I thnk of it like this. I am a computational scientist. I have a new idea, I have claude code churn on this idea for some days, making plots of data, doing random things with it, plotting it, modeling it, etc. This leads to some thoughts about the data as to whether it is worth pursuing or not. This reduces that initial idea exploration from weeks to days. Then when I come upon an idea about the data, I start over. This time I build the analysis myself, the analysis has less overall to do because of the claude exploration. Also, because I have had claude interacting with the data, I can ask it some questions like, which column out of 1000+ columns and tables is the one with X variable? Or I can say to claude, go download the PDFs of a bunch of papers I want to read on the topic, and it might even suggest others. I do think that if people think they will orchestrate AI to do complex scientific endeavors for them, it is not going to work unless it is really a specific thing to do. But the truth is, this is the point of the last iteration of new software, instead of writing software, you write neural networks to solve the problem. And so, if you want to detect some event in some data, the old way is to write a custom algorithm, the new way is to pull a trained neural network off the shelf to find it for you (or train one yourself), and the new new way is to convince an AI agent to do it for you. And so I think that ultimately, there are likely lots of tasks AI agents are going to be happy to do for you, but there are a great deal many, at least from a science point of view, that will be sort of meaningless if the AI does it, because the act of doing it is what builds the scientific meaning. It's a mixed bag. I am both very bullish on AI in the future and rather pessimistic about AI today.

> People were talking about making programming as basic as mathematics and English, but now vibing seems to be the norm.

I think there were similar upsets in mathematics when the calculator and graphing calculator became the norm. For English, I imagine everything from typewriters to automatic grammar tools have felt revolutionary compared to what was important in the past.

> vibing seems to be the norm

I know enough coding that "vibe coding" wasn't something I ever tried/spent time doing.

However, a friend recently stayed in my guest room and used the TV in my office as a monitor. He had time to kill between jobs/housing and using only free AI, he learned how to use Godot enough to get a grip on how the various elements worked and built a working game that would have passed as high-end shareware in the late 1990s.

I was shocked. The guy had nothing but an idea and no prior experience AT ALL in coding or graphics, he just had played a ton of games and had a detailed idea of what he wanted. Copilot ran him in circles now and then, but like any troubleshooting, he found a way to make it work (often copy/pasting between AIs to get one AI's bad code fixed by another).

I could never trust a codebase without reviewing it, but he didn't care. He just made it work.

And what was created in a week by a total newbie would have taken me a month or more if I'd tried to do it by hand. My decades of experience & tricks of the trade do not hold a candle to what AI can accomplish.

Time is the most valuable resource we must manage as humans. So what choice am I making if I keep on the slow/hand-made path?

> Hard to release new languages and frameworks. Now seeding would be needed. Teaching the LLMs about new language and frameworks would require upfront effort from the creators.

This is a very interesting point. I suspect AI will indirectly hinder progress more as time goes on. Doing something new will not be worth this additional cost.