812: AV & Vibe Coding with Tyler McRae

Episodes
1
3 Likes
2

I enjoyed hearing Tyler’s story. I ran sound and did some tech support for my church for several years.

Many years ago I worked for a freight company in Arkansas and had customers in the Nashville area, Mena and De Queen, etc. Beautiful country.

2 Likes
3

Re: AI coding assistance, I have found the following useful:

  • When I have working code but am thinking about refactoring a certain way, AI can save a lot of time retyping things into the new structure. For example, if I have a data structure and several functions that work together, I might decide to create a class object to bring it all together. And then I need to reimplement the object everywhere else in my project. I know how to do that, but it takes time/effort; whereas AI can do it in seconds.

  • When I need to create dummy data with a certain structure, I can just describe what I want and AI does a pretty decent job of creating that. “Create a JSON representation of a list of book objects, each of which is a dictionary with author, publication date, date that I read it (set to None if unread). Give me a sample dataset with 100 books.” Then I use the fake dataset to work on the rest of my program.

  • When I want to look up how to do something in code, it is now more useful to query AI than to read reference manuals. Between my IDE’s built-in support and AI, I rarely have to search the docs myself. AI has the extra benefit that any mock-up code can use the variable names that you specify, so it is easier to integrate with the rest of your program.

  • If I am facing an error message or bug in a program, AI easily explains what might be the cause of the problem and often suggests a solution. It beats spending an hour reading snarky comments on StackOverflow.

  • If I need to convert code from one language to another, AI is good at that.

What AI coding support is not good at is larger-scope software architecture. It is a code-generation engine, not a software-design expert. But if you know what you’re doing, you can use it to build the draft scaffolding for different architectural design choices. By working through those prototypes, you can make a more informed choice about which design pattern to adopt.

Cutting-n-pasting code into ChatGPT or equivalent is not the ideal workflow. Thankfully, companies like JetBrains are building code-specific AI into IDEs like PyCharm. I think this will be the future, with little helpful agents sprinkled throughout the OS and apps. (Anybody remember Clippy?)

2 Likes