OpenAI introduces Codex, its first full-fledged AI agent for coding

Uncivil Servant

Ars Scholae Palatinae
4,726
Subscriptor
"Production-ready code"

Oh I can assure you that phrase does not mean anything remotely close to whatever the hell they think it means, but that's a great Friday quote!

Remember, people, what matters is total lines of code, not dependencies, bugs, security flaws, or implementation!

Test in Prod, what could go wrong?
 
Upvote
179 (188 / -9)

Uncivil Servant

Ars Scholae Palatinae
4,726
Subscriptor
Interesting now how is it going from "I have an idea for a program" to "I have a program that works"?

Concepts of a plan is the same as a plan. And once something is planned, implementation is usually just an afterthought, amirite?

(/s and you have no idea how painful this is to type)
 
Upvote
111 (111 / 0)
Post content hidden for low score. Show…

picklefactory

Ars Praetorian
401
Subscriptor
How's that any different from now?
As Uncle Joe said, "Quantity has a quality of its own."

I suppose 𝓉𝒽𝑒 𝓋𝒾𝒷𝑒𝓈 don't matter so much when a machine must interpret this output. But I don't envy the humans that have to read it that weren't doing the prompt-doodling.
 
Upvote
43 (44 / -1)
I don't personally know any programmers who don't use AI at this point. Even at shops that forbid normal AI (like Disney) they have local agents.

It feels like this is a genie that's not going to be stuffed back into the bottle.
True, though I still argue it's leading to far worse code overall. I've noticed our devs (major tech company) have become highly reliant on it for some core components of our application and those have turned into complete disasters. Bugs started trickling in and now they are flooding in as they vibe code their way to a completely unworkable implementation. We are already having to start from scratch on a few areas.
 
Upvote
122 (123 / -1)
Post content hidden for low score. Show…

Aurich

Director of Many Things
41,040
Ars Staff
True, though I still argue it's leading to far worse code overall. I've noticed our devs (major tech company) have become highly reliant on it for some core components of our application and those have turned into complete disasters. Bugs started trickling in and now they are flooding in as they vibe code their way to a completely unworkable implementation. We are already having to start from scratch on a few areas.
I am fully convinced that all AI—regardless of how good it is, how cool it is, how capable—is fundamentally leading humanity into being more lazy and using our brains less.

I really don't know where we go from here.

But I also recognize the value. Ultimately it's up to the individual and how they use it, but that gets to the core of the problem: faith in the general masses.
 
Upvote
101 (105 / -4)
I don't personally know any programmers who don't use AI at this point. Even at shops that forbid normal AI (like Disney) they have local agents.

It feels like this is a genie that's not going to be stuffed back into the bottle.
I don't think we're waiting for it be stuffed back in the bottle. This entire time AI has been presented as a solution looking for problems. A solution that my testing has shown to not meet any of my case use scenarios so far. I do know some people who use it to find issues with code validated by the tools their work gives them that don't run. Not looking to use AI for anything. What I want is companies to present their new tool which is better than the old tool and if that happens to use AI tech, that's fine, but if they lead with "it's AI", I'm going to look at if skew eyed.
 
Upvote
68 (73 / -5)
I don't personally know any programmers who don't use AI at this point. Even at shops that forbid normal AI (like Disney) they have local agents.

It feels like this is a genie that's not going to be stuffed back into the bottle.
Well, you don’t know me, but I for one am a programmer who doesn’t use AI. Given that I’ve spent the last year using non-AI tools to reduce static-analysis-flagged issues in our code base by 10x(ish), and made the MTBF get boringly large in the process, I don’t think my job security is in any great danger any time soon…
 
Upvote
96 (106 / -10)

Legatum_of_Kain

Ars Praefectus
4,068
Subscriptor++
Criticism of LLMs aside, the release of a model built for one specific task is pretty much the only focused business case use I've seen for AI so far.
But that was a solved problem already, without confabulations, less space, less power usage, and cheaper, in like literally the intelligence product Microsoft had before LLMs.

An absolute insanity would be changing something that worked and gave you good, accurate code and instead using…. Whatever this is. The expensive bullshit machine?
 
Upvote
58 (60 / -2)

Uncivil Servant

Ars Scholae Palatinae
4,726
Subscriptor
When I need software now I go to the App Store. If it’s there for a couple bucks I buy it. If it’s add supported or non existent I ask Claude to build it for me. I have a nice little library of programs now. I’m selling a few of them for a buck. It’s great being able to think of exactly what you need and build it in a few hours.

This is what worries me: the idea that the hours spent writing the code are the majority of the work. I do not understand this attitude. This is not how software development works. This might be fine for developing Call of Duty: Cannibal Zombie Apocalypse, but not so much for anything involving, say, monetary transactions or PHI.

If it really doesn't matter whether your code works, if undocumented and undocumentable behavior is not a showstopper, then AI coding sounds great!
 
Upvote
108 (110 / -2)

jmarlin

Technical Director
411
Ars Staff
Bugs started trickling in and now they are flooding in as they vibe code their way to a completely unworkable implementation. We are already having to start from scratch on a few areas.
So true. In my experience, it's been best to know exactly what you want your class or function or whatever to do (and know how to do it by hand if you had to). Then use discreet comments for each section to prompt the tool. e.g. // Loop over users array, finding those with malformed emails

When I've let it generate entire classes, it's generally been a mess. I weep for the senior engineers that will be un-f*cking some of this stuff as time goes on. Still, I'm interested in checking out this new tool!
 
Upvote
41 (42 / -1)
Yes, it’s Codex! The A.i. sensation that’s sweeping the nation! Only $149500.00 at participating stores! Get one today!

