Intro
The current state of AI is scary, not because we are loosing jobs; not because I don’t like it; it’s because is advancing really fast to levels I think most of developers didn’t expect. Practically every week there are new LLM versions of existing models, or completely new ones, and most of them are really capable of doing a lot of stuff. It is really what happened to the .com boom back in the 2000’s when thousands of new domains and pages were being created on the internet. I have to accept that the tech is pretty amazing, I’ve been using it for about 2 years and it’s just getting better and easier to adopt to your workflow, although at the moment costs are elevated if you are not running local models.
My first interaction with AI
I think it started back in 2024 or in 2023 I don’t remember, it started with ChatGPT from OpenAI. I was just using it to ask questions about software or technical stuff just to see if it was really that great for what at the moment I was seeing on the news or social media. In 2024 I was still using Stack Overflow in order to solve coding problems, or to learn how to do certain stuff I was trying to accomplish; but then decided to try ChatGPT exactly for that in order to not burn my time looking on the web and see how that was. And oh boy, it solved every doubt I had, and not even that, it explained in detail and added more related topics in order to understand more about the concepts I was asking about. That day I decided to give it a try and add it to my workflow when encountering issues.
After some time using it just for questions, New OpenAI models were coming out and people (developers on social media) said that it was getting better at coding, so there I was, now asking the model to solve a bug I had on my code, or to write small parts of code that I new I could do but were tedious, and yep as you expected, it worked. Now, not everything was smooth and the code it recommended not always was the best, so I needed to re-prompt in order to get what I wanted. Keep in mind I was not trusting what it delivered, I was still reviewing the output to look for problems, or that it didn’t adapt to my way of coding. But nonetheless it was really helping me with more code output no my projects because I didn’t want to loose much time on boilerplate code.
Microsoft Copilot
Recalling, I was using just the web version of ChatGPT, at the moment I was copying code from VSCode and pasting it in there. It was just how I started and didn’t look into other stuff at the moment. It was then when while looking on the web I noticed Microsoft had Copilot and it integrated with VSCode in order to access and update your code directly on the text editor. This had a cost, it was 10 bucks at that moment, so It was really not that expensive and you had a lot of models with limited access but you had a bunch to choose from, so I just bought the subscription and started using it.
It was a game changer, having the agent directly on the text editor and being able to link the files or code snippets without leaving the editor really increased my productivity, and it became my main method of using AI, because in reality as a developer I only use it for coding, I don’t use AI on my daily basis for regular activities, sometimes I ask questions still to ChatGPT or Claude but thats really it, I use it the most for code.
Claude and how I currently work with AI
Currently Im working with Claude on my linux machine. I use it on the terminal with it’s TUI or in VSCode with the available chat extension. I changed to Claude because recently Copilot changed it’s way of charging models and now is token based, which is no longer appealable for me. Anthropic has a 20 dollar plan which is still sufficient for my code tasks, and Opus or Sonnet are very capable for coding so I’ve been using them a lot.
Something important to say is that although I use AI, I recently decided to first look on the web for stuff, and then if I don’t successfully find what I need then I try to use the agent. Why? well I was starting to see a problem with AI, and it’s that I started to code less, and I love coding so I decided to handle it in a different way in order to not loose the passion when building stuff. So far its has been a good balance and I’ve been enjoying it more, so the current workflow would be around 40% manual coding, and 60% AI, of course the part of AI is supervised, I don’t accept code I don’t like, and ask to change it whenever I feel it will not adapt to my way of coding.
I’ve been trying to start a home lab. I’m a big fan of owning your own services and data, so while I do that I will also try to run a local LLM in order to reduce my bills for AI. But of course that is still a work in progress.