Rendered at 16:20:26 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
bartread 1 days ago [-]
The thing is, it's not actually that hard to spend at that kind of rate.
I wouldn't say this was the norm, but I have burned through £1k of tokens in a couple of days before. And, on that same gig, I realised a test I was considering running would burn £2.5k worth of tokens over the course of a few hours so spent a couple of hours of my own time re-engineering it to get that down to £150.
The point is that I don't necessarily know that spending a shitload of money on tokens really tells us anything useful. If he's getting a return on that investment, fine, whatever.
Obviously most of us couldn't afford to spend that much ourselves, no matter how much value it might deliver, and many of our employers wouldn't be too happy if we spent it on their dime either. Equally, if you can, and it's delivering real value, then I don't see why you wouldn't?
plusfour 24 hours ago [-]
> would burn £2.5k worth of tokens over the course of a few hours
details please, if it's interesting
deadnomore 23 hours ago [-]
Evaluation loops do this easily. You have to run hundreds of reasoning jobs in parallel and very quickly you can torch thousands of dollars.
bartread 20 hours ago [-]
Yeah, it was basically this, and running a whole lot of agents concurrently to churn through the dataset faster.
sixtyj 22 hours ago [-]
Yup, run multiple agents on expensive model in loop via API.
You can do it if your employer has endless pockets.
Example: $30/1M input tokens (I don’t count output or cache, for simplification), you have several daily agents for everything => 3M tokens daily per agent * 10 agents ~> $900/day => $328k/year (API spending)
hparadiz 23 hours ago [-]
Any 3D work in blender can burn that easily. Add video editing to that mix in any capacity.
nodesocket 23 hours ago [-]
At what point do you buy a Mac studio or AMD Gorgon Halo and try running some of these tasks in local models? Surely don’t need frontier models for all this work.
bartread 20 hours ago [-]
You definitely don't need frontier models for all this work.
And, in fact, I got it from £2500 to £150 by careful model selection (I ran a trial over a small dataset with different models until I found the least expensive/capable model that would reliably do what I needed), using a local model for one step, and paring down the inputs as well so the token count plummeted.
wolvoleo 14 hours ago [-]
Yeah I see the same thing at work. Most of the people burning crazy numbers of tokens aren't doing any amazing work. They're just using fable to write their emails and do their web searches.
They feel like they're super smart but it's just as smart as hiring a $500k/yr PhD as a PA to schedule your meetings and answer the phone. Sure it does a good job but you're just wasting money and abilities.
They feel super edgy and bleeding edge but all they really do differently is switching the AI chat from auto to the most expensive option all the time. When I see what they're using it for they're only grasping the basics of its abilities.
They're also people sending 2 page AI generated emails that make no sense for literally everything where a "Yeah sure let's do that" would have sufficed.
Unfortunately most of these are business leaders who just want to look edgy and "all in on AI" so we can't really limit their tokens. Yet they're also the ones saying we spend too much and want to limit some of our dev teams actually putting it to good use.
bigfatkitten 7 hours ago [-]
I tokenmaxxed in my last role.
Management succumbed to AI psychosis, and so we were directed to show that we were adopting and using AI, but what we were using it for was not considered important. So we went out of our way to burn as many tokens as possible.
If you tell Claude to be inefficient and burn as many tokens as it can, it’ll happily oblige. It will also offer useful advice as to how to use even more tokens.
jorvi 13 hours ago [-]
You don't.
Once you factor in cost of electricity and depreciation of hardware, it is more expensive to run locally.
There are many arguments for running local but cost isn't one of them.
nodesocket 13 hours ago [-]
> but cost isn't one of them.
Let's do quick back of the napkin math.
- Mac Studio M5 Ultra, 80-core GPU, 256GB memory, 2TB storage -- $11,299.00.
- Power (385w continually 100% of month) is $40/mo (my regional) or $480 annually. Let's say power is 3x more expensive in California, which it probably is, that's still only $1,440 annually in power.
