In the 20th episode of my Mastodon series I pivoted to a new topic: LLM-assisted coding. After three posts in the new series, it got picked up by The New Stack. Here’s the full list so far, I’ll update it here as the series continues there.
1 When the rubber duck talks back
2 Radical just-in-time learning
3 Why LLM-assisted table transformation is a big deal
4 Using LLM-Assisted Coding to Write a Custom Template Function
5 Elevating the Conversation with LLM Assistants
6 How Large Language Models Assisted a Website Makeover
7 Should LLMs Write Marketing Copy?
8 Test-Driven Development with LLMs: Never Trust, Always Verify
Although I’ve spent decades working with software, and still regularly write code in one form or another, I’m only an average programmer. There’s no doubt that LLM assistants can help me be better. But how much better? In what ways exactly? That’s what I aim to find out. My day job embeds me in a fast-moving open-source project, Steampipe, to which I want to contribute as much as I can. That ambition was, until recently, tempered by my own limitations. Now, augmented by the collective knowledge embodied in LLM assistants, I’m raising my sights. Larry Wall famously said: “Easy things should be easy, and hard things should be possible.” It’s starting to look like even hard things can be easy. But the devil’s in the details, so in this series I’ll focus on concrete examples of how LLM assistants do (and don’t) help me become a better developer.
Despite my wariness of the hype, I have to admin I had a successful ChatGPT experience having it generate a complex Google Sheets formula after my attempts and searching failed.
It’s not all that different from sifting through code snippets from Stackoverflow et al, where you try one solution, it does not work, you look for another, it mostly works, and then you tweak til it does the job.
As an educational colleague wrote recently (more on using LLM to update content), it’s not a replacement for doing the coding, but can be an accelerant in the dev process
For a lot of my routine scripting it is very much becoming a replacement for doing the coding, which is great, because it enables me to focus on non-routine coding, where the assistance can be helpful in some very different ways.