This is doing more to keep me away from "vibe coding" than anything else. Look, I'm genuinely interested in using AI as a tool and trying to boost my productivity in any way possible - I equate this to activities from the past like learning shortcuts of my editor, learning to type fast, and so on - but, the almost persuasive nature of this README, just pushes me away.
Not to mention using lines of code as a metric of usability is just _whatever_.
> In the last 60 days I have written over 600,000 lines of production code — 35% tests — and I am doing 10,000 to 20,000 usable lines of code per day as a part-time part of my day while doing all my duties as CEO of YC.
LOC will never be a good metric of software engineering. Why do we keep accepting this?
I can generate 1 million LOC if I really wanted to.
As long as LOC is the main metric for these setups, they will never be successful.
LOC is a very-very weak proxy of "how many new features" I've built, and they don't have any other metric that can be measured easily. But it causes serious issues, because equating LOC with productivity leads to inevitable utter bloat, that no agent or human can ever rectify in a meaningful timeframe. I'm pretty sure this 600 000??? LOC could be shrinked to 60 K for the same feature set, but with better readability and performance.
I have written 600 THOUSAND lines of production code. The best and most beautiful production code. The agents negotiate. They want to make a DEAL. I am the best deal maker in the world. Thank you for your attention to this matter! -- GJT
I've been using gstack for the last few days, and will probably keep it in my skill toolkit. There's a lot of things I like. It maps closely to skills I've made for myself.
First, I appreciate how he implemented auto-update. Not sure if that pattern is original, but I've been solving it in a different-but-worse way for a similar project. NOT a fan of how it's being used to present articles on Garry's List. I like the site, but that's a totally different lane.
The skills are great for upleveling plans. Claude in particular has a way of generating plans with huge blind spots. I've learned to pay close attention to plans to avoid getting burned, and the plan skills do a fair job at helping catch gaps so I don't have to ralph-wiggum later. I don't find the CEO skill terribly effective, but I do like the role it plays at finding delighters for features. This is also where I think my original prompting tends to be strong, which could be why it doesn't appear to have a huge impact like the other skills.
I think the design skills are great and I like the direction they're going. DESIGN.md needs to become a standard practice. I think it's done a great job at helping with design consistency and building UIs that don't feel like slop. This general approach will probably challenge lots of design-focused coding tools.
The approach to using the browser is superior to Claude's built-in extension in pretty much every way (except cookie management). It's worth it for that alone.
For people who don't understand this...think of each skill like a phase of the SDLC. The actual content, over time, will probably become bespoke to how your team builds software, but the steps themselves are all pretty much the same. All of this is still early days, so YMMV using these specific skills, but I like the philosophy.
As I said on Product Hunt (which upset Garry quite a lot) --
If he weren't the CEO of YC, this wouldn't be on PH, and it wouldn't be on HN.
This is not an impressive setup, folks. It's overengineered and deeply into its own form -- it will not make your agents better, and is likely to make it worse. There are lots of other people to follow/learn from/mimic for skills/context engineering.
Looking at the README file, my first question would be what's his monthly API bill, with my second question being how much of a discount does he get as the CEO of Ycombinator.
Considering he mentions ten sessions at once and I'm pretty confident he wouldn't tolerate waiting for the quota to reset itself... maybe like high four digits with the discount applied, definitely five without it.
But maybe there is some cool stuff here. A lot of prolific AI-assisted engineers I know have their own advanced plan modes, and the CEO plan mode in the repo is interesting (although very token heavy)
I've been using Claude code for a while, probably written close to 100K+ lines over several months.
It is always a learning exercise to see how other people are using CC and I'm sure I'll learn a lot from this, so thanks for sharing it.
But, I don't understand what 600,000 lines in 60 days mean. Lines of code is one thing, but to do what? There still needs to be a loop where CC generates code, there is test automation, maybe do some code review, and then test/run to see what it's built and if it matches the spec, refine the spec, provide new guidance and so on. Products are not built in isolation and are not just KLoC.
Now, if I were asking CC to, take the Algorithms text book and write all the code in all the language etc. (as an example) 600 KLoC over 60 days would make sense. If it were porting an existing product from one stack to another, maybe. But for new products, at least to me that part doesn't make sense.
The best question is, what's been shipped in the past 60 days with those 600,000 lines.
Lots of people trying things for the sake of it, without really achieving anything with it. Maybe they have 'a setup' but the setup ends up being unproven.
I hope I’m wrong, but I’ve seen this pattern a couple of times with close friends: they get obsessed with a topic, their sleep falls apart, they seem manic, and eventually they start doing really strange things online and crash and burn. They usually recover, but by then a lot of relationships are damaged and they’re left with a lot of shame.
