Is software engineering still worth it because of AI?

I would rather see an AI say, "I don't know." rather than a confident wrong answer! We get that enough from the management types!

Always love your top notch analysis Dee. Amazing

In agent based systems you can assign different roles and your AI is checking (and testing) the code. The output is a lot better than just plain use of LLM.

Well, we’re all discussing “what’s happening today,” but these technologies are steadily improving. it means our fears may become reality in the future. currently, these technologies struggle with reasoning and logic in conversations, but a lot of work is happening behind the scenes. even in the case of "Devin," the company is diligently working to fix issues and release an improved version.

Thank you for the analysis Dee. I agree with every part.

On our interviews at my current job, we give people 4 coding questions that tend to use a specified input and output but leave the how completely open. They are designed to take an hour or two, but we let candidates who pass the first interview take them home, solve them in their own time with internet enabled. The second and final interview consists partly out of a code revoew to see how they take criticism

I work as a software developer for 20 years. Yes, solving complex problem is main part of the job. AI can help by providing most common solutions and starting point what to google, but that's not always applicable, cause code is specific for applications that are far from something typical. Another problem is that AI is great for creating simple applications without context: here is piece of log file, I want to extract some information such as timestamp, values, confidence factors into CSV - ChatGPT can do it and even write some valid regexps. Then I want to plot these and/or find outliers - again, no problem, another 20 lines long Python script that works. But it really fails in tasks that are dealing with math, kinematics, geometry (my current area) and in general problems that have wide context. But knowing limitations means that AI can be still very useful for some auxiliary function, simple python code snippets (i'm not familiar with using pandas, sympy etc.), for rubber ducking or giving overview about some topics, typical solutions, algorithms or libraries. Bad stuff: it takes away simple, yet rewarding tasks that are nice to have to relax for half a hour.

I’ve used AI at work and I’ve realised that the responses that I received from it were usually were usually wrong or not what I was looking for. It work really well when you’re creating something simple from scratch but the more complex the problem is the less likely it is to be helpful. Especially when you have a software with millions of line of code. It seems promising in the future but currently no.

It's no longer worth it. But not because of ai. The market is flooded with developers and there are hardly any jobs. Companies startup, get funding, hire tons of developers, and then lay them all of and outsource for maximum funds. There are hardly any jobs and seniors are taking mid level and junior roles because of this.

I think there is a big problem with how that research paper is being interpreted. If you pay attention, the trends that are discussed do not start in 2023, as you'd expect for something released in late 2022! You can see the shift ongoing in 2021, 2022. You don't even need to know that there was a massive boom in hiring of junior software engineers in that period to understand there is a confounding factor here. I am willing to bet that the main driver behind those trends is in fact the massive increase in the size of software engineering teams we saw during the pandemic, which often meant hiring people who were very junior, often switching careers, with not a lot of training especially in terms of best practices. That said, I agree that AI is best used today as a semi-trusted colleague. I've been using it to research errors I get when writing Rust, and it's been a massive help. I am now able to learn new concepts and understand things in detail much quicker than I would before having to dig through documentation and forums. The main thing is not to copy/paste blindly, but make sure you understood what the code does.

Overall i would say yes its getting sloppy. Im newly graduated and wow ChatGPT was a mess as a teachers aid.

1:19 made me chuckle

Say it gets to point where you give it a 200 page specification to build a non-trivial application and it spits out a full application. Then 6 months later it needs a change. You update the specification and feed it to the AI and it spits out a new version but since the AI has also probably been updated there is almost no chance the new application can just be dropped in as a replacement.

I could go on for days listing all my issues with AI but the biggest problem by far is that it will never just respond with, "I don't know how to do that" it will just keep spitting out code telling you that it is what you need, and in many cases you can see the glaring mistakes in the code as its writing it.

I used to look at Stack Overflow solutions and then implemented this into my code. However I analyzed the code and improved it and make it fit into my existing code. I am doing the same thing with AI code, I look at it, then manually modify it and make it even better. But I can only make it better because I have a history of creating code out of nothing. My brain is trained to create code since the 1980's :-)

Even if... you would have more time for videos 😍 (although you are dropping very frequently :))

AI replacing large part of devs is just a matter of when, not if. Devs are too expensive compared to a high quality AI system which can be used as a service.

Just a while ago I asked ChatGPT to write an encryption function for me. About 10 prompts later I knew that ChatGPT can't code, it can just mash together what it has already seen somewhere. And that was one function. Until "AI" will be able to read und understand thousands of lines of code (that includes how the UI will look and respond) and develop it further according to the users wishes, I will be retired.

I feel like im tuning into a lot channels where I'm more experienced than the author, sorry to say but I feel like its the case , does anyone know where can I get some senior content 😅

AI will demonetize software production, lowering the bar for enthusiast coders and boosting open source software production.