Gabe Newell thinks AI tools will result in a ‘funny situation’ where people who don’t know how to program become ‘more effective developers of value’ than those who’ve been at it for a decade
Gabe Newell thinks AI tools will result in a ‘funny situation’ where people who don’t know how to program become ‘more effective developers of value’ than those who’ve been at it for a decade
35 Comments
TL;DR:
> people who don’t know how to program who use AI to scaffold their programming abilities will become more effective developers of value than people who’ve been programming, y’know, for a decade.
—Gabe Newell
He does go on to say that more programming knowledge/experience still helps, though.
The problem is that it’s all uselessly delicate. The first time you have a problem you can’t explain well, or the AI can’t understand, your code is completely useless. We don’t have a fix for that in sight without real, actual coders
What will actually happen is they will put out garbage and rely heavily on senior programmers to QA. Rip all the low level programmers, but the really good ones will still manage
It’ll be interesting to see the affect of AI on the indy gaming scene in the next few years. I suspect we are going to be seeing a lot of AI made shovelware since it will take even less time then asset flipping + basic coding. But the other side of that is that more creative people with limited coding skills will be able to do more as well.
It’ll probably create a situation where people with great ideas but poor coding skills become more productive than people with poor ideas but great coding skills.
There were a couple recent reports from companies incorporating AI for their coders that it actually led to around 25% decrease in efficiency and time to program. The reason why was that the code often needed fixing, and it often worked for a task but that didn’t scale or integrate with other tasks.
AI vastly decreases effort to put out nonsense, but it’s not capable of handling more critical components. Which makes sense – it doesn’t understand it just apes.
I don’t buy that
It’s fine for simple tasks like building web scrapers but if you don’t understand fundamentals, you can very quickly end up in a mess
I think people who don’t know how to program will be able to create bigger and bigger messes in the code base unless they understand how to ask the AI to refactor and clean up as they add features.
At some point the non programmer needs to become a programmer if they want to write software that isn’t crap.
If you don’t know how to program and the ai is wrong then you won’t know it’s wrong. It’s often wrong.
Gabe said anti cheat is simple to solve with ai yet here we are
It’s not there yet. It’s there for prototyping – but not for building a real application.
AI generated code creates a ton of syntax errors because it has a hard time evaluating the entire code as a whole when adding or subtracting features – and you’ll end up bricking your program. So you still need to know what you’re doing.
I do believe it will get there very soon though. And here I more or less agree – because people who have been gaining experience in other domains will now also be able to create things. But at the same time, there will be many software engineers that will be able to use their systems thinking to also create a ton of value with AI scaling up their production.
Won’t be funny for long when they figure people that can’t code can’t debug errors made by AI when generating said code.
That’s a thing ***everybody*** who know how to code said for a while.
Bizarre take
Have fun fixing the little problems. Now everyone is going to tey to release half.broken apps and games with knownidea how to fix it.
Oh shit, our lord and savior said something positive about AI, how will Reddit react to this conundrum?!
Hell no
But GabeN would know I guess
I’ve been a solo dev for a decade and it helps a TON
But I know my codebase like the back of my hand–not for long!
Nah. It’s not going to happen any time soon. Software architecture is a thing for a reason. Ai still can’t do the big picture. Then, when they can, programming won’t exist as a job any more.
I think it’ll drastically lower the skill floor for introductory/ basic programming, which can be a good thing. But it looks worthless for stuff beyond the basics. And anyone needing to take the step from that lowered floor up to the next level is going to have a significantly harder time.
[to wit](https://www.pcgamer.com/gaming-industry/gabe-newells-daily-routine-is-get-up-work-go-scuba-diving-says-hes-been-retired-for-a-long-time-but-works-7-days-a-week-the-things-i-get-to-do-every-day-are-super-awesome/)
I stopped listening to Gabe Newell, when I realized he’s not really a game developer anymore.
* Steam is a [very good] platform. A+.