Now I know these are symptoms of bipolar disorder/psychosis (they both eventually got professional treatment and told me much later), and I wish I’d known at the time so I could’ve helped. He’s bragging about sleeping 4 hours and joking about having cyber psychosis. [0]
Sleeping only 4 hours is a classic mania symptom.
I’m not as close to Garry, so I don’t know for sure, but some of the behavior feels very similar to what I’ve seen in my friends.
I hope Garry has people in his life who can help. At the very least, you have to sleep — poor sleep is strongly correlated with psychiatric conditions.
For the same reason I care about Elon Musk's decisions after he purchased Twitter: I want more tech CEOs to publish as much of their bullshit online as possible for people to hopefully realise what their "superhuman productivity" actually looks like in practice.
and where's the result? LOC as a side a measure of success is typical for the "omg LLM are amazing and can do it all phase" but once you enter the "actually shipping products people want with human complexity and experience meltdowns" it's usually different....
Well at the current trajectory I'd expect him to release his own OS or something by end of July, his own AWS competitor by October and to close YC applications indefinitely at the end of the year.
But for now I'd be fine with him making his repos public.
edit: There's a few funny threads on other social media. Honestly, though, let a guy get excited, when you find new ways of using new tech; he's one of the lucky 10,000 who has discovered prompt scaffolds. There are better, bespoke tools for more targeted tasks.
Tan is the reason YC batches have gone down hill. I don't think he gets the benefit of the doubt anymore. This is just pure slop for someone way too high on their stash.
Interesting to compare this to Gastown. I also have been starting with a design mode, but I have been doing the ceo side myself. I also rely almost solely on codex for audit - Claude is just too eager and optimistic to make a good auditor.
I think Doll over on bsky has an interesting Claude setup going. Some kind of adversarial mode where they pitch Claude against another model (Gemini, I think) combined with their own “memory” model called Chainlink.
They’ve recently started using their AI pipeline to put out rust-based conversions of tools and it seems to be going pretty well.
Nothing in there about network states, taking over areas of the city and putting people in special coloured shirts to mark them as outsiders and then booting them out, etc
Mostly just markdown-based skills. I've personally had more luck with harnesses, preconfigured permissions, and scripts to automate the frequent workflows, and the repo seems pretty light on that.
Better to have Garry spend his time writing code than using his wealth to attack the very concept of liberal democracy or screech about college students on his blog.
I mean, couldn't Garry spend his time evangelizing YC as providing opportunity to talent that would not otherwise have access to said opportunity instead of either of those? It seems to me like that would be a better use of his time to achieve YC's objectives of funding somewhat competent, highly obsessive, morally flexible founders to create YC portfolio companies others are desperate to invest in at future higher valuations or to acquire. Perhaps I am misunderstanding the purpose of YC though.
He could. But he's also being using his wealth and influence to advocate for the destruction of liberal democracy alongside that activity, so I think it would be a net positive for him to fall into an AI psychosis such that he no longer has time to sabotage society.
The problem with this is that it all runs local on someone's computer, whereas with openclaw you can involve your teammates (e.g. on slack) which is much more powerful.
This reads like a a child telling you about their toys and making up fun little stories about how they all interact together. Or showing you their Minecraft server. How about you explain why anyone should care about this Gary, and no, lines of code aren’t a good reason.
I don’t know. What I do know is that he shares a similar insane accelerationist / libertarian philosophy that is popular in certain drug-fueled circles within tech.
Got it. Thanks for article. Not on West Coast so hearing less about SF things. Don’t see much or anything about the drug part here either - the part I was really curious about.
If these are the people making the decisions (and don't even get me started on the 'technical' folks at a16z...), the cluely-esque enshittification of VC over the last few years makes A LOT of sense.
With that state of mind Gary will be in charge of the FBI in a matter of days. Watch out Kash, there is a new weirdo in town and he got +10 to AI Psychosis.
The way the whole repo is written it’s like he thinks he is the messAIah.
We are all getting sold glass marbles.
When Garry remakes the world into a collection of fiefdoms run by CEO kings, an AI version of dang will be responsible for deciding everybody's "curious conversation."
The speed with which LLMs rot peoples brains is really quite stunning. This is just one of the many reasons why I can't trust anyone whose holding the bag for AI stuff, anyone knee deep in this mess is likely unable to see the horizon.
