Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

An industrial piping contractor on Claude Code [video]

twitter.com

Post reply on HN
Print view View original

71–80 of 104 posts

Re: An industrial piping contractor on Claude Code [video]

#71

Post by pron »

Here's what I find difficult to reconcile with my own experience. I've been using Codex in anger for the past 2 months or so (with gpt5.3-codex and then gpt5.4) on projects of different complexity. It is quite good at debugging, but the (non-trivial) code it produces is really bad. And I don't mean bad stylistically, but bad in the sense that Codex clearly won't be able to maintain it for long (which is how Anthropic's C compiler experiment failed) because it uses an approach of "success at all costs" where it always prefers fixes that treat the symptom rather than going back and rethinking the architecture as features are added.

So the options are: 1. the program involved here is really trivial, 2. it hasn't evolved long enough for the agent to fail at evolution, or 3. others are not seeing what I'm seeing.

Re: An industrial piping contractor on Claude Code [video]

#72

Post by pron »

w10-1 wrote: post #70

Software development usability has always been measured in mean time before black-box failure: something misbehaves that the person can't fix or understand. LLM's shorten that time for every application and every user, but particularly for users from professions that haven't built modeling or debugging skills because they rely on physical reality - like pipes fitting or process supervision - to weed out non-performer…

Except I find that agents are much better at debugging than writing code. When the code gets to the state where the agents can no longer debug it and keep it running (as happened in Anthropic's failed C compiler attempt) humans will likely not be able to save the situation (in an efficient way). And without careful human supervision from the start, the code ultimately gets to that place.

Re: An industrial piping contractor on Claude Code [video]

#73

Post by pron »

steve_adams_86 wrote: post #3

I've more or less accepted this, and I think my future is in making software more resilient, secure, and fault tolerant. These people will likely want to scale these solutions up, tie different solutions together, and generally make their lives increasingly difficult. Often without realizing it. My experience is that Claude starts to make quite a mess in this context, and it'll often cause as many problems as it solv…

You assume that the agents can produce okayish software that isn't very resilient, but that's not quite what we're seeing. Remember Anthropic's attempt to have agents write a C compiler (which isn't a super-complicated task by any means). Despite preparation work that is well beyond what most software can have (specs available to the agent and that the model trained on, thousands of human-written tests available to the agent and that the model trained on, and a human-written reference implementation available to the agent and that the model trained on) the agents still failed to converge. In other words, under conditions that are so favourable as to be unrealistic for almost any kind of real software, the AI agents couldn't even produce so much a workable C compiler. And this is what I also see: the agents take the code in a direction that doesn't converge, let alone make it resilient.

Re: An industrial piping contractor on Claude Code [video]

#74

Post by BobbyJo »

pron wrote: post #71

Here's what I find difficult to reconcile with my own experience. I've been using Codex in anger for the past 2 months or so (with gpt5.3-codex and then gpt5.4) on projects of different complexity. It is quite good at debugging, but the (non-trivial) code it produces is really bad. And I don't mean bad stylistically, but bad in the sense that Codex clearly won't be able to maintain it for long (which is how Anthropic…

This tracks with my own experience. I have to watch what the LLM does in response to any given ask, especially when adding new features to an existing system, because it will do things a different way every time until it no longer knows what patterns to follow.

Re: An industrial piping contractor on Claude Code [video]

#75

Post by pron »

robotswantdata wrote: post #18
steve_adams_86 wrote: post #3

I've more or less accepted this, and I think my future is in making software more resilient, secure, and fault tolerant. These people will likely want to scale these solutions up, tie different solutions together, and generally make their lives increasingly difficult. Often without realizing it. My experience is that Claude starts to make quite a mess in this context, and it'll often cause as many problems as it solv…

Are we sure Claude Scale™ won’t appear next month? A specialist agent that turns your vibe coded mess into a production grade scaled solution on their infrastructure. Expect anthropic to want to capture more of the supply chain over time

If they could they would, and if they can they will. Maybe it will appear next month, and maybe 5 years from now, and we don't know which of these is more likely. But I think that if agents could actually produce good, reliable software than can evolve over time, there's little they couldn't do even beyond software. So it won't be (just) the software developers being replaced, but also software users.

Re: An industrial piping contractor on Claude Code [video]

#76

Post by georgemcbay »

jvanderbot wrote: post #65

As part of my "well shit what's next" arc I've been checking out machine shops in my area. By and large they are dumb manual 30 year contracts or highly automated job shops.. And the owners of those job shops aim for 3 shifts per worker via automation, and mash their own software with AI already. They are ruthless at cost cutting and automation and AI tools are perfect for them. Unfortunately the vibe I get talking t…

The general approval the average person has for tech/software workers has declined precipitously over the last decade (IMO for some pretty good reasons, as an overall industry we've done a lot of societal harm) which just adds to how cosmically funny it is that as a group we are likely to experience one of the quickest mass career disruptions from AI.

Re: An industrial piping contractor on Claude Code [video]

#77

Post by Beretta_Vexee »

For decades, employees have been developing tools using whatever was available to them, and in most cases, this was limited to Excel macros.

AI provides access to much better tools for testing and quickly experimenting with new ideas.

The only ones who should be worried are companies that charge millions for four junior developers and an agile coach, and deliver more PowerPoints than code (I’m looking at you, Capgemini).

Re: An industrial piping contractor on Claude Code [video]

#78

Post by zeckalpha »

xvedejas wrote: post #58
williamcotton wrote: post #57

Earlier quoted context omitted.

”The current market share of custom-built homes is approximately 19% of total single-family starts” https://www.nahb.org/blog/2025/08/custom-home-building-grows...

New multifamily construction in the US that has to undergo design review is arguably fairly custom in that each site will have different requirements. I think it's fair to say that commoditization is a spectrum?

tract != multifamily

Re: An industrial piping contractor on Claude Code [video]

#79

Post by wild_egg »

edg5000 wrote: post #9
steve_adams_86 wrote: post #3

I've more or less accepted this, and I think my future is in making software more resilient, secure, and fault tolerant. These people will likely want to scale these solutions up, tie different solutions together, and generally make their lives increasingly difficult. Often without realizing it. My experience is that Claude starts to make quite a mess in this context, and it'll often cause as many problems as it solv…

What could happen is a reduction in the amount of programs used, with a smaller set of more sophisticated programs doing more work. This maps to what we saw pre-industrial revolution: lots of small family operations doing menial manufacturing work (woodworking, textiles, cooking). This got replaced by large factories. A smaller amount of companies producing the a larger volume of goods. With AI, a smaller group of en…

Factories benefit from economies of scale that favour centralization.

I think smaller groups handling more complexity is on point. But that's because each group will build their own bespoke factory catered to their exact needs.

I very fully expect a mass proliferation of custom programs rather than standardizing on a common set that groans under the weight of being so general to support all use cases.

Re: An industrial piping contractor on Claude Code [video]

#80

Post by PepperdineG »

mathisfun123 wrote: post #33
coldtea wrote: post #32

Earlier quoted context omitted.

> thanks to a new lower barrier provided by LLM. New lower barrier means commodification.

Lol no it doesn't you literally have it backwards - think about the trades, specifically construction, as low barrier to entry jobs and consider that houses/buildings are all different (not commodities).

Both can be true in homes being unique while also functionally being like commodities. Whether a home or a spec home or tract home the pricing is based similarly to sports stats and no matter how one-of-a-kind a home is a mortgage on such a home can then be packaged with a bunch of other homes into a bond where bond investors will look at the stats of the combined homes and who the borrowers are.
Post reply on HN