imo, Agentic lower the bar for people to actually create software they want, however it also introduce or scale problem for devs, since is inevitable for them to use it
Agent coding hype is marketing to push FOMO on devs in order to trick them into turning off their brain in exchange for money. They get hooked at work where they are likely more than happy to let a machine barf out boring code they likely didn't want to write in the first place. Then they bring that practice home. It's digital crack for devs.
They were so burned out that they decided to use an LLM to write this blog post.
It should be embarrassing to people that they can’t even write a whole argument out by hand. It’s not that hard. All you have to do is believe in it.
This blog post is just selling the idea of engineering as craft. It exists to make us feel good about having skills and to get us to regurgitate our priors in the comments. We should at least demand that the sales pitch be manual.
Doesn't read like LLM writing to me and my detector usually fires very quickly. Don't see any obvious signs - want to share what made you think of LLM writing?
Easiest way to keep yourself from overload is get yourself a local model. At 128GB in Mac Studio or AMD395+ and it's large enough to get the sweet context window, but slow enough that you can either process what it's doing or focus on something requiring more careful work.
It's still hard to stop yourself from constant bikesheding, reorganizing, etc, but yeah, mentally it's a different bag of tricks you need to learn.
I can relate to that. If I'm not careful catching up my internal model and understand of the problem space and the specification of my project while the agent keeps spitting out code I feel very uncomfortable. Perhaps I have very high standards for my code quality for various reasons? But its extremely frustrating and tiring to me to check all the implicit assumptions the agent so boldly makes at high speed.
Usually, I start with a solid idea that needs hands on experience to fully developin, so I need to stay in the loop. Than I have to force myself to carefully review the code and point out architectural issues (usually there sneak in a lot of them). If I don't force myself, I end up in a limbo where the agent produces code faster than my mental model following.
Even if I have a well developed domain knowledge and a good idea what I want, I then definitely feel some kind of, what I guess compares to, slot machine effect. "Maybe now I get what I want. Argh, not quite. Okay, feels nearly there. Perhaps only this prompt. Ah, there is something missing...".
I'm really curious how you handle and approach this.
9 comments
[ 2.8 ms ] story [ 27.1 ms ] threadIt should be embarrassing to people that they can’t even write a whole argument out by hand. It’s not that hard. All you have to do is believe in it.
This blog post is just selling the idea of engineering as craft. It exists to make us feel good about having skills and to get us to regurgitate our priors in the comments. We should at least demand that the sales pitch be manual.
It's still hard to stop yourself from constant bikesheding, reorganizing, etc, but yeah, mentally it's a different bag of tricks you need to learn.
Usually, I start with a solid idea that needs hands on experience to fully developin, so I need to stay in the loop. Than I have to force myself to carefully review the code and point out architectural issues (usually there sneak in a lot of them). If I don't force myself, I end up in a limbo where the agent produces code faster than my mental model following.
Even if I have a well developed domain knowledge and a good idea what I want, I then definitely feel some kind of, what I guess compares to, slot machine effect. "Maybe now I get what I want. Argh, not quite. Okay, feels nearly there. Perhaps only this prompt. Ah, there is something missing...".
I'm really curious how you handle and approach this.
It got me back into enjoying things more