It has been 5 months since my last post about AI use in software development and whether I wanted to continue programming. Like all aspects of AI recently, the industry, tools and landscape seem to evolve rapidly, so a new check on things is in order.
I've been thinking about this X post from v0xium (archive link) off and on for the past few days. I can definitely relate to it in some aspects. It is incredibly frustrating when the leadership tells you to "just ship things", often and quickly, without any consideration for quality of the shipped code. The feelings of this from my fellow software developers has been quite mixed. Some people absolutely embrace AI and everything it provides, shipping and publishing code to production daily, with no or low regard for quality or maintainability. And some people, myself included, have grown quite tired of the long PR description slop and the overly complex code and inlined commenting that Claude vomits out.
Let's point some fingers
I think at the core of this issue is that the management is mostly to blame. They see and hear from fellow management executives and AI sales people how LLMs can improve the velocity of their development teams, so they implement these mandates to use AI by all means necessary. They only see the lines of codes produced, and are unaware of the nuances of the quality, readability and maintainability of the output that is generated.
What will happen next?
This is completely a prediction on my part, but I feel like we are on top of an AI-assisted software development bell curve at the moment. We are demanded high velocity from the management and encouraged to use LLMs on absolutely every daily issue. But I think the curve is slowly declining, due to:
- Frustration from developers of the quality of code and associated textual information
- Rising inference costs, which is bound to happen soon(ish)
- Code maintainability and process knowledge are gonna take a huge hit in the long run, but it will only be noticed when it's too late
There is some rebellion and discomfort among developers regarding the current, vast mandate to use AI tools in software development, but I think that more is needed, and will happen. Some people will speak out, management will notice the declined quality of the output and the consensus among executives will be to limit AI usage (especially if inference costs grow 5 to 10 times).
A sort of equilibrium will be reached, where AI will still exist and be a viable development tool, but its overblown usage is discouraged by companies (or perhaps not for all companies, the so-called "true AI believers").
I do hope we reach this equilibrium sooner rather than later. I'm getting quite sick of Claudisms and overuse of em dashes.
Do I want to keep doing this?
In my last post I quite provocatively titled it whether I want to be a programmer anymore or not.
If this current trend of highly encouraged AI usage continues, then no, I want to do something else, other than run AI agents summarizing pull requests and writing code for me, day in and day out.
But I truly believe that this current situation is not sustainable and will gradually change back to old ways, with AI still of course being used as a tool among many others.
So hunker down, learn to not get triggered by Claude's prose and keep the quality of your own code high despite the pressures above. It's gonna be a rocky few months, but things will get better, I'm sure.