The announced death of the software developer

After the hypes of the no-code and low-code tools, now generative AI is the hot new kid on the block that will make everyone a software developer. Marc Andreessen said once that software is eating the world, is it now become the time where the software is eating its creators?

To really understand how the new AI tools have an impact on the work of a software developer, I build a website and I’m now moving into the bigger work of rebuilding a Saas-application from scratch.

I’m pleasantly surprised how AI can augment my skills as a developer and after some trail and error, I had within a day, the website structured and up and running without writing one line of code. The end-result was not quite what I wanted, but it was sufficient for a first version. It is like learning a new programming language to get the AI to do what you intend. All the more important steps of software development are still valid when working with AI. You still need to capture requirements, decide on your tech stack and define a solution. The AI will help you translate it to initial code. It is even helpful and suggests how you might improve on the application it made up.

But then the drawbacks are starting to show up. Getting the last 20% right, really takes an effort and AI is not really helping anymore. Asking AI to tackle certain changes, made everything worse and not for long I had no longer a working application. Bugs introduced by AI, take more time to solve as my coding companion ‘AI’ runs around in circles to fix the bug. I need first to dive into the code to understand the logic, before even can make an attempt to fix the code. A lot of the initial time gain, is lost. On closer inspection, I find the use of deprecated functions, unsafe database queries and some other security issues. Asking to fix/enhance part of the solution, breaks the application in other places. For security reasons, I needed to upgrade a library to a newer version, but AI could not deal with it and executing the upgrade steps manually broke my application. AI was no help at all in fixing the problem, as it always wanted to reverse to the previous version. I found myself rewriting some parts of the application and fix its structure. There the offered code completion was really helpful and improved the speed of writing. Also I found the generation of testing scripts very helpful, as the developer can focus on completeness of the test and not on writing the code.

In this green field environment, artificial intelligence allowed me to build quickly a prototype, but then the human intelligence had to take over to create a robust, reliable system. Do not expect it to be up to speed with the newest trends and technologies.

Generative AI is definitely augmenting and empowering software developers, but is not replacing them fully. It will free up time, so the human in the equation can do the thinking and leave the boring lifting to the machine. We have to watch out that we are not shifting our focus from understanding and learning to completion and become totally depending on AI. Let us no trade long-term understanding for short-term productivity, by always read and understand all AI-suggested solutions. Let this not be an era of poorly build insecure applications, where the real experienced software developer is extinct, and nobody can fix it.