I've been organizing head-too-head match-ups of various flavors of GitHub Copilot and Cursor, to keep from being bored sending resumes out into the void.
In the past I have used AI: Code-completion at MassMutual, code review at SELF, and vibe-coding apps for fun during Christmas break.
The most powerful use I've found with AI? Research Assistant:
- "Here is a list of toolsets. Describe them. Be brief".
- "What are the release dates of these toolsets?"
- "Use corporate tech blogs as primary sources".
- "Cite your sources. Provide links".
These prompts provide excellent documentation that allows me to deepen my own toolset education.
When I ask AI to build this study guide, AI is teaching me how I can teach myself.
When I ask AI to do something for me, the only thing its doing is teaching me is how to craft a prompt.
See, when AI does eventually screw up, the only thing I can do to "fix" it is try to craft another prompt, hoping that maybe this time it can see and correct its mistake. And if it can't... ask it again? And again? Tenth time is the charm?
Where I find AI fouling up time and time again? Creating CI/CD pipelines using GitHub Actions Workflows running automated tests in Android emulators.
- It doesn't realize that GitHub actions have new versions that have been released in the past year.
- It declares you should be using a Mac runner since it is more stable. No, a Linux runner is better! No, a Mac runner! It flip flops on them between code reviews.
- If you ask it to shift to a build / test / report stage, it always forgets to upload the artifacts in one stage so they can be downloaded to another.
- It suggests Intel-based Android emulators to be used on Mac OS runners, only recognizing how it fowled up if you copy-and-paste the error to it.
- It will erase and change the comments you placed in the workflow if you do not watch out.
- It keeps wanting to go out of the box you placed it in, and rearrange disorganized code that runs elsewhere in the project, and change it to pretty code that has hidden errors in it.
Why do they get iOS workflows mostly right and Android workflows mostly wrong? Who knows!
All I can say is that if you are going to add AI into your workflow, learn some breathing exercises. You will need them to work through the frustration you are about to face.
Want to see other projects where I have used AI? Check out the list of programming projects on my blog, where I have sample code I have written for the past ten years at https://www.tjmaher.com/p/programming-projects.html
What do you find AI constantly screws up? How do you fix it? Leave comments below!
-T.J. Maher
Software Engineer in Test
BlueSky | YouTube | LinkedIn | Articles
No comments:
Post a Comment