This is doing more to keep me away from "vibe coding" than anything else. Look, I'm genuinely interested in using AI as a tool and trying to boost my productivity in any way possible - I equate this to activities from the past like learning shortcuts of my editor, learning to type fast, and so on - but, the almost persuasive nature of this README, just pushes me away.
Not to mention using lines of code as a metric of usability is just _whatever_.
This is nothing compared to his self-deifying tweets on the subject.
> In the last 60 days I have written over 600,000 lines of production code — 35% tests — and I am doing 10,000 to 20,000 usable lines of code per day as a part-time part of my day while doing all my duties as CEO of YC.
LOC will never be a good metric of software engineering. Why do we keep accepting this?
I can generate 1 million LOC if I really wanted to.
As long as LOC is the main metric for these setups, they will never be successful.
Garry Tan is a mediocre tech bro with an overinflated ego from being CEO of a VC, of course he'd think shitting out more lines of code is better.
brb unrolling all my loops to get hired by YC
LOC is a very-very weak proxy of "how many new features" I've built, and they don't have any other metric that can be measured easily. But it causes serious issues, because equating LOC with productivity leads to inevitable utter bloat, that no agent or human can ever rectify in a meaningful timeframe. I'm pretty sure this 600 000??? LOC could be shrinked to 60 K for the same feature set, but with better readability and performance.
I have written 600 THOUSAND lines of production code. The best and most beautiful production code. The agents negotiate. They want to make a DEAL. I am the best deal maker in the world. Thank you for your attention to this matter! -- GJT
underrated comment omg. I can hear Shane/DJT’s voice, amazing.
I've been using gstack for the last few days, and will probably keep it in my skill toolkit. There's a lot of things I like. It maps closely to skills I've made for myself.
First, I appreciate how he implemented auto-update. Not sure if that pattern is original, but I've been solving it in a different-but-worse way for a similar project. NOT a fan of how it's being used to present articles on Garry's List. I like the site, but that's a totally different lane.
The skills are great for upleveling plans. Claude in particular has a way of generating plans with huge blind spots. I've learned to pay close attention to plans to avoid getting burned, and the plan skills do a fair job at helping catch gaps so I don't have to ralph-wiggum later. I don't find the CEO skill terribly effective, but I do like the role it plays at finding delighters for features. This is also where I think my original prompting tends to be strong, which could be why it doesn't appear to have a huge impact like the other skills.
I think the design skills are great and I like the direction they're going. DESIGN.md needs to become a standard practice. I think it's done a great job at helping with design consistency and building UIs that don't feel like slop. This general approach will probably challenge lots of design-focused coding tools.
The approach to using the browser is superior to Claude's built-in extension in pretty much every way (except cookie management). It's worth it for that alone.
For people who don't understand this...think of each skill like a phase of the SDLC. The actual content, over time, will probably become bespoke to how your team builds software, but the steps themselves are all pretty much the same. All of this is still early days, so YMMV using these specific skills, but I like the philosophy.
>For people who don't understand this
No one doesn't understand this; the critique is that it is 1) nothing special and/or 2) overengineered.
As I said on Product Hunt (which upset Garry quite a lot) --
If he weren't the CEO of YC, this wouldn't be on PH, and it wouldn't be on HN.
This is not an impressive setup, folks. It's overengineered and deeply into its own form -- it will not make your agents better, and is likely to make it worse. There are lots of other people to follow/learn from/mimic for skills/context engineering.
Wow, the astroturfing by other YC-funded CEOs over there in the comments is quite ridiculous.
> There are other people…
Would you please share a couple? TIA
Check out the strats from the Every team: https://github.com/EveryInc/compound-engineering-plugin (I recommend just learning from their builds and doing your own!)
Simon Willison's blog: https://simonwillison.net/guides/agentic-engineering-pattern...
I don't understand how adding an ad to every single skill is a good idea:
https://github.com/garrytan/gstack/commit/9d47619e4c72136574...
It just unnecessarily clutters the context, in EVERY single skill.
The stanza of co-authored byes in the commit message there is just incredible.
Can you share some pointers of who to learn from? Link? Thx
The Every team: https://github.com/EveryInc/compound-engineering-plugin (I recommend just learning from their strats and doing your own!)
Simon Willison's blog: https://simonwillison.net/guides/agentic-engineering-pattern...
Looking at the README file, my first question would be what's his monthly API bill, with my second question being how much of a discount does he get as the CEO of Ycombinator.
My guesses would be five digits and 90%.
> five digits
before or after the 90%?
Considering he mentions ten sessions at once and I'm pretty confident he wouldn't tolerate waiting for the quota to reset itself... maybe like high four digits with the discount applied, definitely five without it.
