Comment by handoflixue
10 days ago
In terms of startups, predicting tech, and all the things Hacker News is about, it mostly matters what the clever hacker can do, not whether the tool is ready for the mainstream.
If a clever hacker can get 10x results with an LLM, they're gonna outcompete the 90% that can't figure out how to replicate that result, and they'll be able to get about as much work done without that 90%
Factories, Agriculture, etc. - this is hardly the first time that pattern has played out.
So, I'm using AI both at work and for a personal greenfield project, and I have 20 years of pre-AI software development. Which isn't to say I'm amazing, just putting some context here of what my experience level is and the contexts I've used this tech in. First off, I doubt the "10x" number in general (I'm not seeing it personally or from other people), but lets say I pretend it's true for a second. 10x better/faster at what exactly? Like, if you have an unreleased greenfield app then sure, you can bang together features a lot faster. But what if you have an established app with real users? It's not just the time it takes to make the feature, you also have to consider documenting the feature, making sure it works well with the other features, making sure it's something customers actually want, making sure it's part of a larger coherent design, training customers, marketing the feature, etc. Like this notion of "we'll just go 10x faster!" falls apart really quickly when you're talking about making something that people will actually depend on and use.
I keep thinking about that (so far anonymous) company that blew 500 million dollars worth of tokens in a month, and what I desperately want to know is WHAT DID THEY BUILD WITH THAT?! Like, for that sort of money they should have created an earth shaking new business or something instead of becoming a cautionary tale that's rightfully too embarrassed to publicly own it.
The other thing with regard to factories/agriculture/whatever is, in revolutions with those things nobody needed to be convinced. Sure, people were (rightly!) concerned about the societal impact, but the utility of a factory was fairly obvious. And yet last year OpenAI spent more on their marketing budget than Coca Cola. The way this stuff is hyped and pushed has an air of extreme desperation. If it was so good people wouldn't need so much convincing!
> But what if you have an established app with real users?
I recently used GPT 5.5 "extra high" running nearly non-stop for about a week to upgrade a legacy ASP.NET Web Forms app to ASP.NET Core on .NET 10.
This was considered "too hard" (too expensive) for human developers because it is a wholesale rewrite of every web page template. Not to mention that the dependency injection mechanism is totally different, async is more pervasive, etc.
Worse still, the old app was split into a bunch of components with a variety of web API protocols in between them, had stupidly complex Oracle stored procedures, and a whole series of hidden land mines in the codebase. It was an undocumented, unmaintainable mess full of dead and spaghetti code.
I would have estimated 6 months minimum for a human developer to uplift it, but 12 months is more realistic.
Doing the same in a week feels like tapping into some sort of forbidden black magic. It feels like I can't admit this to anyone, lest they think I dabble in the dark arts.
I don't work in coding, but I do a lot of complex tasks that can be automated to some extent. In 2019 I spent more than a month painstakingly building an autohotkey script that would interact with a design app to build a Chinese language workbook with proper formatting, and create indexes. When the script was finally running on its own it felt like magic. Nowadays I use a mix of Claude code /codex/antigravity (I have the 20 Usd sub for each) to build very specific "one use" tools that save me countless hours. I can even be very specific about how to design those scaffolds so the flow just feels intuitive for me. It's insane. It feels like a cheatcode. I think the best use for Ai in a company is to build tools for the humans, not to replace those humans
1 reply →
> The other thing with regard to factories/agriculture/whatever is, in revolutions with those things nobody needed to be convinced
Were you not around for "the internet is for nerds" and "why would I ever need to learn how to use a computer, I work in an office"?
Are you familiar with the word "sabotage", originating from people who were trying to stop factories from taking over the world?
It's part of why I find the anti-LLM pushback so bizarre - it used to be that people using computers and the internet were exactly the ones trying to convince others "hey, this is actually really cool". A decent chunk of us have seen this play out once or twice before.
This is exactly the crowd that should remember how much convincing everybody needed, and how hard it was to get everyone to take things seriously!
