Taking the unconventional path

A staff engineer quit their job at a startup after 2.5 years, citing dissatisfaction with AI's impact on software development. The engineer, who turns 40 next year, argues that large language models replace 90% of the enjoyable coding work, reduce job satisfaction, and raise ethical concerns about stolen training data and high energy consumption.

Two weeks ago I quit my job. I had been working as a staff engineer at this startup for the last 2.5 years. The people, pay, and workload were all good but it was time to move on. I have worked with most of the team before this job and I am sure we will probably end up working together again in the future. So why quit? To be honest, I can't 100% rule out a midlife crisis. I turn 40 next year and with everything that has been going in tech in the last year I feel like it is time for a change. When AI first came out I was quite interested. I have always been a sucker for new shiny tech and AI seemed like a childhood dream come true. I could finally build my own Jarvis as seen in all the Iron Man films. Who wouldn't want a sarcastic AI assistant that is constantly mocking you when you ask for help. Of course making funny assistants was never the goal for the big tech companies. The goal and the reason why their companies are so ridiculously valued, is to replace all workers. To be honest, I wouldn't mind if the goal was to replace workers but still get paid, but that is never going to happen UBI anyone? . As a technology I don't have a problem with AI. Being able to ask questions and get answers for things that you wouldn't be able to search for is genuinely useful. I have even used it for generating small scripts for myself that have saved me time. Using LLMs for professional software development though sucks. I got into programming because I like using my brain, solving problems and turning them into code. I often found while writing the code I would come up with even better solutions or think of edge cases I hadn't considered before. LLMs essentially replaced 90% of the job that I enjoyed with the other 10%, reviewing other peoples code. I have nothing against reviewing human code and providing feedback, it can be great way to mentor junior developers remember those? . Reviewing an LLM's code which is ultimately going to have your name associated with it is really tiring. There is just so much to review and it is difficult to detect the bugs that you would have picked up if you had written the code yourself. The other parts of the job were still there but were greatly reduced. Yes the initial problem solving was still there but all the insights and additional thoughts while writing the code were gone. The big thing that disappeared though was the satisfaction you get from solving the problem and writing the code yourself. When your job is reduced to prompt, review, prompt, review, in a constant cycle there is not much job satisfaction. When you are working for someone else it is hard to justify doing things the manual way. Generating code, doing a thorough review and testing all the edge cases, is still quicker than writing the code manually. If I could type a lot quicker than 65 WPM then it might be different. In most cases though the feature ends up being good enough, which is just depressing for someone who takes pride in their work. Even if coding with LLMs didn't suck, the ethics of AI still troubles me as someone with a strong moral compass. LLMs are trained on stolen data and includes millions of books and websites where no consent was given to include their information in the training data. On top of that, the amount of electricity and water need to run these models is insane. If I could have a local model, that wasn't trained on stolen data and used minimal electricity then I wouldn't have a problem with it. I have tried some of the open weight models and in most cases they are too slow for coding tasks on my otherwise powerful Mac Mini. It does make you think more about the amount of power needed to run these things, when a computer that can do video editing in 4K struggles, to run a model that is a small fraction of the frontier models. I have really missed writing, making videos, learning new things and teaching it to others. So despite being demonitised https://www.alexhyett.com/youtube-just-demonitized-my-channel/ on YouTube, I am going to get back into making videos again and writing blog posts like this one . As someone who took physics at university, because computer science looked "too easy", I really like learning new things. I still like building things with code but I want to work on my own projects where it is not expected of me to use AI to solve everything. I am not sure I could get a job in software right now that didn't require me to use AI on a daily basis. Luckily I have built up some savings while working so I am not going to be homeless any time soon. Also my wife, who has been a stay at home for the last 13 years is going back to work through choice I might add . Luckily AI hasn't reached the industries she is going to be working in and likely won't for a while. This also means me picking up all the parental duties that my wife used to do while I was working including school runs, doctors appointments and sports days. I still need to earn money however, but with my wife working and us living fairly frugally I only need to earn around £22k a year to not dip into my savings. The plan is to focus on making courses for those that still want to learn about software development along with working on my own software products. My first course will be called "From Localhost to Production" and covers everything you need to get your project live and ready for people to use. I figured with so many developers out of work at the moment and the job market looking bleak for juniors, that more people will be looking to launch their own products. If you are interested you can sign up to get notified on my courses page https://www.alexhyett.com/courses/ for when it will be released.