[coldtea]

>Actually writing code was never the difficult part for the majority of software created. It required skill yes, but the really hard part was figuring out what to implement in the first place.

Nah, it was mostly writing the code. One can have a very good idea of "what to implement", even on a feature by feature and architecture basis, but still need many man-weeks or months or even years with a large team to make it into code.

>Which features should the software have, how should they function and interact, which tradeoffs to make given the limitations. Stuff like that.

Stuff like that a person can sit down and plan. Which is what the techical lead or software architect role does.

They still needed to work themselves plus anything from a couple to dozens of coders to make it into actual runnable code.

If the industry was just senior tech lead roles, yeah, it would be affected less. But 95% of it is regular coding roles, even mostly boring coding roles.

[akerl_]

> but still need many man-weeks or months or even years with a large team to make it into code.

That's an argument that writing code was the time-consuming part, which is different from it being the difficult part.

[coldtea]

Well, that itself is a dillema about "which is the most difficult part", whereas the discussion is about whether AI will severely impact/kill the coding job market.

For which question, whether writing code is the most difficult part or not is irrelevant. It's still the part on whch most employed develoopers spend the most time on, and thus, if it gets more/fully automated, a lot of those jobs will be killed.

[magicalhippo]

> For which question, whether writing code is the most difficult part or not is irrelevant.

I find it exactly opposite, in that because most coding is not very difficult in most cases, it is the part that will be replaced by AI.

[magicalhippo]

> but still need many man-weeks or months or even years with a large team to make it into code.

We planted our own hedge. The hard part was finding the right type of bush that we liked but could grow well here yet not require too much maintenance, and how to properly plant them. The easy but really time consuming part was digging the trench, preparing the soil (including drainage where needed), putting the plants in, fill back up and then make sure they didn't die the following week.

> But 95% of it is regular coding roles, even mostly boring coding roles.

Which AI can now do very well if guided by someone who knows the requirements and targets very well, and knows how to test the product well.