it at the wrong level and everything after is a pit.

The model part also has a pit. Settings to Model, plus sign in the top right, provider set to Custom, fill in four steps and save. Saving doesn't mean it takes effect. You have to go back to the top right of the chat window and manually switch the model over, otherwise it's still using the old one. I posted about this on October 21 and still haven't figured out whether I missed a step or it's just like that.

The Enterprise side is a different setup. The admin backend can centrally purchase, allocate seats, and adjust quotas; the console centrally configures models — Hunyuan, DeepSeek are both selectable — and can also upload enterprise private skills and MCP connectors. MCP connectors are the opening to hook in internal hospital systems; our department didn't connect it, we don't have that permission. The quota thing is quite practical — I gave two colleagues read-only, they can view the spreadsheet but not edit, and only I can edit. That way the ledger won't have two versions.

Daily ops is just one thing: go through the config once a month. Because the Skills layer has higher priority than global, changing global doesn't take effect, and over time you forget what you changed. I usually glance through it at month-end when reconciling shifts.

One more pit to avoid. Don't keep clicking continue on long tasks. The post said continuing repeatedly burns credits and run quota. Same for me when processing large batches of spreadsheets — throw dozens of files in at once and it tends to get stuck midway; splitting into two batches is actually more stable. Being in a hurry doesn't help.

The official line is "the moment AI enters an enterprise it naturally has context" — you say "that project" and it knows which project. Sounds nice, but the more context you give, the tighter permissions have to be pulled. These two things move together.

My judgment is that this kind of robot living in the chat list will become the default form within a year — whether to use it isn't a multiple-choice question. But whether it can get into a department depends on whether permission tiering is done, and the line between Personal and Enterprise will get clearer and clearer.