LLMs like ChatGPT and copilot are like those super saturated junk food like a pizza or burger which feels good in that moment (ready code snippets or answers) but over a period only accumulates weight gain, sugar and diseases (technical debt, brain drain).
We have stopped reading or even looking up official documentation, that has become an extinct skill today. And why would we even if an LLM does it for us and tells us only what we need to be told to create that release or fulfill that urgent deadline.
What happened with AWS outage recently is only a brief foreshadow of what might eventually come to pass if this trend continues. Imagine a world where most programmers are primarily LLM prompters with a very shallow understanding of core programming skills or even operational skills pertaining to an app, framework or library. What will we do if a major outage or technical issue occurs then and no person around knows what’s really going on?
And that’s not even mentioning the replacement of human workers problem which is the most discussed topic these days. Eventually, the senior/mid management will think why do we even need these “prompt engineers”, let an agent do that work. After that, senior management will think why do we need these “prompt managers”, let another agentic AI that controls other agents do it! Eventually, the company will be run entirely by robots and shareholders will enjoy their wealth in peace!
As dystopian as the above scenario sounds, that’s the world we are eventually heading towards with all the progress in AI and the commerce oriented environment it’s evolving in. But it’ll still take decades at least considering the state of prevailing systems in public and private sectors. But until that happens, let us programmers equip ourselves with real old school skills which have stood the test of time - like scavenging documentation, referring to stack overflow and wikipedia for knowledge, etc. and coding with humility and passion, not this LLM crap.
I wrote a couple of applications in C# a few years ago and then ChatGPT couldn’t help me with my questions, I think it has improved a lot since then, but the fact that it can make mistakes and do things incorrectly is a problem. There is a need for specialists who will double-check what exactly he writes, because he can simply forget about security and not use private in his program, which will cause users to lose data. In general, my opinion is that AI is good, but everywhere and always you need control and you can’t do without people.
If you have a random bullshit generator, this may still be very valuable. As the bullshit will make you think and find new angles to your problem. So despite its limitation, I find AI a good sparring partner. But if you swallow the bullshit without thinking yourself, I agree.
This is not about AI (IMO). It is more of avoiding to put all eggs in one basket. By diversifying, you reduce the risk. The bigger centralization, the bigger risk.
Progress hurts. Horses vs Cars and so on. Accept or not. Embrace or not. Change or not.
I think it depends on the programmer. Some will become more effective using AI and some less effective. Those that use it effectively might rise above the others.
Documentation has become less useful, sometimes useless. Good documentation has become a rare skill.
There will continue to be a need for experienced programmers, at least for a while. If beginner programmers do not get experience then that will be a problem that needs to be solved.
I think you are correct that in the future AI will likely be able to generate applications from specifications that non-programmers can write and understand. Grace Hopper would be very excited by the possibilities; a primary intended feature of COBOL was for a language that non-programmers can understand.
Half a century ago people were saying that most programmers will be obsolete but then software was able to become more complex. We do not know what will happen.
Go for it. Do not expect everyone to comply. The ones that make the best decisions will prevail.
Post hoc ergo propter hoc. Unless ive missed something, the AWS outage had nothing to do with AI. It was an automation failure; but do not conflate automation with AI.
That said, “progress hurts”… if its progress. Ask those that tossed themselves onto laserdisks. Or those that said we’d all be using VR for everything by now, or…
Progress often hurts. But not every innovation meets its lofty ambitions. Time will tell.
TBH LLM is very much a layer of automation. It takes a prompt and generates ready code as output, a process which can (and will) be automated though it presently works with human prompters.
In fact, agentic AI and Model Context Protocol (MCP) which is a hot AI topic today is exactly about that - automation of not just coding but other high level processes like code reviews, testing, delivery management, etc. by models conversing with other models and agents using protocols like MCP.
Again, post hoc ergo proptor hoc. I suggest you look it up.
What you have just said equates to the following:
“Robots make a lot of things. In fact, a lot of factories are making cars with robots, This problem was caused by a car from a factory. So it’s a robot’s fault.”
Automation != AI. AI != automation. They CAN be used together. They arent ALWAYS. And nothing in AWS’s problem, again, to my knowledge, has said annnnnything about AI.
That analogy is very illuminating and illustrative. Imagine what happens when you take humans entirely out of the loop and only robots control all processes? In that case, it will truly be the robot’s fault - or the fault of whoever trained it eons ago and is nowhere to be found today.
This is what’s happening in the software industry today. As more and more parts of the development and delivery process get taken over by LLMs, there will be eventually nobody left in the company to take the blame but the LLMs themselves.
Except it’s… not what happened at AWS. Which was the point.
If we actually got to the point where
Then yes, it would be. But we’re not there. And we wont be there for a loooong time. Because at that point, you’ve invented Skynet, and the Terminators are coming for you.
I get that point — it’s definitely a double-edged sword. AI can boost productivity, but if you rely on it too much, you stop thinking deeply about problems. The best programmers still use it as a tool, not a crutch.
LLM is another term that people are using for stuff much broader than what it is defined to be. LLM is an acronym for Large Language Model. They are a type of Natural Language. People are now using LLM as if they include knowledge bases and expert systems and things like that. If I ask AI about the terms knowledge bases and expert systems then I get confusing answers. There should be a better term than LLM but unfortunately everyone prefers to misuse a term than create something. I know we are not supposed to post answers from AI but I hope the following is appropriate. The following are suggestions from Google Gemini. I do not like any of them but I think they get the idea across.
I totally agree! Using AI can make tasks easier, but it’s important not to rely on it too much. It should be a tool for quick tasks or checks, while deep understanding comes from hands-on experience and learning the basics. Good developers will always need strong fundamentals, especially when facing complex issues. AI just helps with the tedious stuff!