The Studio M5 Ultra has a worklife of at least 3 years.
jcelerier 5 hours ago [-]
it's going to be quite slower than remote providers, won't be able to run full models, only quantised, and for more cost than the highest Claude subscription tier where you can do roughly 60-100M token per month at 200$, e.g. 7200 for three years. It makes sense if you prioritize local AI but not really otherwise sadly.
sho 10 hours ago [-]
As many others have said, that's not even slightly impressive. Someone maxing out a single claude ultra plan on opus 5.5 is already doing ~$160k+ per year if the massively inflated API prices are used. I personally have 5 and the only thing stopping me getting more is concern about being banned, I could easily double my usage if it was safe and convenient. Hell, I could probably 10x it or more.
Once you make the jump to agents managing agents the only limit is your budget and your harness's ability to coordinate. We'll all be using far more tokens than this in a year or two.
mi_lk 9 hours ago [-]
What do you do that you are maxing out 5 accounts
sho 7 hours ago [-]
Day job, multiple side projects which are their own entire public apps, research, personal agents, running and developing the custom harness that orchestrates all of the above..
It sounds like a lot when I type it like that, but it's not actually that much if you delegate brutally. Like I said, I'm constantly token starved and could do a lot more!
iLoveOncall 7 hours ago [-]
Feeling productive while achieving nothing.
avaer 9 hours ago [-]
I want to know if there are any returns on that. Even with the insane discount on API prices.
Or are people paying $1K/mo (subsidized) just because AI is fun? Can you imagine a video game charging $1K/mo?
roryirvine 7 hours ago [-]
I absolutely can imagine a video game doing that, especially if there were some way to make the consumption be conspicuous.
In Tan's case, I think the "fun" thing is the right way to look at it. He gets to play at being a big shot developer and show off on Twitter, so clearly he feels he's getting some value from his spending even if no-one else does.
There is a more serious question about to what degree the AI vendors have a duty of care to prevent addicts from getting themselves into financial trouble - perhaps they might in future face the sort of regulation applied to high street bookmakers. I doubt Tan would fail any putative affordability test, though, so it would still be up to him to seek help for his problem if he feels its getting out of control.
sho 7 hours ago [-]
Well, I run a number of apps that have users who appreciate them (I hope?) and they simply wouldn't exist without AI giving me the leverage to build them. So.. yes? What's a "return", anyway? It's worth it to me, obviously!
But you might have hit the nail on the head - yes, it's extremely fun and maybe that's the main reason after all. And FYI many, many people pay a lot more than $1k/m for video games.
killingtime74 8 hours ago [-]
If you claim it as a business/work expense (In Australia, that gives you a 47% discount), 1K is not even one day's salary for a well-paid dev and just about one day for an average paid dev.
tarballing 31 minutes ago [-]
Average base compensation for SWE is ~140k (in the US w/ the highest earning SWEs), so the average is roughly half of that. Earning ~275k is comfortably in the top 20% of all SWEs; 1k a day is easily in the top 5%.
Zanfa 8 hours ago [-]
Are Australian software devs really averaging 250k USD per year?
bigfatkitten 8 hours ago [-]
No, though a very small minority are in that range with equity etc from Big Tech.
killingtime74 8 hours ago [-]
Its bimodal. If you are at one of the big ones yes. Its on levels.fyi (you can narrow to Australia)
HWR_14 9 hours ago [-]
Even if it was just for fun, $12,000 a year for a hobby is not cost prohibitive for a lot of people.
tarballing 41 minutes ago [-]
Median average single salary in the US is ~64k a year. Spending ~19% of all income on a single hobby is completely out of the question for the vast majority of people.
dimbletimbers 4 hours ago [-]
That there are people like you in a world with starving children sincerely turns my stomach.
tuesdaynight 48 minutes ago [-]
If that is the case, how do you use HN or any social media without vomiting all day?
tikimcfee 4 hours ago [-]
There really is no excuse, is there?
This is my daily intrusive thought. "Why the hell am I doing any of this? Why am I not yelling louder at people that have so much and give so little?"
This website is full of people that could save millions of human lives in aggregate, but choose not to and actively argue against the idea because, and I do swear it's a damned curse at this point, "there is no incentive to do it."
