I can only say I find it useful for coding, and its way faster to ask it questions instead of searching documentation. It can read the code base, and explain it to me instead of me trying to understand the cryptic 2 and 3 letter variable names the last a hole used, in their 57 state state machine, all states just numbered, no names (why a state machine in python? Some people…) Then when I want to change something in the code that is substantial, I can ask it to write a draft that I then refine, saving keystrokes on boiler plate. It can suggest data structures and algorithms I’ve not yet used or heard of, and then I can learn about them, making me smarter as well.
I did this all on my own before with a lot of grep and find commands, reading python/perl/c++/tcl/git/cvs documentation. Then tracking down someone to explain the piece Im not understanding. It turns a few weeks worth of hard effort into a relaxed few days of feeling more productive.
Even just linting, I can ask it, why is this function not giving me the expected outcome (in terms that simple), and it finds the 1 off error faster than me, like in 5 sec in 500 lines of code.
Its like having someone with perfect recall that has read all of the code base, and all comp science info on the web, sitting next to me. Its not a great coder, but I can get the information i need to be the good coder I am faster than google and grep. Not using it now is like insisting that O’Reilly books (which i have read for fun in the past) are better than searching the online docs or google.
AI can‘t read or write. It processes and computes data in entirely different ways than we do: Based on probability. It doesn‘t understand context at all. We‘ll see how well vibe coding holds up in due time when more of our infrastructure is vibe coded and fewer solutions are actually understood.
I wouldn’t call what im doing vibe coding, would you?
Ill also say, who cares if it reads or writes? The point is it makes me more effective at my job, the fact it is presented as a conversation as opposed to another format is an implementation detail not a critical feature.
A loop and a data structure to hold the result of parsing. The data structure will have a state you can interrogate. No need to build a state machine to explicitly name each way the parsing can develop.
I also find it very useful to bounce ideas, like an interactive rubber duck. Even when it’s wrong it can help me think out loud or elaborate ideas.
It’s also very useful to help me set up tests, raise and delete environments, write documentation, etc. all things that I can do on my own, it can help make implementation faster.
AI can be useful when you use it like a tool. I know it can and will make mistakes, but like all tools it can help make things faster when used correctly.
In my field its common for my colleagues to not have enough experience outside of their role to provide the feedback im looking for. They help in conversation for sure, but the wide knowledge base that is available through AI is my biggest win.
That and I cant over utilize it (well at least as a rubber duck). I can bug my colleagues too much and get in their way if im too talkative.
When does it “actually make … sense”?
I can only say I find it useful for coding, and its way faster to ask it questions instead of searching documentation. It can read the code base, and explain it to me instead of me trying to understand the cryptic 2 and 3 letter variable names the last a hole used, in their 57 state state machine, all states just numbered, no names (why a state machine in python? Some people…) Then when I want to change something in the code that is substantial, I can ask it to write a draft that I then refine, saving keystrokes on boiler plate. It can suggest data structures and algorithms I’ve not yet used or heard of, and then I can learn about them, making me smarter as well.
I did this all on my own before with a lot of grep and find commands, reading python/perl/c++/tcl/git/cvs documentation. Then tracking down someone to explain the piece Im not understanding. It turns a few weeks worth of hard effort into a relaxed few days of feeling more productive.
Even just linting, I can ask it, why is this function not giving me the expected outcome (in terms that simple), and it finds the 1 off error faster than me, like in 5 sec in 500 lines of code.
Its like having someone with perfect recall that has read all of the code base, and all comp science info on the web, sitting next to me. Its not a great coder, but I can get the information i need to be the good coder I am faster than google and grep. Not using it now is like insisting that O’Reilly books (which i have read for fun in the past) are better than searching the online docs or google.
AI can‘t read or write. It processes and computes data in entirely different ways than we do: Based on probability. It doesn‘t understand context at all. We‘ll see how well vibe coding holds up in due time when more of our infrastructure is vibe coded and fewer solutions are actually understood.
I wouldn’t call what im doing vibe coding, would you?
Ill also say, who cares if it reads or writes? The point is it makes me more effective at my job, the fact it is presented as a conversation as opposed to another format is an implementation detail not a critical feature.
deleted by creator
as someone who would also reach for a state machine when doing simple parsing… what’s the better alternative in python?
A loop and a data structure to hold the result of parsing. The data structure will have a state you can interrogate. No need to build a state machine to explicitly name each way the parsing can develop.
that’s what I’d do for a state machine implementation too
I also find it very useful to bounce ideas, like an interactive rubber duck. Even when it’s wrong it can help me think out loud or elaborate ideas. It’s also very useful to help me set up tests, raise and delete environments, write documentation, etc. all things that I can do on my own, it can help make implementation faster. AI can be useful when you use it like a tool. I know it can and will make mistakes, but like all tools it can help make things faster when used correctly.
i have colleagues for that
In my field its common for my colleagues to not have enough experience outside of their role to provide the feedback im looking for. They help in conversation for sure, but the wide knowledge base that is available through AI is my biggest win.
That and I cant over utilize it (well at least as a rubber duck). I can bug my colleagues too much and get in their way if im too talkative.
I have an actual rubber duck for that. I hate it when my colleagues interrupt me just because they forgot how lambdas work
that also helps. truth be told we’re all too busy to do actual rubberducking unless we work on something together so proxies are a good alternative.