4 ms·

This might be a contentious take but I have concluded that people that work in tech need to adjust to the idea that "coding" as a sole technical proficiency wil

This might be a contentious take but I have concluded that people that work in tech need to adjust to the idea that "coding" as a sole technical proficiency will become increasingly less valuable as time goes on, and other technical skills that don't necessarily involve "coding" (such as OS knowledge, networking, etc.) will become more valuable. In other words I feel really bad for bootcamp people right now.

HumblyTossed 3y ago
The actual act of coding has always been such a small part of the job anyway.
- JohnMakin 3y ago
  I agree, and my least favorite - so it's weird that the majority of interviews for DevOps/SRE positions typically involve a large amount of programming tests.
  - codemonkey100 3y ago
    [dead]
  - Aeolos 3y ago
    If you have ever been on the other side of the interview divide, and noticed that >50% of "programmers" cannot write a single line of code, you might not find it so weird anymore... Not just self-proclaimed programmers either. I've had a chap claim 5+ years of C++ experience working on GE nuclear reactor software, who couldn't explain how C++ memory management works, or delete vs delete[]. So yes, programming interviews it is, unless we get some different form of accreditation like doctors or lawyers.
    - jwestbury 3y ago
      Oddly -- I've never had this problem in interviews. Is it possible that it's better screening happening pre-interview at FAANG and FAANG-like places? For what it's worth, I like practical programming problems. The problem comes when the programming interviews don't match the job. Why do I need to write a sorting algorithm in an interview for an SRE position? (I worked at Dropbox for a couple of years, and my interview there was probably the best SRE coding interview I've had - detect duplicate files in a filesystem, with follow-ups about improving performance.)
disqard 3y ago
An excellent point! It reminds me of how drafting was thoroughly disrupted by AutoCAD: https://rarehistoricalphotos.com/life-before-autocad-1950-1980/ https://rarehistoricalphotos.com/life-before-autocad-1950-19... How much of (our, as in, programmers') work is the "writing code" part, vs. all of the other stuff? Build consensus, gather feedback, communicate ideas, resolve ambiguities, make non-zero progress in the absence of clear direction, etc.
- ska 3y ago
  > How much of (our, as in, programmers') work is the "writing code" part, In my experience, For a junior/entry level, 80%+ For a (actual, not just title) Sr. dev? Typically <= 20-25%.
  - nradov 3y ago
    There's no way that even juniors are spending as much as 80% of time actually writing code. They spend >20% attending meetings, completing mandatory training, reading user stories, participating in code reviews, waiting for test execution, etc.
    - ska 3y ago
      That's highly dependent on scale of org, but fair enough. Sometimes it's closer to 60. And sometimes your seniors are only seeing 10-15, for the same reason.
  - didntknowya 3y ago
    idk it'll closer to 60% coding, of which >50% would be looking thing up rather than actually writing anything
- kragen 3y ago
  https://skeptics.stackexchange.com/questions/17224/do-professional-software-developers-write-an-average-of-10-lines-of-code-per-day https://skeptics.stackexchange.com/questions/17224/do-profes... 10–20 lines of code per day means the “writing code” part is about 2 minutes out of an 8-hour day, if you only count the code that ends up being delivered. maybe 10× that including unit tests, drafts that get replaced, duplicate code that gets refactored away, test code to see what a library will do, etc. but ai can obviously handle a lot of that stuff too
mattgreenrocks 3y ago
We are overdue for something to nuke the amount of boilerplate code we have to deal with, especially with anything that touches the web.
- cgdub 3y ago
  We’ve had Lisp for 60 years, but instead of using that, we’re going to have LLMs write the boilerplate for us. You will still have to read the boilerplate of course.
  - mattgreenrocks 3y ago
    Another week, another article lauding how simple Golang is, and everyone nodding in agreement.
    - starttoaster 3y ago
      As a professional Golang dev (among other responsibilities), I see more hate for Go on the internet than praise. Like most things people complain about, I usually can barely discern what people's exact complaints are, as most people seem to just complain about the unknown. I take it you get the opposite on your travels around the internet, and I'm just wondering, how do I get into your newsfeeds? The internet algorithms seem to be putting us both in the wrong places, and I'm wondering if we might be having a Freaky Friday moment, if you will.
      - nradov 3y ago