I'm not even remotely rich. I have no "savings" or "investments". Yet I have a comfortable life in a house with access to food and care as I need. Sometimes I really hate it.
randysalami 3 hours ago [-]
To your last part, then you’re ok man. Sounds like you have your kindness and empathy intact. It sounds like given the avenue, you would organize for what is right. You have perspective but don’t let it consume you. You need to be healthy because if you care, there might be a chance in the future where you can put that care toward change. And you need to be in your best state to have the largest impact.
There are many people in this world who have more, view people as dirt, and dedicate their lives to keeping people down. Good, kind people who are disempowered by a system that betrayed us all are not at fault. Feel better man.
tuesdaynight 51 minutes ago [-]
You cannot change the world, you need to accept that. However, you can have a hell of an impact on your surroundings, like your close family, neighborhood or city. Focus on that. Social media made us forget that.
You can get access to DeepSeek 4.1 Flash for almost nothing, on a $1, $10, or $100 subscription. That can complement your 5 Claude accounts, or replace some. DS Flash 4.1 is so fast and cheap it’s ridiculous
sho 42 seconds ago [-]
All such sites are data harvesting operations. They'll collect your transcripts and sell that data to model companies to train on. It's a big business.
I occasionally get my agents to do sensitive things; I'm not comfortable having my transcripts collected and sold, so I don't use them. They say they anonymize them but who can ever tell? Not for me, thanks.
fantasizr 24 hours ago [-]
I can't recall software being conspicuous consumption until the advent of the AI token. Maybe the first "I am rich" app in the app store for $999.
johnfn 23 hours ago [-]
Pretty sure Joel Spolsky said “we purchase the best machines money can buy for our developers” like 20 years ago. So it’s hardly a new concept.
fantasizr 23 hours ago [-]
I personally recall hardware flexing going back to SGI, but nothing like "look at my salesforce spend".
isolatedsystem 11 hours ago [-]
Not sure why that counts. An engineering workshop buying mint quality lathes would just be considered "Buy once, cry once" in my eyes.
asdff 8 hours ago [-]
Lathe sure. But what about shiny macbook pro to ssh into aws
trymas 5 hours ago [-]
eh, compared to salaries even shiniest macs wouldn't register in USA, especially hubs like SV. Especially if it's issued once a couple or four years.
With AI spend - especially at Anthropic/OpenAI API prices - spending 0.5-1 macbook a month worth of tokens per employee seems normal (from what I've seen, heard, what happens in this forum and thread and what my employer does).
Tokenmaxxxing (followed by commit/PR-maxxxing) is celebrated. Don't think - just launch orchestrator that will create dozen child agent sessions to do useless work to impress your superiors. If buy the end of the day you will ask to increase your token allowance - you're a superhero.
raincole 10 hours ago [-]
> machines
That's hardware.
ASalazarMX 23 hours ago [-]
I've seen the other side of that coin: "my program is blazing fast in my machine, I don't know why those naggy users are complaining!"
fantasizr 2 hours ago [-]
Getting your app to run on the worst devices was artistry. Assuming the lowest IE and windows environment.
addaon 12 hours ago [-]
> I can't recall software being conspicuous consumption
Hardware for sure. Big statement giving the entire company macs vs pc. Apple spelled it out in those ads.
itishappy 24 hours ago [-]
AWS spend and other service bills?
fantasizr 24 hours ago [-]
My take was when people would post those large bills it was mostly to complain about Amazon or a process having gone wild for comedic purposes.
itishappy 23 hours ago [-]
My take is that disclosures like these are always humble brags and/or marketting (look at how much I can afford to spend) with minor complaints rolled in to disguise this fact.
> “It’s too expensive right now,” he said, adding that he knows the amount of money he spends on AI agents is “a crazy privilege that not everyone can do.”
maxbond 24 hours ago [-]
Superchats and other livestream gifts
fantasizr 24 hours ago [-]
Agree I meant mostly in the vein of "look how much I can spend on XYZ because I'm rich and I can" with questionable value.
bix6 24 hours ago [-]
> he’s suffered from “cyber psychosis” and sometimes barely sleeps
Sleep is one of the most important things in the world. Why not have the agents work while you sleep?
hackyhacky 24 hours ago [-]
> Why not have the agents work while you sleep?
Even better, why not have the agents sleep for you?
