Comment by conartist6 | Hacker News Reader

Comment by conartist6 | Hacker News Reader

Parent Full thread

Right but now you're talking about 5 or 20 or 100 or 1000 companies building CRM software. They're basically doing the mostly the same work over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and over and (I would like you to know that I typed every single one of these "and over"s with my very own fingers) and over and over and over and over and over and over and over and over and over and over and I think only the AI companies really benefit from that.

I feel silly explaining this as if it's a new thing, but there's a concept in social organization called "specialization" in which societies advance because some people decide to focus on growing food while some people focus on defending against threats and other people focus on building better tools, etc. A society which has a rich social contract which facilitates a high degree of specialization is usually more virile than a subsistence economy in which every individual has all the responsibilities: food gathering, defense, toolmaking, and more.

I wonder if people are forgetting this when they herald the arrival of a new era in which everyone is the maker of their own tools...

The thing is that also empowering individuals to do specialized activities by way of a tool (instead of themselves having to specialize) is a hallmark of progress? Like I don’t need a “professional” to wash my clothes, I don’t need to wash my clothes myself. I use a washing machine.

I don’t need to hire a programmer. I don’t need to be a programmer. I can use a tool to program for me.

(We sure as hell aren’t there yet, but that’s a possibility).

> (We sure as hell aren’t there yet, but that’s a possibility)

What makes you think so?

Most of the stuff I've read, my personal experience with the models, and my understanding of how these things work all point to the same conclusion:

AI is great at summarization and classification, but totally unreliable with generation.

That basic unreliablity seems to fundamental to LLMs, I haven't seen much improvement in the big models, and a lot of the researchers I've read are theorizing that we're pretty close maxing out what scaling training and inference will do.

Are you seeing something else?

This seems really vague. What does "totally unreliable" mean?

If you mean that a completely non-technical user can't vibe code a complex app and have it be performant, secure, defect-free, etc, then I agree with you. For now. Maybe for a long time, we'll see.

But right now, today, I'm a professional software engineer with two decades of experience and I use Cursor and Opus to reliably generate code that's on par with the quality of what I can write, at least 10x faster than I can write it. I use it to build new features, explore the codebase, refactor existing features, write documentation, help with server management and devops, debug tricky bugs, etc. It's not perfect, but it's better than most engineers I've worked with in my career. It's like pair programming with a savant who knows everything, some of which is a little out of date, who has intermediate level taste. With a tiny bit of steering, we're an incredibly productive duo.

I know the tech is here to stay, and the best parts of it are where it provides accessibility and tears down barriers to entry.

My work is to make sure that you don't need to reach for AI just because human typing speed is limited.

I love to think in terms of instruments versus assistants: an assistant is unpredictable but easy to use. It tries to guess what you want. An instrument is predictable but relatively harder to use. It has a skill curve and perhaps a skill cap. The purpose of an instrument is to directly amplify the expressive power of its user or player through predictable, delicately calibrated responses.

Maybe we're working on different things?

My experience has been much worse. Random functions with no purpose, awful architecture with no theory of mind, thousands of lines of comprehension debt, bugs that are bizarre and difficult to track down and reason about...

This coupled with the occasional time when it "gets it right".

Those moments make me feel like I saved time, but when I truly critically look at my productivity, I see a net decline overall, and I feel myself getting dumber and losing my ability to come up with creative solutions.

Yeah, I haven't had any of that, especially in the last ~6 months. I'm using Cursor and lately Opus 4.5

I have used Claude to write a lot of code. I am however already a programmer, one with ~25 years of experience. I’ve also lead organizations of 2-200 people.

So while I don’t think the world I described exists today — one where non-programmers, with neither programming nor programmer-management experience, use these tools to build software — I don’t a priori disbelieve its possibility.

Your washing machine can only deal with certain classes of clothing. It will completely destroy others, and has no way to determine what clothing has been put into it. Meanwhile, the average untrained-but-conscientious human will, at worst, damage a small portion of an item of clothing before spotting the problem and acting to mitigate it. (If the clothing is "absolutely must never come into contact with water" levels of dry-clean only, they might still trash the whole item, but they aren't likely to make the same mistake twice.)

Programming is far more the latter kind of task than the former. Data-processing or system control tasks in the "solve ordinary, well-specified problem" category are solved by executing software, not programming.

Using an AI is still like hiring someone to do programming work for you. It's going to cost money. Why would you waste money? We have sewing machines, but you don't make all your own clothes do you?

If the cost of the raw materials and worker were less than the price tag at the store, sure, I would probably opt to make my own clothes. They would fit me perfectly, and I can get the right shade of blue instead of bluish.

In the case of AI, Claude costs $100 or $200/mo for really good coding tasks. This is much less expensive than hiring someone to do the same thing for me.

That sounds like a nice hobby.

Which part is the hobby? Clothes making or using Claude to generate real production code?

Both. I would note that "real production code" is not necessarily a high bar. For example it does not rule out gross negligence. Most of the companies that outsource their thinking and working to Claude will die of it.

I have a different point of view. Claude code is extremely good at creating and maintaining solid, everyday code including Ansible playbooks (used in production), creating custom dev/ops scripts for managing servers (again, used in production), creating Grafana dashboards (again, production), comparing database performance between nodes, etc. Just because a person did not hand-write this code does not make it any less production ready. In fact, Claude reviewed our current Ansible code base and already highlighted a few errors (the files written by hand). Plus, we get the benefit of having Claude write and execute test plans for each version we create. Well worth the $100/mo we pay.

And to your note that real production code is not necessarily a high bar, what is "real production code"? Does it need to be 10,000 lines of complex C/rust code spread across a vast directory structure that requires human-level thinking to be production ready? What about smaller code bases that do one thing really well?

Honestly, I think many coders here on HN dismiss the smaller, more focused projects when in reality they are equally important as the large, "real" production projects. Are these considered non-production because the code was not written by hand?

All it sounds like to me is that Ansible is production-ready, Grafana is production ready, the compilers and runtimes you're using are production-ready.

Each of those things is a mountain of complexity compared to the molehill of writing a single script. If you're standing on top of a molehill on top of a mountain, it's not the molehill that's got your head in the clouds.

God, yes.

I see so many people quote that damnable Heinlein quote about specialization being for insects as if it's some deep insight worth making the cornerstone of your philosophy, when in fact a) it's the opinion of a character in the book, and b) it is hopelessly wrong about how human beings actually became as advanced as we are.

We're much better off taking the Unix philosophy (many small groups of people each getting really really good at doing very niche things, all working together) to build a society. It's probably still flawed, but at least it's aimed in the right direction.

Not only AI companies benefit the companies themselves benefit too from getting software that actually meets (more of) their needs rather than whatever some dev imagined they might need.

That's right. Infact, I see more outsourcing to happen, due to risk delegation and complexity management. AI would only make humans lazier and risk averse. Complexity of regulations, government reach, security risks would only increase. Risk can't distributed to AI employees (agents). A supervisor of AI agent populations can't be held responsible for all the bugs and complexity in a AI-generated product.

Companies have always written their own software (or abused spreadsheets).

The tool is the computer; writing software to solve your own problems is advanced tool use, not making your own tools.

Obviously this doesn't apply to every kind of software, but I would argue that anything someone might be tempted to build in Excel is fair game.

Reply on news.ycombinator.com