I don’t really listen to Neil Druckmann too much, or Bethesda’s Todd Howard, but at least they each have created innovations and whole games, multiple actually in the last 20 years.
*Why does Newell even get any press?* What is that they (Valve) do? What do they make?
* Steamdeck is still a specialized, not so universal box (buy a Sony / MS console FFS).
I realize that as a member of the unwashed masses, I need be drowned and cleansed. Maybe this is not a subject for a casual gamer like myself? I’ve been sitting here playing FPS PC Games since Columbine, running my Steam system since 2004/2005. There isn’t a month that goes buy on my PC systems where Steam is not running.
If this is what Gabe Newell thinks about AI versus traditional programmers, he ought to be ignored for the rest of time.
I think a large portion of developers could dramatically increase their productivity with the use of tools like snippets and generators. Ironically, I think a large portion of what slows development down is not establishing architecture and coding standards, encoding them in generators and snippets, writing unit tests and automation tests, doing proper domain modelling, and writing clear and complete requirements before starting work. Basically, they skip everything that will make their lives easier to move faster.
He’s completely right, but that is a very *very* small window.
When your code is all vibe based, you will then need to hire real developers to fix your problems because the AI can’t figure out how to help, and will keep breaking things more and more.
At which point companies will realize these vibe coders are a bigger liability than a traditional developer.
It’ll take a couple of years, but it wont take long. And people who actually know these systems and how to program will be able to charge a premium *because you will need us*.
Ofcourse, management are usually pretty dumb, so who knows how this will all play out in reality.
Yeah, doubt it, the most important part of learning to code and practicing code is in learning how to think critically and logically. Learning any given language is far less of an issue.
Did he interview from one of his many ships?
Sure, in the sense that AI tools will be much more effective tutors than either university professors or youtube videos.
But no, there’s no shortcut where you don’t learn how to program and just have the AI do it for you.
Another way to look at it is that AI coding tools are like scientific calculators. They’re way better than having to pull out the book of trig or log tables, but having a “sin” button isn’t useful at all if you have no idea what an angle is.
Right up until it’s time to start diagnosing and fixing bugs.
I’ve programmed with chatgpt and while it does work, it does help to have some background knowledge to help navigate it to the right direction. Otherwise you just get horrible code.
Programming is about LEARNING to use the tools you have at your disposal. Even libraries update and change.
The good ones are already using AI. They’ve been automating everything for a while now.
AI is not correct 100% of the time. You still need someone who understands when it’s wrong and know how to fix it’s mistakes.
Is he rolling a d20 in the image?
Not with the current state of AI. Any app made with AI is currently unmaintainable. It’s typically so broken that if you don’t know how to fix it on your own, it stays broken. And instead of fixing it (because they don’t know how), the AI “programmer” just prompts the AI to work around problems. The end result is layers upon layers of broken system over broken system.
I use AI to summarize information. For me, it’s another layer on top of Google that saves me a bit of time reading. And I spot errors at least 20% of the time. “Hallucinations”. Because I can spot them, I can discard them. But someone who doesn’t know better? It goes into the app.
So ya, maybe 10-20 years from now we’ll see solo/small team indie devs using AI to create things that would have previously been impossible simply because the programming part will be much easier, but if you’re expecting a massive rush of awesome indie games made with AI, you’re headed for disappointment. The only people claiming today’s AI can make viable apps are people who don’t know the difference between good code and bad. You don’t want games made by those people.
I really don’t understand rich guys obsession with AI.
Hiring humans has made Gabe several billion dollars. A lot of his genius is based on hiring adroitly and having Steam making ridicule money with only 200 employees.
So why would you think there’s something better? There’s no possible path to things being better than they already are for him.
It’s really baffling.
Coders have been standing on the backs of other coders forever, that’s how it works
Thats all well and good until something doesn’t work right and they have no idea why and the AI can’t help them. Using AI makes you stupid.
same goes for art and writing.
The minute something needs to change, what are these fundemental-less people going to do?