EDIT: for those who say this is not realistic, I've had Claude sleep for 10 hours in just 30 seconds. Saved me an entire night.
Freedom2 23 hours ago [-]
Care to share your agent setup and harness? I've been trying to do something similar but my Claude only sleeps for about 2 hours before waking up.
hackyhacky 22 hours ago [-]
I use a supervising agent, so if my sleeping agent has insomnia, the supervisor puts it back to sleep. It's helpful to have the supervisor put on relaxing music or read it something boring [1].
I can't tell if posts like this are parody or serious
zombot 10 hours ago [-]
Even if they're parody (this one is), there are enough people now who take them seriously and put them into practice. I have never seen any other tech that surfaces that amount of idiocy.
r_lee 10 hours ago [-]
could you make a skill for this? Thanks
cozzyd 15 hours ago [-]
It helps to increase your network latency
thephyber 23 hours ago [-]
If his brain is the thing malfunctioning, then his brain may not be capable of (1) correctly diagnosing the problem as a lack of sleep or (2) honestly prescribing "do things to get better sleep" as the fix.
People in psychosis don't exactly weigh all of the options with a level head. He may not recognize that he has "cyber psychosis" or that his lack of sleep is a bad thing.
killingtime74 8 hours ago [-]
If he was good at it, he would be sleeping more because you would be doing less work.. The fact that he's sleeping less means he's not as good as optimizing his agents as he thinks he is.
Gud 8 hours ago [-]
Or, he is operating more agents by staying awake longer.
Don’t make me defend Garry Tan…
itishappy 24 hours ago [-]
That's probably exactly why. Some of my latest nights have been attempting to prepare a long-running overnight task for an agent.
It was productive, but as you say, it's dangerous to gamble with sleep.
ks2048 24 hours ago [-]
YC seems big on quantitative metrics. Is there anything that says he is more efficient than 3 or 4 years ago. Is YC as a whole?
bwfan123 14 hours ago [-]
> Is there anything that says he is more efficient than 3 or 4 years ago.
This is a difficult question to answer ! I would ask a related question which is equally hard. Has AI-use increased your human cognitive capability or decreased it ?
__s 9 hours ago [-]
Decreased it. I no longer carry the entire code base in my head. Used to be someone could describe issue to me & I could know exact line where bug was. Sometimes laying in bed I'd realize some line was wrong
I accept this loss because the productivity is there. & hope to retire in a few years if things don't go south. For people entering the field I worry
asdff 8 hours ago [-]
He will say he is and that will be the end of it.
glennfiedler 24 hours ago [-]
He’s gotta pump those numbers up, those are rookie numbers in this racket
asdff 8 hours ago [-]
Yeah, why not 400k? Low energy.
glennfiedler 46 minutes ago [-]
1.2 or GTFO
Vishal_Max 4 hours ago [-]
For a normal person outside the Startup world this news is super confusing and frustrating while most people who already use ai heavily think this is nothing and they are surprised by the fact that he is not even spending that much!
alpineman 3 hours ago [-]
Someone needs to tell this guy about the subscriptions
bluegatty 13 hours ago [-]
the question is not spend it's outcome.
what has he produced with that?
clickety_clack 12 hours ago [-]
He must have the best automatic daily slack summary.
bluegatty 11 hours ago [-]
Probably not even that, there are off the shelf things.
I don't think he has produced anything.
Tokenmaxxing was a dumb trend he has not figured out he's out of the loop yet.
fakedang 11 hours ago [-]
Revenue for Anthropic to use in their S-1.
bluegatty 9 hours ago [-]
what does that have to do with anything?
Gary spending 250K/year on nothing is not strongly related to pros using $200/plan to do much better.
whattheheckheck 10 hours ago [-]
You haven't seen tanstack? He presented to Stanford he wanted to scale that idea to an entire organization not just a solo founder.
He's gonna hit the halting problem
bluegatty 9 hours ago [-]
it's a hulking pile of garbage.
up2isomorphism 24 hours ago [-]
So what has he built?
Jtsummers 24 hours ago [-]
A site where he complains about rich people being charitable, among other things.
bitwize 16 hours ago [-]
1920s, Henry Ford: "I want to build automobiles any man can afford, even the workmen in one of my factories."