        I don't hate Go, I'm just disappointed in it. Decades of programming language research to draw from, and the best they could manage was a warmed over C++? I know that Go is practical and useful but it feels like a lost opportunity, a "local maximum" if you will. Ultimately I think it will prove to be a dead end with no path to the future.
        starttoaster 3y ago
paxys 3y ago
The "writing code" part of coding has never been valuable in the first place, so AI isn't going to shake things up nearly as much as people think. So you built an AI that can parse a perfectly written spec and produce code? The only thing that changed is that the person writing the spec is now a programmer.
- dunefox 3y ago
  > The only thing that changed is that the person writing the spec is now a programmer. Yes, but that is not necessarily the software developer of today.
- gumby 3y ago
  Indeed. The point of FORTRAN was that you just typed your FORmulas in and the compiler TRANslated them to machine code. It was a simple alternative to programming! So I guess there have been no programmers since the 1950s.
- tshaddox 3y ago
  "Perfectly written spec" is already a description of what computer code is.
- circuit10 3y ago
  The impressive part of AIs like ChatGPT is their natural language understanding, they can infer things that aren't explicitly said and ask follow up questions if needed. They aren't powerful enough to write complex programs yet, though.
- JohnMakin 3y ago
  > The "writing code" part of coding has never been valuable in the first place of course it has been, even very recently. where do you think all the demand for bootcamps and bootcamp graduates came from years ago? Why does basically every SWE interview consist of a large % of programming tests? As for the rest of your post, I didn't say anything like that at all - but it seems like a trivial conclusion to say that "coding" will become less valuable as the interfaces become simpler to it (such as human language interfaces). This is hardly controversial and probably axiomatic.
gardenhedge 3y ago
Code is a tool used in companies to satisfy a business requirement. It has always been like this.
codemonkey100 3y ago
[dead]
jimbokun 3y ago
Will AIs not know about OS and networking?
- starttoaster 3y ago
  Sure they will, they'll just be trying to tinker with firewalld settings on Ubuntu servers, or ufw on RHEL.
tarkin2 3y ago
I’m not sure how safe networking and similar folk are either. You’ll still need experts who understand the generated network code, and experts who can untangle the generated code, but not so many. Code will be a tool used by non-experts; fewer special-case experts will be needed. Where we’ll be when the experts die out, I’m not sure.
sterlind 3y ago
I don't think domain expertise is a competitive bottleneck. These LLMs started with the sum total of human knowledge in their training data, but (currently) lack the ability to synthesize that knowledge, to reason and plan to achieve goals. GitHub Copilot and ChatGPT-4 have roughly the same breadth of knowledge, but Copilot can only make superficial suggestions, while ChatGPT-4 is smart enough to write basically any toy script as long as it fits in a page and doesn't require inventing anything new. The difference between the average bootcamp graduate and the average engineer isn't intelligence, it's knowledge. Once AI makes bootcamp graduates obsolete, I see no reason why engineers in general won't be obsolete too.
bloppe 3y ago
To add contention: if you find LLMs are super useful for helping you do your job, you should start looking for a new job. Not necessarily a completely different job, just one where AI won't be able to help all that much without another technological breakthrough, like working with large (several million LOC) codebases.
jahsome 3y ago
As someone who dropped out of CS, I've always viewed coding as a "foot-in-the-door." Coding is a means to an end, and when someone only cares about writing a bitchin' line/method/program struggles to identify and understand (or more importantly communicate) the end, what good is the means?
timeon 3y ago
> and other technical skills that don't necessarily involve "coding" (such as OS knowledge, networking, etc.) will become more valuable. You can pass job description for Advent of Code task to these curve fitting tools and it will return solution. It is not just about coding. Other technical skills may have become less relevant too.
emporas 3y ago
Algorithms and data structures however? Still important! Rust implements some algorithms [1] for the same API with the sole purpose to be interchangeable and readily replace one with another. Does a computer know how to do that? I find that not to be the case at the moment. Algorithms were important in the past, but a lot of programming effort had to be put on mastering every compiler/OS/architecture and their quirks in every case. Now the focus of programmers, will shift more to knowledge about algorithms and data structures. https://dhghomon.github.io/easy_rust/Chapter_32.html https://dhghomon.github.io/easy_rust/Chapter_32.html

HumblyTossed 3y ago

The actual act of coding has always been such a small part of the job anyway.

