[bcherny]

The advice isn’t about coding vs managing. What John is saying is to deeply understand why you’re building something, so that you can build it better. If you over focus on the what — the implementation, the language, the approach — you won’t be as good, and your work may be increasingly replaced by AI.

[klabb3]

> The advice isn’t about coding vs managing.

Definitely. Carmack is no dummy, but I’d argue this comment section proves that he gave a pretty bad answer here (bad for the audience, not if you know Carmack and what he means).

I guess it’s the impostor syndrome, but many programmers have an out-of-place reductionist view of their work. It’s not simple, and crud boilerplate proves little about the future prospects.

Managers OTOH really are in the zone of GPT parity. At least a much larger subset of their day-to-day activities. So are many soft skills. In fact, soft communication is where LLMs shine above all other tasks, as we’ve seen over and over in the last few months. This is supported by how it performs on eg essay-style exams vs leetcode, where it breaks down entirely as it’s venturing into any territory with less training data.

Now, does that mean I think lowly of managers? No, managers have a crucial role, and the ones who are great are really really crucial, and the best can salvage a sinking ship. But most managers aren’t even good. That has a lot to do with poor leadership and outdated ideas of how to select for and train them.

[ilyt]

> Definitely. Carmack is no dummy, but I’d argue this comment section proves that he gave a pretty bad answer here (bad for the audience, not if you know Carmack and what he means).

I dunno, I got what he meant from the start, and the same advice was given by many people in many forms, usually in variant of "well, the business doesn't give a shit about details but the end product".

> Now, does that mean I think lowly of managers? No, managers have a crucial role, and the ones who are great are really really crucial, and the best can salvage a sinking ship. But most managers aren’t even good. That has a lot to do with poor leadership and outdated ideas of how to select for and train them.

I joked some managers could be replaced by forward rule in mailing system, ChatGPT is an upgrade on that.

[ScoobleDoodle]

I agree but I think I’d call it the “how” rather than the “what”. You might mean “what tool”, but I also think of “what feature”.

[nosianu]

EDIT: Was the comment edited, or did my brain miss something? I think I perceived something else there when I wrote my response.

It's still "how". Only on a higher level. For example, instead of placing the form elements exactly and designing them you describe data flow and meta info about the data to be gathered via the form, and how it looks and where elements are placed on various screens happens automatically.

Writing code in a higher level vs. assembler still is coding, but you worry about very different things. Just compared with assembler, since looking back is easier than looking forward. Instead of worrying about (the few) registers and interrupts and in which RAM cells you place what you now think about very different things. It still is programming though, and you still tell the machine "how". Only on a different level.

When you lead a large company instead of working with a machine on the factory floor the work is very different, you still need precision and know "how", only on a different level. Even if you have "underlings" who can think, and you can let them execute the sub tasks you create, you still have to know what you are doing, only on a higher level.