AI has increased the perception of productivity to unwitting managers, which is a dangerous pit to be falling into.
It's like a lot of those fabled 10x engineers that I've run into over the years; they may seem like they're incredibly productive and managers love them, but in reality their work is so riddled with issues and rushed half-baked ideas, that they end up costing the other engineers on the team 10x the amount of their time to review and fix it all. And it's those other engineers who do all of the invaluable but underappreciated cleanup work who are getting the boot right now.
Oh I'm sure they exist, that's why I said "a lot", not "all". My point is more about how managers perceive the employees they manage. They all want to believe that they found the goose that lays the golden eggs, but in reality so many perceived rockstar engineers are just over-confident or over-eager people mass producing garbage.
That's the important parallel with using AI in production: on the surface it may seem like the solution to all productivity problems, but in reality it still takes a lot of human brainpower to review, assess, fix and maintain whatever the AI model spits out. Especially since generative AIs like to present their work with supreme confidence, and it's getting increasingly hard to separate the chaff from the wheat.
I think a sizeable % of engineers can be 10x-ers as ICs / leads - but it's kind of an "in the zone" thing, when they're experienced in the stack and the product, can navigate the org, are motivated and aren't held back by issues with team or management (or personal issues etc).
But it's like growing plants, in the sense that it only takes one crucial element to be missing to limit growth (or "output").
I've met a few ppl who were _reliable_ 10xers (could navigate a much wider range of techs / domains / situations than others) but I've seen even the best crash and burn at times when the circumstances aren't right.
Widespread use of tools like Copilot hasn’t been happening for long enough for you to draw these generalisations with an appropriate degree of confidence. This just comes across as you misrepresenting your personally-aligned hypothetical as reality.
I have come to the conclusion that a lot of the 10x developer stuff is basically down to the 10x person being the original author of the code.
Every other developer on the project has to do at least 3 times as much work, they have to read through the original code and try and understand what the original person was thinking of when writing it before even getting to solving the problem at hand.
A true 10x developer would write code good enough that every other developer could maintain it as easily as the original author.
I have only used the free version of ChatGPT and I'll admit that it has been helpful, but only for maybe 50% of the stuff In ask it.
Maybe 15% of the time it has been a hindrance by making up stuff. It's basically a faster version of Stack Overflow / web search.
A lot of IT work is maintenance work, and it knows nothing about those systems. It's good for me for JavaScript as I spend 80% of my time on the backend. I know enough JavaScript to debug other peoples work, or to add some simple jQuery / vanilla JS to an HTML page (which is all that is needed in the majority of cases), but if I was to build a React App from scratch, it would likely look like beginner level stuff.
Has anyone had experience with the paid version of ChatGPT? How does it compare?
I feel I am one of those developers that codes 100 times faster since ai tools came out. I was not slow before by the way. But now I can take a well defined interface and class example, ask gpt4 for a new well defined variant, and out it comes. one tweak and a test later its done.
Some things that would have taken me weeks take days. And some things that used to take days take 1-3 hours.
Novel work can't yet be automated. But... most of the work I do is not novel, and theres files and files of preexisting context for copilot to consume.
As for this effect being widespread and noticed enough to be related to the firing. I dont think its the case.
The people I work with don't seem to get nearly as much out of the tools as I do, and not for lack of trying. They're constantly using the tools too. Even if companies are considerably more productice with the tools, I do not think management would have been attuned enough to performance and costd to have instigated layoffs from it.
You’re benchmarking yourself based on your own experience of course. Not to say this is horribly wrong but yeah, doesn’t tell the rest of us much about productivity.
Like saying that you run faster after a cup of coffee, cool ?
Much of what I've done with ai tools I have had to do before and struggled over days. I was a python dev mostly but I had cases at work where I needed to write java, c, or c++. The genicam compliant industrial camera nonsense it got me through in a few hours was insane. And using a fuzzy logic library. I did not have to spend tons of time reading the documentation, struggling with cmake, or figuring out how to install and link dlls or even syntax for that matter. I can start to type a comment of a high level description of a small codeblock and copilot spits it out.
This works in any language.
It isnt that it speeds up every single operation I do. When I get in the zone I toggle it off. But 80% of the time i hit a snag it instantly short circuits me through the issue. I can and do ask follow up questions about how the thing works so the next time I come across it it wont even be a snag. So I have had slowly compounding speed boosts from it for two years. I code faster with it also because it taught me things, and I have learned when to turn it off (often)
It's starting to replace some low-level jobs, like copywriting and art-based positions. For development, the only thing I've found it good for is giving me example usage for custom Apis and other things that I can't just Google/download.
It's marginally better than Stackoverflow (which I've only ever used to give me an idea that I then turn into my own code, usually written from scratch, because most code is littered with security vulnerabilities).
It's a long way from replacing developers or increasing productivity on a measurable level. It's also like building your business on quick sand. Some feature you use today might be put behind a higher pay-tier (or removed) at any time.