JohnMakin 3y ago
I agree, and my least favorite - so it's weird that the majority of interviews for DevOps/SRE positions typically involve a large amount of programming tests.
- codemonkey100 3y ago
  [dead]
- Aeolos 3y ago
  If you have ever been on the other side of the interview divide, and noticed that >50% of "programmers" cannot write a single line of code, you might not find it so weird anymore... Not just self-proclaimed programmers either. I've had a chap claim 5+ years of C++ experience working on GE nuclear reactor software, who couldn't explain how C++ memory management works, or delete vs delete[]. So yes, programming interviews it is, unless we get some different form of accreditation like doctors or lawyers.
  - jwestbury 3y ago
    Oddly -- I've never had this problem in interviews. Is it possible that it's better screening happening pre-interview at FAANG and FAANG-like places? For what it's worth, I like practical programming problems. The problem comes when the programming interviews don't match the job. Why do I need to write a sorting algorithm in an interview for an SRE position? (I worked at Dropbox for a couple of years, and my interview there was probably the best SRE coding interview I've had - detect duplicate files in a filesystem, with follow-ups about improving performance.)

disqard 3y ago

An excellent point! It reminds me of how drafting was thoroughly disrupted by AutoCAD: https://rarehistoricalphotos.com/life-before-autocad-1950-1980/ https://rarehistoricalphotos.com/life-before-autocad-1950-19... How much of (our, as in, programmers') work is the "writing code" part, vs. all of the other stuff? Build consensus, gather feedback, communicate ideas, resolve ambiguities, make non-zero progress in the absence of clear direction, etc.

ska 3y ago
> How much of (our, as in, programmers') work is the "writing code" part, In my experience, For a junior/entry level, 80%+ For a (actual, not just title) Sr. dev? Typically <= 20-25%.
- nradov 3y ago
  There's no way that even juniors are spending as much as 80% of time actually writing code. They spend >20% attending meetings, completing mandatory training, reading user stories, participating in code reviews, waiting for test execution, etc.
  - ska 3y ago
    That's highly dependent on scale of org, but fair enough. Sometimes it's closer to 60. And sometimes your seniors are only seeing 10-15, for the same reason.
- didntknowya 3y ago
  idk it'll closer to 60% coding, of which >50% would be looking thing up rather than actually writing anything
kragen 3y ago
https://skeptics.stackexchange.com/questions/17224/do-professional-software-developers-write-an-average-of-10-lines-of-code-per-day https://skeptics.stackexchange.com/questions/17224/do-profes... 10–20 lines of code per day means the “writing code” part is about 2 minutes out of an 8-hour day, if you only count the code that ends up being delivered. maybe 10× that including unit tests, drafts that get replaced, duplicate code that gets refactored away, test code to see what a library will do, etc. but ai can obviously handle a lot of that stuff too

mattgreenrocks 3y ago

We are overdue for something to nuke the amount of boilerplate code we have to deal with, especially with anything that touches the web.

cgdub 3y ago
We’ve had Lisp for 60 years, but instead of using that, we’re going to have LLMs write the boilerplate for us. You will still have to read the boilerplate of course.
- mattgreenrocks 3y ago
  Another week, another article lauding how simple Golang is, and everyone nodding in agreement.
  - starttoaster 3y ago
    As a professional Golang dev (among other responsibilities), I see more hate for Go on the internet than praise. Like most things people complain about, I usually can barely discern what people's exact complaints are, as most people seem to just complain about the unknown. I take it you get the opposite on your travels around the internet, and I'm just wondering, how do I get into your newsfeeds? The internet algorithms seem to be putting us both in the wrong places, and I'm wondering if we might be having a Freaky Friday moment, if you will.
    - nradov 3y ago
      I don't hate Go, I'm just disappointed in it. Decades of programming language research to draw from, and the best they could manage was a warmed over C++? I know that Go is practical and useful but it feels like a lost opportunity, a "local maximum" if you will. Ultimately I think it will prove to be a dead end with no path to the future.
      - starttoaster 3y ago

paxys 3y ago

The "writing code" part of coding has never been valuable in the first place, so AI isn't going to shake things up nearly as much as people think. So you built an AI that can parse a perfectly written spec and produce code? The only thing that changed is that the person writing the spec is now a programmer.

