A task is more than a prompt
A few notes on the work that happens before asking an AI to build something.
Sample text for the reading-page design. This is not a published article by Anh.
An empty prompt box makes every task look like a writing problem. Find the right words, send them, and wait for the result. But deciding what to ask is often the larger piece of work.
Imagine asking an assistant to improve a page. That could mean shortening the copy, making the navigation clearer, or fixing a layout that breaks on a phone. A polished result can still solve the wrong problem.
Name the thing that should change
Start with an observation someone else could check. The title wraps awkwardly on a narrow screen. A reader cannot tell which card opens an article. There is no way back to the collection. Each observation gives the task a useful boundary.
Then describe what a successful result would let someone do. For a writing gallery, that might be choosing a topic, opening a post, and returning to the same collection. A screenshot can help explain the visual direction, but it cannot explain that whole interaction.
The useful question is what someone should be able to do when the work is finished.
Leave room to discover the answer
A task does not need to prescribe every implementation detail. It needs enough context to explain the goal and enough constraints to rule out unwanted results. The person or agent doing the work still needs room to investigate.
The first result is a chance to check the original question. If it misses the point, revise the task with the specific thing that was missing. That observation is more useful than another request to make it better.