2020s, techbros: "Here's why it's a huge problem for non-billionaires to have things."
georgemcbay 23 hours ago [-]
Looks like this is getting downvoted, but it is absolutely true.
Garry Tan is fully aligned with Elon Musk in going after MacKenzie Scott's charitable giving, making spurious arguments that she's doing it in the wrong way when they are all just running cover for the fact that she's one of the few at that level of wealth being charitable at all these days.
Silly woman, she should give her money away the right way, which I guess is by attempting to buy political power.
Jtsummers 23 hours ago [-]
https://news.ycombinator.com/item?id=47042593 - HN discussion (5 comments) on that particular rant of his. And it looks like you were one of the few people to comment on it, so really this is for other people who are missing the context of our comments.
Finally, the perfect yes-man, the thing execs and VCs always wanted.
jgalt212 24 hours ago [-]
a bubble.
flyingsky 8 hours ago [-]
Might make sense for him to invest in hosting his own models locally. Prices are only gonna keep going up.
_-_-__-_-_- 14 hours ago [-]
As someone who has only ever used these free AI agents and who is not a software developer or computer scientist by trade, could someone please explain the concepts of tokens and their worth and harnesses and what exactly all of those things accomplish?
ygjb 14 hours ago [-]
None of these are effective questions for this forum. You should copy and paste your post and my response into your free chatbot of choice.
Evaluate the original post and response and build a high level learning plan with citations and references that covers each of the topics in the original post. Start the learning plan with an overview of effective prompting techniques with specific examples to research individual topics.
void-pointer 11 hours ago [-]
A Token is a unit of data used by large language models. Tokens used to be pretty analogous to 3-6 letters of English text, but things have become more complicated with the advent of multi-modal generation
Large language models are usually priced on MTok or “million tokens”. You’ll often see LLMs priced at $X/MTok for input and $Y/MTok for output.
A harness is a program that interfaces with a large language model and provides it tools and context to achieve a goal. Codex and Claude Code are good examples of harnesses.
What exactly all of this accomplishes is not something I can answer for you :)
x-complexity 9 hours ago [-]
> the concepts of tokens
Tokens are what you get when you have to map a chunk of text to specific numbers, so that the LLM can actually work with it.
[Your text] --> [List of numbers (tokens)] --> [List of [List of numbers describing a concept/term/thing/job/item/etc] (embedding)]
That final thing (List of embeddings) is what gets fed into the LLM, so that it can produce (List of embeddings most likely to follow afterwards), which is then turned back into text.
> and their worth
Tokens are priced based on the model you're using. In general, expect to pay more when you need to use larger models. (Note: Exceptions exist, but treat them as exceptions)
The space as it is now prices tokens based on:
- how many tokens it needs to process (input),
- how many it needs to generate (output), and
- how often you send the same chunks of text that their system already has on file (cached).
> and harnesses
A harness is something that adds in tools & directions for the LLM, so that it can do its tasks. In a dumbed-down semi-visual summary:
---
[Text about what the LLM should be & what they're allowed/not allowed to do]
[Text about available tools & how to call them]
[Your text here]
---
The harness, if its part of a program/app, then looks at what the models generate in response, and does the (file writing/reading, tool calls, web searches, subagents, etc.), based on the formatted output that the LLMs should produce.
> and what exactly all of those things accomplish
Depends on the goal of the user. Some use them as assistants, most use it for coding.
> As an early OpenClaw adopter who is now a power user, Tan says one thing he’s learned is not to treat AI agents like traditional software. Some people are so worried about “repeatability and safety that they actually trade off all of the beautiful parts of working with an AI agent, which basically can be so similar to a human,” he said.
This is insane. He's turning the failure of his AI assistants to do what he wants every single time and to do it in a safe manner into a positive thing. Failure as a positive!
thephyber 23 hours ago [-]
To be fair, using the scientific method to disprove superstitions is what the Enlightenment was all about and that is the most significant engine of growth and flourishing humankind has known.
That said, I approach AI with deep skepticism. I try to learn it, including its strengths and weaknesses. But I don't blindly trust it.
