Will we even read code once AI just generates platform native executable binaries using a blackbox AI optimized compiler we have no access to?
Seems we will be writing programs using some kind of logical formal methods[1] and generating the code without ever looking at it and testing the system properties. I imagine AI will make its own intermediate programming language that benefits it the most instead of human readability
I find that LLMs can write poorly structured code if not given a good foundation to do so. The existing code in a repo is what the LLM will see and generally output similar code to.
Given that, I'd recommend the mainstays on good general software design. One of my favorites is "A Philosophy of Software Design" by John Ousterhout. Good general design practices.
Other than that, I think a lot of books generally stay relevant. If it's topic specific, it'll give you a good background on understanding of that topic so you can make sure you or an agent are making sound decisions. Architecture of course is more valuable than ever.
Respectfully, I don't understand this way of thinking. You learn and get better at programming by programming. Now we are using LLMs to write the code to the point where we need to assign ourselves homework to keep our skills?
Why not just use the LLM in a way that doesn't prevent us from absorbing the knowledge?
yes, you get better _at programming_, specifically your own programs, and in a general sense. but is there a way to hone the reading of code that is not yours?
> Now we are using LLMs to write the code to the point where we need to assign ourselves homework to keep our skills?
yes!
> Why not just use the LLM in a way that doesn't prevent us from absorbing the knowledge?
the economic viability of this is unproven in our new future.
Will we even read code once AI just generates platform native executable binaries using a blackbox AI optimized compiler we have no access to?
Seems we will be writing programs using some kind of logical formal methods[1] and generating the code without ever looking at it and testing the system properties. I imagine AI will make its own intermediate programming language that benefits it the most instead of human readability
[1] https://forge-fm.github.io/book/2026/
I find that LLMs can write poorly structured code if not given a good foundation to do so. The existing code in a repo is what the LLM will see and generally output similar code to.
Given that, I'd recommend the mainstays on good general software design. One of my favorites is "A Philosophy of Software Design" by John Ousterhout. Good general design practices.
Other than that, I think a lot of books generally stay relevant. If it's topic specific, it'll give you a good background on understanding of that topic so you can make sure you or an agent are making sound decisions. Architecture of course is more valuable than ever.
Design Patterns by Gamma et al.
Recently helped me solve a non-obvious code-stink design issue.
Respectfully, I don't understand this way of thinking. You learn and get better at programming by programming. Now we are using LLMs to write the code to the point where we need to assign ourselves homework to keep our skills?
Why not just use the LLM in a way that doesn't prevent us from absorbing the knowledge?
yes, you get better _at programming_, specifically your own programs, and in a general sense. but is there a way to hone the reading of code that is not yours?
> Now we are using LLMs to write the code to the point where we need to assign ourselves homework to keep our skills?
yes!
> Why not just use the LLM in a way that doesn't prevent us from absorbing the knowledge?
the economic viability of this is unproven in our new future.
obligatory SICP