I could be underestimating both by a digit.
Yes
That's an absolute insane amount of code 'created', but the natural follow up is: for what? Are there examples of what this software has created?
LLM generated READMEs hurt my eyes
But maybe there is some cool stuff here. A lot of prolific AI-assisted engineers I know have their own advanced plan modes, and the CEO plan mode in the repo is interesting (although very token heavy)
https://github.com/garrytan/gstack/blob/main/plan-ceo-review...
I've been using Claude code for a while, probably written close to 100K+ lines over several months.
It is always a learning exercise to see how other people are using CC and I'm sure I'll learn a lot from this, so thanks for sharing it.
But, I don't understand what 600,000 lines in 60 days mean. Lines of code is one thing, but to do what? There still needs to be a loop where CC generates code, there is test automation, maybe do some code review, and then test/run to see what it's built and if it matches the spec, refine the spec, provide new guidance and so on. Products are not built in isolation and are not just KLoC.
Now, if I were asking CC to, take the Algorithms text book and write all the code in all the language etc. (as an example) 600 KLoC over 60 days would make sense. If it were porting an existing product from one stack to another, maybe. But for new products, at least to me that part doesn't make sense.
The best question is, what's been shipped in the past 60 days with those 600,000 lines.
Lots of people trying things for the sake of it, without really achieving anything with it. Maybe they have 'a setup' but the setup ends up being unproven.
I hope I’m wrong, but I’ve seen this pattern a couple of times with close friends: they get obsessed with a topic, their sleep falls apart, they seem manic, and eventually they start doing really strange things online and crash and burn. They usually recover, but by then a lot of relationships are damaged and they’re left with a lot of shame.
Now I know these are symptoms of bipolar disorder/psychosis (they both eventually got professional treatment and told me much later), and I wish I’d known at the time so I could’ve helped. He’s bragging about sleeping 4 hours and joking about having cyber psychosis. [0]
Sleeping only 4 hours is a classic mania symptom.
I’m not as close to Garry, so I don’t know for sure, but some of the behavior feels very similar to what I’ve seen in my friends.
I hope Garry has people in his life who can help. At the very least, you have to sleep — poor sleep is strongly correlated with psychiatric conditions.
[0] https://youtu.be/W3YpC4Dvzso?t=929
Can confirm this experience, as someone who took 10 years to be diagnosed with bipolar type 2 (the median amount of time, unfortunately).
But, if he is bipolar, he would have experienced hypomania/mania before. This wouldn't be the first time...
Poor sleep is correlated with lower life expectancy. 12%?
[flagged]
couldn’t?
Maybe it's accurate, one could conceivably care less about them if they cared more about paying them attention?
They could also care at n+1 in the set of whole numbers where n/max is infinitesimally small.
Why should anyone care about this?
For the same reason I care about Elon Musk's decisions after he purchased Twitter: I want more tech CEOs to publish as much of their bullshit online as possible for people to hopefully realise what their "superhuman productivity" actually looks like in practice.
and where's the result? LOC as a side a measure of success is typical for the "omg LLM are amazing and can do it all phase" but once you enter the "actually shipping products people want with human complexity and experience meltdowns" it's usually different....
Well at the current trajectory I'd expect him to release his own OS or something by end of July, his own AWS competitor by October and to close YC applications indefinitely at the end of the year.
But for now I'd be fine with him making his repos public.
Omg, this is like god mode.
edit: There's a few funny threads on other social media. Honestly, though, let a guy get excited, when you find new ways of using new tech; he's one of the lucky 10,000 who has discovered prompt scaffolds. There are better, bespoke tools for more targeted tasks.
Tan is the reason YC batches have gone down hill. I don't think he gets the benefit of the doubt anymore. This is just pure slop for someone way too high on their stash.
>In the last 60 days I have written over 600,000 lines of production code — 35% tests — and I am doing 10,000 to 20,000 usable lines of code per day
That would be considered a huge liability and shameful historically.
Interesting to compare this to Gastown. I also have been starting with a design mode, but I have been doing the ceo side myself. I also rely almost solely on codex for audit - Claude is just too eager and optimistic to make a good auditor.
I think Doll over on bsky has an interesting Claude setup going. Some kind of adversarial mode where they pitch Claude against another model (Gemini, I think) combined with their own “memory” model called Chainlink.
They’ve recently started using their AI pipeline to put out rust-based conversions of tools and it seems to be going pretty well.
Nothing in there about network states, taking over areas of the city and putting people in special coloured shirts to mark them as outsiders and then booting them out, etc
Mostly just markdown-based skills. I've personally had more luck with harnesses, preconfigured permissions, and scripts to automate the frequent workflows, and the repo seems pretty light on that.