Warning: Pregnant women, the elderly, and Elon Musk should avoid prolonged exposure to Codex.


Caution: Codex may suddenly accelerate to dangerous speeds.

Codex contains a madeupium core, which, if exposed due to hacking, should not be touched, inhaled, or looked at.

Do not use Codex on anything. Ever.

Discontinue use of Codex if any of the following occurs:

itching
vertigo
dizziness
tingling in extremities
loss of balance or coordination
slurred speech
temporary blindness
profuse sweating
or heart palpitations.

If Codex begins to smoke, get away immediately. Seek shelter and cover head.

Codex may stick to certain types of tech journalist.

When not in use, Codex should be returned to its special container and kept under refrigeration. Failure to do so relieves the makers of Codex, OpenAI, of any and all liability.

Ingredients of Codex include all sorts of stolen code used to train it on.

Codex has been shipped to our air traffic controllers in Newark and is being dropped by our warplanes on Cleveland.

Do not taunt Codex.

Codex comes with absolutely no lifetime warranty.
 
Upvote
25 (46 / -21)

Uncivil Servant

Ars Scholae Palatinae
4,726
Subscriptor
Why wouldn't you still test? Especially when the instructions say to test?

Because they're claiming to produce "production-ready code". Code that is ready to proceed to QA/SIT/UAT is usually considered in development, yes?

I was under the impression that there is an important distinction between alpha/beta/testing/dev/prerelease code, whatever words one wants to use, and production code?
 
Upvote
54 (60 / -6)

MrRtd

Ars Centurion
290
Subscriptor
I'm getting the feeling that these frequent announcements from OpenAI have more to do with keeping the hype up more than anything else.

This is a situation where tech reporting falls far short. A lot of these things should be advertisements not articles - they should be paying you to promote their product (or maybe they are and you're just not telling us the reader?)
 
Upvote
61 (70 / -9)

Aurich

Director of Many Things
41,040
Ars Staff
Well, you don’t know me, but I for one am a programmer who doesn’t use AI. Given that I’ve spent the last year using non-AI tools to reduce static-analysis-flagged issues in our code base by 10x(ish), and made the MTBF get boringly large in the process, I don’t think my job security is in any great danger any time soon…
I mean, I hope that for you, and all my friends.

I think job security in programming is probably not going to be great going forward, but how that plays out in practice I couldn't possibly say. The lower you are on the totem pole the more someone is going to wonder if a better programmer and an agent could replace your whole team.

I'm talking with Jason in Slack right now about this topic and what we're really wondering is what does the next generation look like? Because there's going to be a lot of students faking their way through courses with AI and graduating into the work force with impressive portfolios they couldn't really explain when asked about.
 
Upvote
66 (68 / -2)

TylerH

Ars Praefectus
4,957
Subscriptor
I don't personally know any programmers who don't use AI at this point. Even at shops that forbid normal AI (like Disney) they have local agents.

It feels like this is a genie that's not going to be stuffed back into the bottle.
Hi, it's me. I'm the programmer. It's me.
 
Upvote
74 (77 / -3)

TylerH

Ars Praefectus
4,957
Subscriptor
I'm getting the feeling that these frequent announcements from OpenAI have more to do with keeping the hype up more than anything else.

This is a situation where tech reporting falls far short. A lot of these things should be advertisements not articles - they should be paying you to promote their product (or maybe they are and you're just not telling us the reader?)
Oh, obviously. As much as Sam Altman may despise Elon Musk, he's clearly seen how successful Musk's (and Trump's, to an extent) constant over-hype train has led to massive financial success for his companies. Tesla is a meme/hype stock; it doesn't matter how poorly he runs that company. Altman is hoping he can achieve the same reality for OpenAI.
 
Upvote
43 (44 / -1)
D

Deleted member 221201

Guest
I mean, I hope that for you, and all my friends.

I think job security in programming is probably not going to be great going forward, but how that plays out in practice I couldn't possibly say. The lower you are on the totem pole the more someone is going to wonder if a better programmer and an agent could replace your whole team.

I'm talking with Jason in Slack right now about this topic and what we're really wondering is what does the next generation look like? Because there's going to be a lot of students faking their way through courses with AI and graduating into the work force with impressive portfolios they couldn't really explain when asked about.
There are also a lot of students who dont understand memory allocation or have touched c or asm and will look like a deer in headlights

Jon Stokes excellent book "Inside the Machine" should be a required reading at this point.
 
Upvote
29 (30 / -1)
I mean, I hope that for you, and all my friends.

I think job security in programming is probably not going to be great going forward, but how that plays out in practice I couldn't possibly say. The lower you are on the totem pole the more someone is going to wonder if a better programmer and an agent could replace your whole team.

I'm talking with Jason in Slack right now about this topic and what we're really wondering is what does the next generation look like? Because there's going to be a lot of students faking their way through courses with AI and graduating into the work force with impressive portfolios they couldn't really explain when asked about.
I do certainly agree that the future looks uncertain especially for the young’uns as you say! I’m old enough that it doesn’t matter so much… I may become the equivalent of a COBOL programmer in a few years… get laid off, retire for a couple of years, then return or go elsewhere as a consultant at exorbitant rates, fixing slop. I foresee that it will be a target-rich environment…
 
Upvote
43 (44 / -1)

Fatesrider

Ars Legatus Legionis
25,111
Subscriptor
So… the enormous sums of money needed to sustain LLM development and AI will come from replacing programmers? We should prepare for a tidal wave of buggy software.
I'd predict the rise in cases of suicide on the part of highly over-worked, over-stressed and underpaid QC personnel first. Once THEY'RE replaced by AI, then, yeah, the buggy software will be everywhere.

Same timeline, a few months later.
 
Upvote
-6 (5 / -11)