I was there for the early days of the internet, but honestly nobody needed to tell me the internet was cool. I tried it once and it was very obviously something I wanted, even with a shitty 28k modem and waiting hours on 2k/s downloads.
AI is useful for some things, but to me it's not "internet" level of useful, especially since in a lot of ways it's just a weird probabilistic wrapper on all the knowledge that was already available for free on the internet anyway. Actually, yesterday I was trying to lookup an API function in a game engine and I had google hallucinate the function call twice in a row, and all I could think was "I really just want good search again".
A lot of what I see happening with coding agents isn't actually that different from how people were (badly) building web-apps in the past -- grab a fuckton of stuff off NPM and copy paste from stack overflow (RIP) with a bit of glue code. Now the agent does that for you, but it's not like a lot of thought or skill went into that style of development in the first place. I feel like the main people saying that coding is dead are the ones that weren't doing it very well in the first place. Even in my personal project where I let AI off the leash a lot more than I would in a professional job.. almost all the features that I managed to bust out quickly were more to do with a rich open source ecosystem than Claude's glue code. (IE, I didn't need to invent a text editor because code mirror is already good, I didn't need to write a bunch of frontend components because there are plenty of good UI libraries, I didn't need to write a fancy graph editor because react-flow is fantastic -- 90% of my leverage is a good ecosystem, not AI.
Anyway, as I suppose a bit of a hater, I don't actually hate LLMs themselves, I hate everything around it -- the annoying grifty hustle culture, the incessant hype, the forced usage, the slop apocalypse, the fact that it's about to set the economy on fire and most of all the intent to degrade peoples skills and intellect in an awful race to outsource our thinking to things that don't think. It's not that there isn't interesting tech in there, it's that it's just that when you consider its effects the positives don't outweigh the negatives.
> It's not just the time it takes to make the feature, you also have to consider documenting the feature, making sure it works well with the other features, making sure it's something customers actually want, making sure it's part of a larger coherent design, training customers, marketing the feature, etc.
Seems like a conflation of responsibilities. An IC that implements a feature is rarely the same person who decided on the feature to build in the first place. This is really only true in exceptionally small teams where everyone wears multiple hats. If a programmer is also doing marketing, they're basically a sole proprietor. Since most people work with others who make product and marketing decisions, I don't feel like this line of thinking is relevant.
Consider programming with a dumb text editor (no syntax highlighting, no integrated build/test/run cycle, no symbol/reference tracking, no auto-complete) and an IDE. The magnitude in performance or efficiency between that and the difference between programming with an IDE and programming with AI is relatively similar. Sure, the AI can do more than the IDE in absolute terms, but it's not going to be training customers or doing your architectural planning. (At least, I sincerely hope not -- not yet, anyway.)
That's my perspective on the "10x myth". AI isn't 10x, just as dumb editor to IDE wasn't 10x. It's only a modest improvement.
> The way this stuff is hyped and pushed has an air of extreme desperation. If it was so good people wouldn't need so much convincing!
You can ignore the hype. I ignore every advertisement I reasonably can. It doesn't matter who's peddling, it's always toxic.
But more than that, people do need convincing. Entrenchment, momentum, and bureaucracy make it nearly impossible to change anything. That, in turn, leads to the well-known adoption curve. You're using AI, you're an early adopter. The marketing and hype are there to chip away at the friction keeping the adoption curve from an early majority.
> I keep thinking about that (so far anonymous) company that blew 500 million dollars worth of tokens in a month
First I've heard of this. Do you know any more or could you maybe point me to a source? Kind of astounding if true! Agree that it would be very interesting to know what kind of output they got for all that effort.
https://finance.yahoo.com/sectors/technology/articles/compan...
no it's not true. The fact that people can even believe this is.... wild. It's a bunch of circular articles jerking eachother off about some anon company.
using $15 / mtok (Clause output using 4.6) as the metric, it would take 33.33T Tokens for you to spend $500m.. . Or enough output for like 250m books, or a developer uses 10 million tokens a day doing non-stop agentic programming, it would take that single developer 9,132 years of continuous coding to run up that bill.
1 reply →