I am seriously worried about this guy's mental health at this point.
Why's a CEO writing code, doesn't he have a company to run?
Better to have Garry spend his time writing code than using his wealth to attack the very concept of liberal democracy or screech about college students on his blog.
I mean, couldn't Garry spend his time evangelizing YC as providing opportunity to talent that would not otherwise have access to said opportunity instead of either of those? It seems to me like that would be a better use of his time to achieve YC's objectives of funding somewhat competent, highly obsessive, morally flexible founders to create YC portfolio companies others are desperate to invest in at future higher valuations or to acquire. Perhaps I am misunderstanding the purpose of YC though.
He could. But he's also being using his wealth and influence to advocate for the destruction of liberal democracy alongside that activity, so I think it would be a net positive for him to fall into an AI psychosis such that he no longer has time to sabotage society.
Missing a satire disclaimer
The problem with this is that it all runs local on someone's computer, whereas with openclaw you can involve your teammates (e.g. on slack) which is much more powerful.
LOL imagine if this guy was your boss. Scary thing is though, my boss is starting to sound like this :D
In the last 60 days I have written over 600,000 lines of production code
No you haven't.
> In the last 60 days I have written over 600,000 lines of production code — 35% tests — and I am doing 10,000 to 20,000 usable lines of code per day
and what is there to show for it? absolutely terrible metric
OK, I am an AI accelerationist, but this quote... Wow. Are we really back to measuring KLoC?
Wow, HN taking a marketing play at face value. Shocker.
This reads like a a child telling you about their toys and making up fun little stories about how they all interact together. Or showing you their Minecraft server. How about you explain why anyone should care about this Gary, and no, lines of code aren’t a good reason.
VC investor metric brain. Right?
If you can call a ketamine soaked, amphetamine-fried tangle of nerves a brain.
Interesting. Is he a known user? Also curious if you are more broadly against some of that overall?
I don’t know. What I do know is that he shares a similar insane accelerationist / libertarian philosophy that is popular in certain drug-fueled circles within tech.
Proof here: https://48hills.org/2024/05/the-alarming-agenda-of-the-big-m...
Got it. Thanks for article. Not on West Coast so hearing less about SF things. Don’t see much or anything about the drug part here either - the part I was really curious about.
Right.
If these are the people making the decisions (and don't even get me started on the 'technical' folks at a16z...), the cluely-esque enshittification of VC over the last few years makes A LOT of sense.
“Gary, this is a text file.”
https://www.youtube.com/watch?v=Q6nem-F8AG8
>somewhere right now, an LLM is saying 'great work' to a man who just committed a text file to github
this is fantastic, my exact thoughts looking at this repo
LOL
"it's a bunch of files telling Claude to pretend to be different people"
I swear that was my analysis as well, verbatim.
This needs to be higher
Garry clearly is.
With that state of mind Gary will be in charge of the FBI in a matter of days. Watch out Kash, there is a new weirdo in town and he got +10 to AI Psychosis.
The way the whole repo is written it’s like he thinks he is the messAIah. We are all getting sold glass marbles.
Get some damn sleep Gary.
I swear to God I'm making a script to write \n to a file and call it productivity increase on social media.
What a disgrace, hacker culture died to this
I’m still hanging around to see it all crumble down honestly. Get the popcorn.
This was flagged. Why did the mods remove flags? How is this anything other than nepotism?
When Garry remakes the world into a collection of fiefdoms run by CEO kings, an AI version of dang will be responsible for deciding everybody's "curious conversation."
Recently flagged posts that were unflagged: this. Recently flagged posts that are still flagged:
Teens sue xAI over Grok's pornographic images of them (bbc.com) - https://news.ycombinator.com/item?id=47406721
Turns out the DOGE bros who killed humanities grants are sensitive about it (techdirt.com) - https://news.ycombinator.com/item?id=47412848
DOGE didn't cut government waste, it was government waste (techdirt.com) - https://news.ycombinator.com/item?id=47415103
Curated Female Founder Cohort in SF (jointheden.co) - https://news.ycombinator.com/item?id=47418184
At this point I lost any gram of respect I used to hold for the mods here.
Anyone who brags about how many lines of code he creates has already lost the plot.
Is any of it trustworthy?
The speed with which LLMs rot peoples brains is really quite stunning. This is just one of the many reasons why I can't trust anyone whose holding the bag for AI stuff, anyone knee deep in this mess is likely unable to see the horizon.
[flagged]
[flagged]
[flagged]