[t43562]

When google turned up (or before) we started to get more productive by being able to look up information and other people's code examples and so on. I couldn't have done my job properly in my whole career without search to help me.

Anyone could look up an internet reference and copy some lines of code without being special. Or could download a python library that does what was needed and write almost no code.

GPT is a kind of improvement of search which again makes the common knowledge base more usable. It obviously lets people do less work just as installing a module saves a mountain of work.

Yet jobs have not seemed to get more scarce since search engines appeared. We have built out a huge internet infrastructure and that might have skewed things because it needed such a lot of effort and people.

We've simply (IMO) done more and more amazing things because there has been an appetite for them. Perhaps there is a limit to software and there might be a time when there are enough people doing it that the salaries look less amazing but it need not fall off a cliff suddenly.

We tend to do the same things many times over anyhow,

[caeril]

I've had the opposite experience. Not just google, but StackOverflow. Did these tools make me more "productive" in the immediate sense of closing feature request tickets? Sure, absolutely.

But these are all crutches for the mind. I don't know about you, but my capability and skill as a developer has only ever been improved by meeting dead-ends, and doing the cognitively demanding work of actually understanding the system I'm working on at a deep level, and deriving solutions myself.

Google, StackOverflow and the like have definitely improved raw feature delivery rates, but they have indisputably made me dumber and less skilled.

As for LLMs, I'm on the fence. I am using GPT-4 in some code generation utilities I've written to speed up mundane tasks, and I do not yet believe it impinges on my ability to learn, as these are just mundane tasks. But there's a good chance this will change if I ever get access to the 32K token model, or GPT-5, GPT-6, etc are released and will be much more capable.

[t43562]

I agree 100% that being given the answer is useless because then you haven't been forced to understand the problem.

I'm thinking of 2 special cases: 1) you can read manuals and get modules online. I lived in Africa in the days of 2400 modems and information was very hard to get - always in books which I could not buy. Now anyone in the world can get information so there's a way in which that allowed the software industry to expand but fortunately there was huge demand.

2) Bugs which other people have solved before - a particular error message from Anypoint Studio today was because I had run brew and I updated maven without realising it - past the version that Anypoint supports. Searching saved hours of frustration. When this works I think it's a true saving.