Hacker News Reader

Comment by zusammen | Hacker News Reader

The irony is that AI, as a replacement for an even average programmer, doesn’t work. You can use it helpfully if you are already a programmer, but if you’re not, you won’t know when it’s generating insecure or even non-working code.

And yet it’s causing massive disruption. The bosses have such a hardon for layoffs that they’re doing this on the slim chance that it might work for them. Having to hire programmers back in a panic is, as they see it, future someone else’s problem.

View on HN

I'm oversimplifying it horrendously with dumb math, but this is how I think of it:

Let's say you're trying to build a product/feature. You have general timelines for the launch. About 5 years ago, it would take a team of 10 to deliver it. Now, let's assume there were productivity gains, and you can deliver it with a team of 8.

This would not necessarily be a problem if for 20% of the workforce cut, you would also start getting investments and create 20% more companies/features. The big uncertainty is, will this ever happen, or it'll end up with 10% increment in such activities, so you still have that extra person hanging around.

That's absolutely true, but it doesn't mean AI won't lead to job losses just in terms of increased productivity: if your existing programmers can get more done you may find yourself reducing (or at least freezing) your team size.

I continue to hope that the net result of programmers getting more productive is an increase in demand for our skills, as employers realize they can get a whole lot more value for their investment. It might take a while for that to shake out though - we need companies to realize they can take on more ambitious projects now.

100% C suite executives are not going to sit in front of a chat agent to fix a bug in an application or solve a problem.

Managers would do what they always do. Give somebody requirements. Just in this case it's "something"

Managers give very loosey-goosey, easy to misinterpret requirements.

Which the AI will just make a stab at without digging in and asking clarifying questions or examining all aspects of the codebase to spot what challenges there are with doing it that way with the specific custom business logic needs needed for the rest of the application, and then get it very wrong and not anywhere close to what was intended (although possibly exactly what was actually asked for).

Meanwhile even a slightly seasoned developer will know to ask for clarifying questions on various aspects of it, or point out challenges with the existing codebase to match the requested requirements (sometimes after taking a stab at it themselves and realizing it doesn't work).

There are ways to prompt the AI to produce better results, with very specific and detailed requirements that include tricky aspects of the technology (that developers are aware of and can provide when they write prompts), but the typical requirements that come down from management ain't anywhere near that.

Not saying AI won't ever get to that point, but we're not there yet.

Being helpful to 2/3 might be enough to lay off 1/3

The only change this quarter is the use of AI to justify this behavior :)

> The irony is that AI, as a replacement for an even average programmer, doesn’t work.

Yet.

AI needs to fundamentally change before this happens. We're not even close to capturing actual reasoning in AI models.

Disagree. I've been coding with o1 and o3-mini-high

If you give it a prompt which resembles a requirements document produced by a competent manager, you will get what you request and it will run first time.

And how many C suite executives will be doing this?

I think you overestimate how many people could produce this document to the degree you're talking about

Doesn't matter how many people can produce the requirements document when an AI can produce it.

No manager and coder pair can iterate on an idea 5 times per minute, even if the first attempt needs correction, the timescale disparity makes AI agents appealing for nearly all noncritical coding tasks.

Now you're talking about someone who can describe what they want in enough so that an AI can generate a strictly defined requirements document with enough detail that another AI (or the same AI) can generate working code?

It's just AI all the way down, and each step still needs to be validated because AI is not reliable

lol. The hardest part in corp programming is getting the requirements right when literally no one involved can seem to think themselves out of a paper bag.

It isn’t a matter of iterating either, it’s being able to figure out what is politically workable and actually implementable.

What about examples where [technical] management actively makes it harder for engineers to do their job by enforcing nonsensical design decisions (when explicitly warned against it)? :)

No AI can solve that one.

Not sure I agree here completely. On its own perhaps not, but if AI is able to deliver nonsensical design decisions at a faster rate than a human, you can iterate to a final better design faster.

What language?

I used to pick new languages when I started projects with AI for learning but lately I've been using ruby for everything possible and I generally prefer it's output as it writes stuff more idiomatically than I do (out of laziness)

Ah, I was curious if o1 is running code in background and doing an error loop.

Do you have a chat example you can share where you found it to work great? I feel that the performance varies a lot between domains.

Maybe. But if one founder and a couple of good devs can handle the requirements and management, and handhold the agents when needed to make a product 80% as good as a competitor with 200 employees, it's still gg for that competitor.

It's irrelevant anyway, because AI companies are not HR firms. They don't get rich by lowering the cost of labor, nobody in tech does. They get rich by implementing the IBM/Oracle/Salesforce model for enterprise B2B: selling a "platform" with plenty of "smart plugins", which will require plenty of humans to sit in on meetings, write technical requirements, do QA and the usual stuff that pushes IT costs way past budget.

Reply on news.ycombinator.com