Disclaimer: this post is written without any AI assistance (except for the generated image cover) to share the real events from my daily experience. It is happening right now.
Hey everyone.
I'm a Senior Software Engineer working in a medium-sized startup, and a recent introduction of AI is slowly destroying my team under the false claim of improved efficiency.
We've always had a pretty small tech team with under 20 engineers with a good mix of senior, middle and junior developers. We've always shared some cool technologies, pair-programmed when possible, brainstormed some new ideas and were pretty connected overall.
Unfortunately it all changed with upper management doubling down on integrating coding agents and making them practically mandatory. I started to realize a huge difference after a couple of months of this forced integration:
People almost completely stopped talking to each other. There is no more "let's work on this problem together" or "let's refactor this code" - it is all done in a silent way, something you would expect more from external consultants rather than a seasoned group of people who worked together for quite a few years. I almost don't want to come to the office anymore because I can't stand these constant chats about AI interaction rather than discussing the actual technical solution. It is borderline unbearable to me.
Our developers started to rely on AI coding agent to the extent of running shell commands like "npm run x". A few days ago I asked my colleague on how to run a particular set of newly-introduced tests to replicate what is done in continuous integration pipeline, and the response shocked me: "Just ask an agent to run a test, I don't know". Since when have we built a reliance to such a degree like running something locally as common as tests? Why do I have now to ask a cloud service to analyze my prompt on some external servers and generate a local command? Who said it is a new normal?
The more AI code is getting merged, the more I don't understand fully what's going on. We have a fairly complex financial logic running around, and every variable name and function impacts the understanding and readability of all internal algorithms. A few days ago I had to debug some core issue where calculations didn't make sense, and it took me ages to understand what was going on. It all looked logical, but in the same time - not at all. I was reading a code in a style of DeepDream - it looks so familiar, but it is not. I realized that it bypassed all manual code reviews because in small batches it makes sense, however when a bunch of generated merge requests are getting combined together - it becomes a slop!
Management is publishing big numbers like how many lines of code was merged, plus how fast the tasks are being done, yet I feel a deception. That's true, we started to work faster, but people forget how to write a clean code and many times just follow the flow without switching their brain on. Engineers are degrading on my eyes.
Do you remember the Cloudflare outage? When suddenly half of the internet switched off? When ChatGPT was stuck because the captcha was blocking access and people were losing their minds since the "Big Brain" was offline? That's what we are playing with right now. It is very easy to get hooked on something like coding agent, and then a simple outage will make you feel like you haven't coded for a while.
In conclusion: I believe AI is amazing. I use it to do things which would require me ages to find API documentation, understand the framework etc. But I completely decline to use it on autopilot because not only it makes our skills weaker, but it also destroys a healthy developer environment and I'm observing its impact daily.
Top comments (0)