Coincidentally, my stock brokerage just yesterday sent an email reminding users that anything their agents do using the user's credentials is the responsibility of the user.
zombot 10 hours ago [-]
> Failure as a positive
Since failure is such a likely outcome in a startup, "failure as a positive" is a mindset that a startup accelerator must propagate to keep founders going. And when you sell a message for long enough, you start to believe it yourself.
2001zhaozhao 13 hours ago [-]
This figure really depends on whether it's API rates or actual spend. One Claude Ultra subscription ($2400/year) gives you nearly $100k a year in tokens if fully utilized.
stuaxo 13 hours ago [-]
Anyone worked out the carbon footprint of that ?
A_D_E_P_T 1 days ago [-]
Come on. If you have a Claude Max 20x plan ($200/month) you're massively subsidized and your API cost is anywhere from $5000-14000 per month.
If you hit your limits on two Max 20x plans, you could very well be exceeding $300k/yr in token use at current API rates.
pu_pe 24 hours ago [-]
Is that how these figures are being calculated? Because if so, that's laughable. I always interpret those to mean that they are literally paying $200k/year to Anthropic, not ~$2k.
TiredOfLife 22 hours ago [-]
Not only that. They simply multiply tokens used with API price per token, but ignore that about 90% of tokens are cached ones that are significantly cheaper.
asdff 7 hours ago [-]
Api would be cheaper still if there were no subscriptions
yfw 10 hours ago [-]
An employee uses 200k of electricity. Do we fire or promote him?
asdff 8 hours ago [-]
Sun Tsu would probably say do nothing. World sees you have an ordinary employee using 200k of electricity and you don’t care. Quite a show of force.
ShadowOfThePit 1 days ago [-]
> ...his agent also helps him run his life. “It’s read all of my email and all of my Dropbox. It’s read, like, all of my Apple Notes, even my therapy notes going back 20 years”.
> He describes using AI agents as getting help with “almost anything that you do in your day-to-day life. (...) Almost anything you want to do, it knows you, and it’s there to help. And it will actually sort of run ahead and say, ‘What about this?’ Or it’ll help you see around corners.”
Wow, impressive. Both the capabilities and how much he's willing to share.
Is it worth $17k per month? For me, absolutely not. But people spend more on luxurious cars and such, so if you've got money to spare...
> Y Combinator is also using an agent that is “rebuilding the way we operate at YC. We now have an internal AI agent, for instance, named Stray Light that kind of has a personality, and then it has a company brain. It actually has the total context of every founder we’ve ever backed, anyone who’s ever emailed us, like all the applications.”
Holy anthropomorphization. But besides that, what's the goal of doing this? Keeping a handy overview of when and why someone said something?
switchers 24 hours ago [-]
Naming your internal AI agent after the massive run down home of an uber rich but probably insane family of confusing clones who made god level AIs (from Neuromancer)... Classy.
zbentley 4 hours ago [-]
Major “we built an AI to help our business called Torment Nexus” vibes, yes.
dreadsword 23 hours ago [-]
That was my thought --- an odd and offputting choice of identity
timmytokyo 22 hours ago [-]
When they show you what they are, believe them.
pythonaut_16 21 hours ago [-]
Still doesn't explain where $17k of tokens is going. I can believe someone could burn that many tokens, but most of what he's described there could be managed pretty easily on one of the $100-200 a month subscriptions I'd think.
The only way I can imagine to burn that many tokens is just to run every task fully through the most expensive LLM with no real strategy. Like indexing emails should be a combo of scripts and small summarizing prompts I'd think.
randysalami 1 days ago [-]
“ Holy anthropomorphization. But besides that, what's the goal of doing this?”
Because they can. It’s as simple as that. To find meaning, to feel in control, to win. To dominate.
dgellow 7 hours ago [-]
I like how the can absolutely never provide anything specific. It’s always „it helps with everything!“. Like what exactly Gary?
27183 24 hours ago [-]
He must be the most annoying person to try to interact with. Clodslop for days.
SpicyLemonZest 24 hours ago [-]
> Holy anthropomorphization. But besides that, what's the goal of doing this? Keeping a handy overview of when and why someone said something?
