[freediddy]

> AI is probabilistic and any probabilistic system can inherently never be perfect as it also still makes numerous mistakes. Work still needs to be reviewed by humans.

I don't know why people keep saying this. I don't believe code needs to be reviewed by humans. Are humans infallible? I think 80%+ of code is basically human slop. Do we review the assembly language that is generated by compilers?

I think given where AI-generated code is, and where it's going, all we need to do is test the outputs, basically turning us into QA. As someone who loves to code, yes I find it distasteful, but we don't get paid for our code, we get paid for our programs. I prefer being paid.

If you want to write beautiful code, you can do it as a side hustle at home, but I think very quickly, this is what our industry is turning into. In January, someone told me that an expert told him that software engineering would be dead in 3 months, and I scoffed at the time, but now I think it's pretty accurate. It definitely won't survive this year.

[SpicyLemonZest]

To me it's not a theoretical issue. I have personally caught multiple issues in code review this month that would have caused an outage had I allowed the PR to be merged. The agent that generated the PR made plausible-sounding but incorrect assertions about what the output should be. So if I don't want production outages I have to do code review.

I've seen some people suggest that I should be require agents to produce an extremely detailed PR description and review that instead, which perhaps would have worked. But I don't understand the appeal. Why would I want to review a lossy, imprecise description of what the code does instead of reviewing the actual code and what it actually does?

[100percentjake]

Compilers are deterministic, though. If you don't put garbage in, you won't get garbage out. With a probabilistic system you can have varying outputs for the same input, thus the need to spend (waste?) effort verifying that your output is quality.

No comment on the rest of your comment. The assembly piece just bothered me in particular.

[freediddy]

There are still bugs in the compiler as well as in optimization that can be created. The only thing stopping bugs in production are unit tests. That's exactly the same as with AI-generated code as well.

[olalonde]

The top AI labs still employ thousands of software engineers. If it was dead, they'd be among the first to stop employing them. If by "software engineering is dead" you mean that using AI coding tools is now part of the job description, I think that's largely correct.

[bLorax]

I really wanted to write a well structured answer to your comment but then I saw the username and was like "nah I'm fine"

[yoyohello13]

> In January, someone told me that an expert told him that software engineering would be dead in 3 months, and I scoffed at the time, but now I think it's pretty accurate

I mean, maybe your 3 month prediction is correct. But by definition that expert's 3 month prediction wasn't accurate because software engineering is still around and it's after April. We are 3.5 years in '6months until programming is solved'.

[fedeb95]

wait just three more months and software engineers will be out of job on time.

[archagon]

What we get paid for is entirely socially determined, not some immutable axiom.

And AI is not an abstraction.

[fedeb95]

humans are way better at managing big contexts than a machine. The human mind is way more complex than any AI model + infrastructure.