Comment by testerius | Hacker News Reader

Comment by testerius | Hacker News Reader

Parent Full thread

I do not get it. I know that there are different people but... Do you really have that many issues, problems with using AI for coding and solving problems in your jobs or hobbies? In my company using Codex and Claude help us a lot with our software. We were able to fix legacy software which we could not do earlier without AI because of current tasks going on. We migrated in 2 weeks* (read below) to new versions of framework and libraries. If we have to do it without AI, then it will be work for few months or even year - only if doing that work, not other work such as fixing bugs, maintaing systems and doing current work such as new features. In our case AI tools help us a lot - every single developer.

Notice: first working prototype was done after approximately 2 weeks and was done by one developer. Of course it had bugs and was not perfect but it was huge improvement compared to what was before. Imagine that you do not regularly update framework/libraries (as it should be) because of various reasons (client does not want to pay for it or you focus on different things). We understand our decisions and so on, we know it but real life sometimes can not be perfect. Thanks to Codex/Claude we can move forward. Of course we know the LLMs/AI tools are not perfect and magic problem solvers but we could not say that these tools make our lifes worse or something.

The problem with AI is that it works.

Think of it this way, before your company would need to hire more people if it wanted to do more. But now the same team can do things like this without hiring more:

> We migrated in 2 weeks to new versions of framework and libraries. If we have to do it without AI, then it will be work for few months or even year...

Maybe long-term this is amortized. But short-term the tendency is lower salaries and layoffs.

Long term the tendency is even lower salaries.

Right now AI still isn't autonomous enough to do all of the work itself, so it needs developers to carefully manage it to do what it must.

If things remained like this, then sure, there'd be a point when "AI-management for area X" would become a profession on its own, similar to how similar processes happened multiple times in the past, and that would result in a new balance.

But AI won't remain like that. The end-goal is that a high level executive coils type on, Claude v25 something like:

"Update all my legacy codebase to the most recent everything. Make sure it works perfectly and doesn't break compatibility with anything. Also, while you're at it, these DOCX our customers sent list new features they'd like, so implement it all while you're at it."

And come back in the morning with it all executed perfectly. Which definitely isn't a recipe for human developers to start earning any better.

Human earning ceiling will be how much it'll cost for AI to do it at the same quality and efficiency level. If you charge less than that you'll have work, but not otherwise.

Yeah. I hate government intervention as much as the next guy and I would love to live in a distopian reality where humans just enjoy life while robots do the work.

But this is a big problem.

Robots replaced our arms, and we didn't care because we still had our brains. Now they are also increasingly replacing our brains in labour.

you forgot "make no mistakes".

The angst is less about right now and more about the obvious conclusion if the trend continues that few, if any, of us will be gainfully employed in this space.

Sure, we can all claim to be the ones who actually are able to hold it right, or that we'll shift to doing something slightly a bit different. And that will be true for some subset of people. But the reality is that it's like that old trope of look to your left, look to your right, 2 of you won't be here at the end of the semester.

there are categories of tasks that were massively enabled through AI.

there are other parts where its not unless youre playing fast and loose.

Most devs against AI argue that the former tasks were never that big of an issue, but they likely worked at enterprises where those tasks were someone elses job through dedicated teams preparing framework upgrades etc

There is also a gigantic delta between using AI on eg Github copilot through API Pricing and having a Subscription model via Anthropic or OpenAI. Through Github you'll be constantly be worrying about costs, with the Sub youll likely mostly be using SOTA and the quality diffence is staggering, especially since the last model iteration (Opus 5.5)

Reply on news.ycombinator.com