Discussions
QwenWorkThe five debates every rollout has
Not a forum index — a summary of the arguments, with the position that usually wins in practice and the cases where it does not.
1. Should personal space be allowed at all
Security teams want it off: unmonitored storage is a data-loss channel. Practitioners want it on: people need somewhere to think before they share. The position that usually wins is keep it, with one written rule — nothing that came from a client goes in there. Removing it pushes the same behaviour to personal cloud drives, where you have even less visibility.
2. How do you judge output quality
The debate is between 「read everything」 and 「spot-check the facts」. In practice, documents get read once and sheets get spot-checked. The rule that survives: the reviewer reads numbers, dates, names and conclusions, in that order. Prose quality is a distraction from the four things that cause actual incidents.
3. What happens to the legacy macros
Three camps: rebuild as automation, rebuild as a script, or keep one Windows machine with Excel on it. All three are in production somewhere. The deciding factor is not elegance but whether anyone left who understands the logic — if nobody does, the machine that runs it is the cheapest form of documentation.
4. How much automation is too much
The signal is not the number of tasks but the number of failures nobody reads. If a task's output folder has not been opened in a month, it is either unnecessary or broken. A useful test: for every automation, name the person who would notice it failing. No name, no automation.
5. One model for everyone, or let people choose
Standardising gives predictable cost and comparable output; letting people choose gets better results on specialised work. Most organisations land on a default with an allow-list, and revisit when a team can show a specific task where another model is materially better. Deciding this on principle rather than on a task is how you get a policy nobody follows.