> Tan/AI built the website so that when a user visits, their browser makes 169 server requests for various assets totaling 6.42 megabytes in size. For comparison, the minimalist Hacker News homepage (also run by Y Combinator) makes seven requests for data totaling just 12 kilobytes.
Lisp failed in the previous AI winter because it was so damn efficient, when the C-suite just likes big stupid numbers. Tell Garry Tan you have a tool that generates 2 million lines of code per day and he'll probably give you a billion dollars on the spot.
(I hope one day he sets out to vibecode Hacker News 2.0 and he'll have to convince our mods to switch over. Oh, that would be so funny.)
I don’t understand what he’s actually building with 37k loc/day. In a month that’s 200k loc per project. I tried to go look at his blog but it didn’t load on my phone.
If you're looking at the code at all, or even the project itself, then you're doing something wrong. AI is smarter than you and agentic teams are more productive than you.
Who are you to question what they build? You merely supply the prompts. /s
About as bad as I expected. AI can write code I don't feel like - but I still review and understand every line like it is a proven untrustworthy coworker (most are proven trustworthly and get a lesser review). I have found too much outright bugs, race conditions, changing standard required constants and the like.
I still don't understand this trend of measuring productivity by lines of code shipped. Goodhart's law can apply to any metric but this is an exceptional bad and gameable one.
I dont get all this obsession over LOC. Having a crystal clear vision is the most important thing. If you've got that with conviction, shipping is trivial and not limited by people - you can go raise the money and hire the people to do it.
LOC is not the measure of productivity people think it is. Getting closer to actualising a vision is the true goal.
Im convinced there's two kinds of innovators
1) people who throw stuff at the wall and see what sticks
2) a much much smaller group of people who are very deliberate, methodical and think strategically before investing a dime in producing
So we’ve finally reached the typing monkeys - since one line in a million might be Shakespeare.
I went all in on Claude for a week to see what would happen and ended up being unimpressed with results and saw it massively over complicating simple issues or creating unnecessary liability.
All these people are doing is pumping out tons of tech debt and liability that they don’t even have time to digest or understand and for what? Where is the upside or the profit, is it just the reward of seeing line go up in terms of “productivity”.
> Tan/AI built the website so that when a user visits, their browser makes 169 server requests for various assets totaling 6.42 megabytes in size. For comparison, the minimalist Hacker News homepage (also run by Y Combinator) makes seven requests for data totaling just 12 kilobytes.
Lisp failed in the previous AI winter because it was so damn efficient, when the C-suite just likes big stupid numbers. Tell Garry Tan you have a tool that generates 2 million lines of code per day and he'll probably give you a billion dollars on the spot.
(I hope one day he sets out to vibecode Hacker News 2.0 and he'll have to convince our mods to switch over. Oh, that would be so funny.)
I don’t understand what he’s actually building with 37k loc/day. In a month that’s 200k loc per project. I tried to go look at his blog but it didn’t load on my phone.
> I don’t understand what he’s actually building with 37k loc/day. ... I tried to go look at his blog but it didn’t load on my phone.
That 37k loc/day must be JS in his blog. /i
Neither does he, that's the beauty of it.
If you're looking at the code at all, or even the project itself, then you're doing something wrong. AI is smarter than you and agentic teams are more productive than you.
Who are you to question what they build? You merely supply the prompts. /s
About as bad as I expected. AI can write code I don't feel like - but I still review and understand every line like it is a proven untrustworthy coworker (most are proven trustworthly and get a lesser review). I have found too much outright bugs, race conditions, changing standard required constants and the like.
I still don't understand this trend of measuring productivity by lines of code shipped. Goodhart's law can apply to any metric but this is an exceptional bad and gameable one.
It's part of the AI agent mania, making big number go up fuels the mania even more.
Just once I wish they would talk about the product they actually shipped
They will... if they ever do ship.
Haha.
I dont get all this obsession over LOC. Having a crystal clear vision is the most important thing. If you've got that with conviction, shipping is trivial and not limited by people - you can go raise the money and hire the people to do it.
LOC is not the measure of productivity people think it is. Getting closer to actualising a vision is the true goal.
Im convinced there's two kinds of innovators 1) people who throw stuff at the wall and see what sticks 2) a much much smaller group of people who are very deliberate, methodical and think strategically before investing a dime in producing
[dead]
I wonder if there is a universal limit for speed per lines of code. The one the we think the light has
I get it. This is an 10x ai developer. It's the additional line count to get one good line of code.
10x more code but 10/11 of all code is now garbage.
Would like to know how much it costs per day.
I ship 10x that per hour. 369,999 of that are empty lines, then a slick one-liner :)
But HN is still partying like it's 1999...
Couldn't he offer a 1/4 day for a HN revamp?
>argued that Gregorein belongs to a fading era of software development, when humans still checked code, line by line, before shipping.
I guess people are going to hurt themselves badly before they realize this is still a requirement.
[dead]
So we’ve finally reached the typing monkeys - since one line in a million might be Shakespeare.
I went all in on Claude for a week to see what would happen and ended up being unimpressed with results and saw it massively over complicating simple issues or creating unnecessary liability.
All these people are doing is pumping out tons of tech debt and liability that they don’t even have time to digest or understand and for what? Where is the upside or the profit, is it just the reward of seeing line go up in terms of “productivity”.
Total madness to me.
[flagged]