CloudBees CEO says customers are slowing down on 'black box' code from AIs

Anuj Kapur, CEO of DevOps darling CloudBees, reckons that AI could retest the founding assumptions of DevOps as a whole, but warns against the risk of creating black-boxed code in the pursuit of greater efficiency. He also says that some customers who rushed into AI-generated code for fear of missing out (FOMO) are starting to …

[tiggity]

Not that I am in favour of "AI" coding (bar as an aid on repetitive easy stuff)

But if they are worried about some of the things they mentioned e.g. code coverage / unit tests, there are plenty of tools you can add to your pipelines to fail a PR if unit tests do not cover the changed code

If your devops processes include such things then problem code hits a brick wall until the issues are rectified.

[Doctor Syntax]

"the founding assumptions of DevOps as a whole"

There were founding assumptions? Other than vendors being able to sell a whole lot of stuff, training and consultancy, of course.

[Anonymous Coward]

Second order effects of AI?

I know one 2nd order effect:

Future AI will be trained on the output of current day AI.

We know what happens when AI is trained on AI output. The new output will be worse than the previous output, getting worse every generation.

As too much of the current code is already produced by AI, the future collapse of AI is now.

[Ace2]

“… the quality that comes out at the back end…”

Couldn’t have put it better myself.

[ecofeco]

Well played. Have my upvote.

[Don Jefe]

The black box nature of LLMs is their growth engine. Once you start dissecting the box, even through the relational forensics in use now, it’s like reading the pamphlet that comes with prescription drugs. It takes the marketing shine off and makes the whole thing less appealing.

[drankinatty]

Did you ever notice...

People that have very little to say like to misapply technical terms to sound important? "the velocity that they need to prosecute this trend at"? Really, "velocity"? And just how do you calculate the dot, cross or triple product of that "velocity"? Use "speed" or "pace" when you mean speed or pace. (</language nerd rant>)

I guess when what you are saying is "we've figured out that creating systems based on AI generated code can lead to significant reliability problems later on because nobody really knows HTF the code is put together" ... you have to come up with a whole lot of words to obscure what was readily available to most early on. But hey, you're drawing CEO pay, so you gotta make it sound complicated.

[LordHighFixer]

Lets refactor that.

I use AI often when writing code, simple basic stuff. But I am often having to say things like "well, that would be great, if that command existed in this language", or "Lets simplify that 64 lines of case statements in to a while loop with an array", and the like. AI is very helpful in laying down you first few pages of code, but you will actually need to know what you are doing to make "good" code.

[CowHorseFrog]

Re: Lets refactor that.

Isnt writing a monolithic function or method thats pages long wrong ?

Seems to me, you dont know good code when you see it.

[Anonymous Coward]

"but what happens when too many things get broken?"

If it's customer or client trust that is broken no amount of super glue or locktite is going to mend that nor all the king's horse and all the king's men.

I don't suppose there is a lot of trust in the software industry, or the IT industry generally, to begin with; but losing that little will only make rebuilding so much more difficult.

[AnAnonymousCanuck]

How Do You Teach People to Review Code?

If all the code is written by LLM's?

YMMV

AAC