dunefox 3y ago
> The only thing that changed is that the person writing the spec is now a programmer. Yes, but that is not necessarily the software developer of today.
gumby 3y ago
Indeed. The point of FORTRAN was that you just typed your FORmulas in and the compiler TRANslated them to machine code. It was a simple alternative to programming! So I guess there have been no programmers since the 1950s.
tshaddox 3y ago
"Perfectly written spec" is already a description of what computer code is.
circuit10 3y ago
The impressive part of AIs like ChatGPT is their natural language understanding, they can infer things that aren't explicitly said and ask follow up questions if needed. They aren't powerful enough to write complex programs yet, though.
JohnMakin 3y ago
> The "writing code" part of coding has never been valuable in the first place of course it has been, even very recently. where do you think all the demand for bootcamps and bootcamp graduates came from years ago? Why does basically every SWE interview consist of a large % of programming tests? As for the rest of your post, I didn't say anything like that at all - but it seems like a trivial conclusion to say that "coding" will become less valuable as the interfaces become simpler to it (such as human language interfaces). This is hardly controversial and probably axiomatic.

gardenhedge 3y ago

Code is a tool used in companies to satisfy a business requirement. It has always been like this.

codemonkey100 3y ago

[dead]

jimbokun 3y ago

Will AIs not know about OS and networking?

starttoaster 3y ago
Sure they will, they'll just be trying to tinker with firewalld settings on Ubuntu servers, or ufw on RHEL.

tarkin2 3y ago

I’m not sure how safe networking and similar folk are either. You’ll still need experts who understand the generated network code, and experts who can untangle the generated code, but not so many. Code will be a tool used by non-experts; fewer special-case experts will be needed. Where we’ll be when the experts die out, I’m not sure.

sterlind 3y ago

I don't think domain expertise is a competitive bottleneck. These LLMs started with the sum total of human knowledge in their training data, but (currently) lack the ability to synthesize that knowledge, to reason and plan to achieve goals. GitHub Copilot and ChatGPT-4 have roughly the same breadth of knowledge, but Copilot can only make superficial suggestions, while ChatGPT-4 is smart enough to write basically any toy script as long as it fits in a page and doesn't require inventing anything new. The difference between the average bootcamp graduate and the average engineer isn't intelligence, it's knowledge. Once AI makes bootcamp graduates obsolete, I see no reason why engineers in general won't be obsolete too.

bloppe 3y ago

To add contention: if you find LLMs are super useful for helping you do your job, you should start looking for a new job. Not necessarily a completely different job, just one where AI won't be able to help all that much without another technological breakthrough, like working with large (several million LOC) codebases.

jahsome 3y ago

As someone who dropped out of CS, I've always viewed coding as a "foot-in-the-door." Coding is a means to an end, and when someone only cares about writing a bitchin' line/method/program struggles to identify and understand (or more importantly communicate) the end, what good is the means?

timeon 3y ago

> and other technical skills that don't necessarily involve "coding" (such as OS knowledge, networking, etc.) will become more valuable. You can pass job description for Advent of Code task to these curve fitting tools and it will return solution. It is not just about coding. Other technical skills may have become less relevant too.

emporas 3y ago

Algorithms and data structures however? Still important! Rust implements some algorithms [1] for the same API with the sole purpose to be interchangeable and readily replace one with another. Does a computer know how to do that? I find that not to be the case at the moment. Algorithms were important in the past, but a lot of programming effort had to be put on mastering every compiler/OS/architecture and their quirks in every case. Now the focus of programmers, will shift more to knowledge about algorithms and data structures. https://dhghomon.github.io/easy_rust/Chapter_32.html https://dhghomon.github.io/easy_rust/Chapter_32.html