Raw LLM Responses

I used to be an unbeliever, but not any more. I see prompt engineering becoming part of the interview process, if not the entire interview, in future job applications. The prompts you wrote to guide your coding agent seem rather sloppy, which is why you got sloppy code. For the past 4 months (September to December 2025), I was tasked with building out a web app at my work. The frontend was written in Astro + React, while the backend was based on Litestar with Sqlite and Redis. AWS cognito was used for authentication, and I also included a custom Admin dashboard that was only made possible thanks to SQLAdmin. Throughout the process of building out this monstrosity (solo), I never wrote a single test. Fast forward to this January, I decided to start writing tests for the backend after having released the alpha version for our customers. I knew what tests I wanted to write, but I didn't want to actually write them. So what did I do? I installed test deps and wrote one test module for one of the API endpoints. Then, I asked Copilot to read the code base and generate agent prompts for a QA test engineer. After it did, I took the markdown file and made some modifications to it by including how test files should be created (location, naming scheme) and how their content should look, including how to create test fixtures. I also included a snippet of the test I had initially written to give it an idea of how to format the tests. Ngl, my journey to achieving the prompt isn't as straightforward as I make it sound. It took about a week of casual prompting to build this test agent, but the results make it worth it. Less than 5 minutes it takes to generate complete tests for each endpoint. It then takes me anywhere from 30 mins to two hours to read the tests and make sure they are sound. Also, when I want to generate a test, I don’t have to feed it long lines of prompts to do so. I just write something like: "Generate tests for the /api/foo endpoint," and it knows what to do, from file creation (if needed) to writing the actual test. I'm not saying that AI will replace engineers, but you would be a fool to think you'd get far in this field if you don't learn how to use it. I think the biggest challenge engineers are facing is that we've gotten used to doing everything on our own, so that when faced with the challenge of having to explain our thought process to an AI (which is really tantamount to writing well-documented code), we subconsciously put up a resistance to it thinking that it is there to hinder our progress. AI won't take over your job, but you will be replaced by someone who has mastered the art of prompting.