Coding with AI

Warning: This is a post about AI.

I’ve always said I’m open to trying new things. And, to be fair, I have done. Over the past few years I’ve tried:

• Various PHP frameworks
• Making PHP composer packages
• Various Javascript frameworks
• Python
• C
• Rust
• Flutter/Dart
• Various cloud providers
• Various cloud services and APIs
• AI for creating code
• AI APIs

Some I’ve used to make software, others I’ve just worked through a tutorial, or tried out on a small test.

While all the languages, frameworks and services have taught me something and expanded my skills and usefulness, AI is probably in a different league of usefulness. Apart from anything, it’s not platform dependent.

My experience with AI for coding didn’t have the best of starts. I had a very complex bug that I had to fix, and AI was no help whatsoever. Then, I didn’t use it for a while, before going back and trying it again relatively recently.

The first thing I tried the second time around was to paste simple error codes into the message box to see if it could tell me how to fix them. It wasn’t terrible at very simple errors, I’d say about the same as googling the same thing, but it still wasn’t great for the errors I tried. If Google didn’t know how to fix, my specific problem on page one, the chances were that ChatGPT wouldn’t know either.

Next, I’d spent a few hours working on an algorithm the previous day, so I tried getting ChatGPT to make some code that did the same thing with the same inputs and outputs. This worked much better, in my opinion. While my code was clear and very readable for me, ChatGPT created some code that was much more concise in a fraction of the time. It even gave an explanation with a slightly different way of doing the same thing. As a human being, I prefer my solution but there is no denying that had I tried ChatGPT first I’d have got an algorithm that worked in a fraction of the time. Then, I could spend some time refactoring or changing the code slightly to suit my purpose.

Having had some AI success, I continued my AI journey by trying to ask ChatGPT to explain to me how I’d set up a specific type of CI/CD deployment. I’d not seen the exact process I wanted anywhere else online, so I thought I’d try AI. I don’t think I tried what it gave me and I think I had to prompt it a few times, but the outcome looked ok. It gave a discussion, explained what to do on the various services, showed you how to install the various packages and gave the example setup file.

Finally, my most recent attempt at AI was to ask ChatGPT for a list of free AI APIs. Looking at the list it gave me, I decided to use Microsoft Azure’s Text-to-Speech. So, I asked ChatGPT to tell me how I would use Text-to-Speech to create an audio file of text I had in a certain format, with code in a certain language on a certain type of server. While I found Microsoft’s docs were good, trying to figure out how I’d get it to work in my specific case may have taken me a while. With the AI assistance, I was able to make an audio file in a few minutes.

The Text-to-Speech AI API was extremely quick. Creating a 3-minute audio file took, what seemed like a second or less. Without AI this would potentially take me weeks to find someone to read the text, then wait for them to do it. The only issue with the text I gave it was that it was just a series of sentences, no paragraphs, so I’d have to figure out a way to add pauses. But, in general, a lot cheaper and faster if I just need an audio file and I don’t care about the tone or quality of the speech too much.

These are just a few examples of using AI in different software development tasks. There will be plenty more AI scenarios that AI can be used for. What I would say though, is that for commercial projects every single case I’ve seen so far would have needed checking. At the very least the code would probably need reformatting. Often the first prompt didn’t quite give you what you wanted, so you had to keep refining the prompts until you got something you wanted. And if the code was being used in an existing codebase, you’d need to integrate it into the existing code in an acceptable manner.

I’d say that every industry currently has some anxiety regarding AI, and that is natural. People fear the unknown. And who knows what the future will hold.

In summary, so far, in terms of using AI for coding, there is nothing to be afraid of if used with caution.