What you're describing sounds to me like absolute hell on earth.
I'm not interested in reaching the finish line with maximum speed and bypassing the hard work of struggling with and solving problems myself.
Partly this is because working this way has real benefits that are difficult to quantify. One example: I've recently dumped an enormous amount of time into investigating performance problems in the tools my team use. I've spent more time making dumb mistakes than actually improving anything. I've also learned a tremendous amount, to the point that I was able to diagnose in seconds the cause of a serialization error in one of the tools we use for testing. Others were convinced that these crashes were expected. I was able to show them that, and why, this was wrong. I've likely saved multiple people on my team days' worth of confusion and struggling, because they were trying to solve the wrong problems. If they'd charged ahead with their intended fix, I suspect the result would have been an outage in a global service that has stringent requirements for availability.
An LLM may have been able to tell me in seconds how to solve the performance problem that started my investigations and dumb mistakes. But I'd have learned basically nothing.
If your goal is to make something specific and code is both the obstacle and the means of reaching that goal, sure, great, I'm glad LLMs work so well for you.
I just want to program. I want to solve problems, understand, and become better at working with programming languages, software, and systems. I haven't seen any evidence that LLMs will help me do this. As far as I can tell, they'd do the opposite. They strike me as a layer of awful, chipper bureaucracy between me and what I actually want to work on. I call this meeting-based programming -- and if that's what software engineering becomes, I'd rather leave the field than adopt that style of workong. And maybe that's a good thing. Maybe LLMs will enable more people to make better stuff faster, and maybe that'll be better for everyonr.
I suspect it won't though. I think it would be a dangerous Faustian bargain, and I'm pretty sure I'd rather die than cede intellectual work -- the thing I love most -- to a machine.
I agree there --- if you want to program, don't use an LLM!
Sometimes I do turn off the LLM on purpose because it is intrinsically enjoyable to program. I like to do things like Project Euler and I would never see the point of having an LLM do it for you, unless you were explicitly reading its code to try to learn something new.
Easier said than done though. Like many programmers, I'm finally facing upper management that expect everybody to start using AI tools. I'm half expecting to lose my job in the near future for refusing.
The future is coming, they keep telling us, (or it's already here) and if I don't actively strive to turn their fantasies into reality, I think they'll have no use for me.
I'm not interested in reaching the finish line with maximum speed and bypassing the hard work of struggling with and solving problems myself.
Partly this is because working this way has real benefits that are difficult to quantify. One example: I've recently dumped an enormous amount of time into investigating performance problems in the tools my team use. I've spent more time making dumb mistakes than actually improving anything. I've also learned a tremendous amount, to the point that I was able to diagnose in seconds the cause of a serialization error in one of the tools we use for testing. Others were convinced that these crashes were expected. I was able to show them that, and why, this was wrong. I've likely saved multiple people on my team days' worth of confusion and struggling, because they were trying to solve the wrong problems. If they'd charged ahead with their intended fix, I suspect the result would have been an outage in a global service that has stringent requirements for availability.
An LLM may have been able to tell me in seconds how to solve the performance problem that started my investigations and dumb mistakes. But I'd have learned basically nothing.
If your goal is to make something specific and code is both the obstacle and the means of reaching that goal, sure, great, I'm glad LLMs work so well for you.
I just want to program. I want to solve problems, understand, and become better at working with programming languages, software, and systems. I haven't seen any evidence that LLMs will help me do this. As far as I can tell, they'd do the opposite. They strike me as a layer of awful, chipper bureaucracy between me and what I actually want to work on. I call this meeting-based programming -- and if that's what software engineering becomes, I'd rather leave the field than adopt that style of workong. And maybe that's a good thing. Maybe LLMs will enable more people to make better stuff faster, and maybe that'll be better for everyonr.
I suspect it won't though. I think it would be a dangerous Faustian bargain, and I'm pretty sure I'd rather die than cede intellectual work -- the thing I love most -- to a machine.