[jspdown]

At this stage, AI is no longer a tool that enhances your ability to ship code, it has replaced you entirely in that role. You don't control what is shipped, and you can't verify if it's correct. That's a serious problem! As software engineers, we remain accountable for code we no longer fully understand.

Then, what comes next feels less like a new software practice and more like a new religion, where trust has to replaces understanding, and the code is no longer ours to question.

[misja111]

Speak for yourself, I don't ship any code that I don't fully understand. Yes that requires less autonomous AI and less frequent merging. But I don't even want to think about the disasters that could happen if you really get into the habit of shipping code you can't verify or understand.

[voidUpdate]

Out of interest, what's the speedup between having an LLM write code for you and then having to go through and understand it, vs writing code that you understand immediately because you wrote it?

[misja111]

It depends. If all I want is some prototype or pet code project, my LLM can write most by itself. The speedup could be 10 times or more. However, if I'd let a LLM write code for my work, I'd have to very thoroughly review it and most likely ask it to rewrite it several times. Each time this would require a new review of course. There would still be a speed up but I guess at most somewhere around 25%.

In practice I try to combine the best of both worlds. I write some code by myself and rely on my LLM for parts that are not too big and where I expect it to do a pretty good job.

[miningape]

Not OP but I hold myself to that standard, and the honest answer is that at best it's the same.

[kolinko]

Or formal methods and other tools for verifying the code security?