Where do you stand on coding agents?
A question has been roaming my mind lately.
On what is the better use of one's time and effort in approaching programming and development in general (career wise not nerdy wise). It goes like this: Is it the better investment to focus more on abstract problem solving and understanding architectural, engineering and systematic thinking by the more utilizing natural language and only using coding agents once fully understanding the problem and the flow YOU CHOOSE and just let the agent as more of a translator. Because I know without I can do it it will just take time, but by utilizing AI I can have more throughput in the thinking described earlier and maybe in a longer term could be more beneficial since coding agents will only get better from now. I also believe you should do the debugging and understand why a thing went wrong, also understanding the code generated by AI.
The only regret feel is that by coding manually in a dull way you learn in a much harder way that could stick to your head better, but is it the best investment in this era? is there a better approach? I wanted to get this out of my mind and have more of a discussion about it, because I am really interested in other's point of view.
You should always be investing on abstract problem solving and understanding architectural, engineering and systematic thinking.
Even with agents, you're still better off using structure to describe your system than pure natural language.
I have been coding since I was 11 which was 45 years ago, which I think is an important note.
In theory I can write all of the code that Claude is spitting out for me. In practice I have no desire to write the thousands of lines of axaml markup for a C# Avalonia application (maybe there's a GUI editor for it, I don't know).
I can now create solutions to problems I wrote off. And I can do it on my phone while I watch TV.
But I do have concerns about an imaginary 11 year old me today who is interested in this stuff. What is the pipeline to coding in assembler for him? (although I had the same concern before LLMs as I started out on 8bit home computers where dropping into assembler from BASIC was as easy as typing a [ and then LDA 1 etc.).
Who will write the library code that the LLMs join together?
There still needs to be innovation to drive change. People will still need to understand statistics to know if their results are valid.
So where do I stand? A smaller circle of humans will provide new things and prompt engineers will glue them together to make interesting products. We're on this ride, let's see where it goes.