I am a non deep tech person. But I used to learn programming at university and I wrote code as part of my job in my early years. So I know how to code, how to set up a team based software dev project etc. With the progress of my career I shifted more and more in management roles and of course you don’t do anything code related there anymore. So it must have been somewhen around 2016 when I wrote my last line of code – until 2024.
First steps in 2023…
But playing around with all sort of AI technology let me again dive into the field of software development and I dig into writing code again. Of course with the help of AI, which actually means that AI wrote the code and I checked for functionality and output. My first steps with that approach was late 2023 with chatGPT and the output was not quite satisfying. Basically the code produced by the AI was very ugly and just worked after manual bug-fixing. I was not very impressed about AI in software dev back then.
But as the model improved it really also progressed the application in software generation. Comparing the quality of the model from end 2023 to mid 2024 is a huge step forward. I used now again chatGPT but as well Claude. The generated code was not the best in class, but working and additional conversations with the AI did actual work quite smooth to further evolve the code and create bigger projects. Claude Sonnet 3.5 especially is really the leading model here. I was able to create with it simple tools for my consulting business like a QR code generator, a funnel generator and other small single page concepts.
Going a step further: AI first IDE
For this small tiny projects AI generated code works pretty well. I also started to use cursor – an AI first integrated development environment based on VSCode. This make the entire process of querying the AI and put the results back into the code base super smooth. This is at the moment the approach that works the best. There are additional tools, where you can create prompts based on design inputs, which you can then put into Cursor or an AI interface to generate code.
The downside
Beside being really surprise from the progress of AI based software development, there are also major downsides I also experience myself. First of all, you don’t know the code, as it is not yours but from the AI generated. To change minor things, which would normally a 1 minute tasks take longer or you just ask again the AI. This can be really problematic when second you run into the typical problem of a bug, that the AI is not able to fix. This is a condition that you will sooner or later experience and the AI is not helpful any more. Means you have to dig into a code, that you hardly know, and fix a bug manually. Software dev probably know this situation when there is the need to fix a bug in someone else’s code and there is no or very little focus on code quality or style. So this can be very painful and time consuming.
So it is in my opinion super important that you actually have coding skills to overcome to above described problem situation and work properly with AI generated code.
Leave a Reply