*I don’t believe you need to become a “prompt engineer” to succeed with AI coding - by contrast, all you need to do is write very clear, simple specs. *
Some of the most common pitfalls that my team has been noticing, come down to users overcomplicating their prompts and confusing the AI.
Roundabout wording, with long sentences, can make it unclear which files should be edited and why they’re being mentioned. Instead of long paragraphs, I recommend structuring clear sentences, using a formula such as “Given, Where, When”.
Some people fall into **the trap of combining catch-alls with specifics **in their prompts. Each prompt should be either specific or general. It’s also OK to start off with a general query - let’s say, search the repo for dependencies on a component, before you write the specs to change said component. Or if you prefer, after giving a task, follow up with a catch-all question such as, “Now that we have changed ____, what else can or should be edited in this file to match?”
The most ridiculous “prompt engineering” that I see is when users end up contradicting themselves. “Fix this bug but don’t change any of my code”. Well, how is the bug going to be fixed?
**Also - you don’t need to start every prompt with “You are a senior software developer” anymore. **It’s 2025. (At least not when you’re using Fine… I can’t speak for every platform out there).
Remember, in most platforms you can add overriding rules or instructions which form a part of the agent’s flow with every task. In Fine it’s project based - you can add custom instructions (style, language, commenting etc) which apply to all tasks on the repo, but don’t confuse the agent when reading the specific task.
Author Of article : Dror Wayne Read full article
