Prompt engineering wasn't on the list. I asked Qwen what I should learn if AI changes my work, and it came back with five skills in order, plus three things to skip. Build one ugly thing is where I'd start.
What Qwen told me to learn
Qwen put task decomposition first, ahead of prompt engineering. You run each step through AI and see where it breaks, and that failure teaches you more than a course.
The skip list is short: an AI bootcamp, collecting tools instead of mastering the one stack you already have, and waiting until you feel ready. Start this week.
- Task decomposition. Cut a messy work problem into steps an AI can handle reliably. Pick one recurring workflow and break it into 5 to 8 steps.
- Build one ugly thing. Not a startup and not an app. A bot, a pipeline, a script or an internal tool that saves you 2 hours a week.
- Audit, not only generate. Know when to trust AI output and when to re-verify it.
- Deepen one domain instead of widening. AI makes general knowledge cheap, so the judgment-heavy knowledge in your field is the moat.
- Enough code to not get stuck. Read errors, reuse scripts, debug when something breaks. Python is the shortest path.
Tools and links
- QwenThe model I asked what to learn if AI changes my work.