You'd be surprised how important such an overview can be in relationship-based businesses. People spend a lot of time trying to track down a vague memory they have of someone telling them something. CRM platforms are big for a reason, after all.
dd3ddaa 13 hours ago [-]
[dead]
bill10 12 hours ago [-]
Interesting he doesn't get free accounts?
flyingsky 8 hours ago [-]
Whats in it for the labs if they give Gary Tan free tokens for uhm, Gstack?
stego-tech 24 hours ago [-]
Headline aside, I agree with him wholeheartedly that the future of AI is personal agents, and that costs have to come down for that to happen.
Not cloud costs, though; consumer costs. These agents must run on-device for privacy protections, because anything less will be a global non-starter. That means we need models that excel on phones, tablets, laptops, and desktops; we need home compute modules to run larger, “frontier”-class models for bigger tasks but without guzzling electricity (the Mac Mini’s power envelope is an ideal target); we need smart harnesses that can foist off inference to bigger models elsewhere on your private network of devices as needed without the user prompting for specific thinking modes or model names. All of this has to happen automatically, privately, securely, without leaking data to advertisers or surveillance companies.
It’s hard work, but it’s definitely the future of AI. I would argue the business case is solved with being able to build bespoke stacks of kit and run models through neocloud providers with ZDR policies. It’s the consumer angle that’s hard to crack, because everyone wants a repeat of Web 2.0 for ad-dollars instead of a privacy-centric modern tech stack that forces companies into lower revenue streams.
randysalami 2 hours ago [-]
Had an idea for rich people where they buy their own AI server. The value add is the handling of hardware and software is all provided by the company. It’s a luxury experience like the iPhone so inherently privacy focused. It’s E2E so non-technical users just see AI at the end delivered in the best way possible with any customizations they want. You build the most comprehensive profile on their life which every AI company would love but it’s possible because you’re actually privacy-focused. Call it MyAI.
Dr_Birdbrain 23 hours ago [-]
I don’t think the privacy-centric vision is going to happen. Very few non-tech people care enough about their privacy to figure out the necessary infrastructure.
The future of personal assistants is most likely akin to Muse Spark, and yes I expect Meta is training on all of those interactions and using them to target ads.
argee 23 hours ago [-]
> because anything less will be a global non-starter.
I beg to differ.
"People just submitted it. I don't know why. They 'trust me.' Dumb fucks."
- Mark Zuckerberg
tibbydudeza 6 hours ago [-]
So what has he shipped - let me guess - nothing of wrong.
opengrass 22 hours ago [-]
That's only 1 SWE or half with sloppy results.
jgalt212 15 hours ago [-]
His tools, gstack, are 360K lines of code. That's probably a lot of tokens and context.
Look at this guy using shell commands instead of asking his agent to count the lines. You’re going to get left behind bro! Your boss won’t pay you to do that!
dgellow 7 hours ago [-]
Markdown, not code
hartjer 24 hours ago [-]
Kind of low isn't it?
jrflowers 13 hours ago [-]
I love that the world’s dumbest guys all get up and pay the computer to tell them they’re big special boys then publicly say stuff like “my brain is bad in a video games way”
> he’s suffered from “cyber psychosis” and sometimes barely sleeps
asdff 7 hours ago [-]
At his level the skill is about getting money out of the right pockets, not really about being particularly smart or even very good at managing business.
rhines 2 hours ago [-]
It reminds me of how we laugh at scam emails for being poorly written and obviously a scam, but the theory is that this filters out all but the most gullible people so the conversion rate on anyone that replies is high.
If your business model is taking money from gullible investors, maybe it's reasonable to appear (or actually be) incompetent so that you don't accidentally waste your time on people who haven't drunk the Kool-Aid.
m3kw9 23 hours ago [-]
I wonder how many of those agents get lost in a loop and just uses tokens, asking the same questions over and over and responding and asking
asdff 7 hours ago [-]
I wonder how much this happens in these huge multi agent jobs. Two agents quietly changing the same file back and forth.
bamboozled 10 hours ago [-]
What does he have to show for it? It helps him out to do research, seems like an expensive helper.
lbrito 23 hours ago [-]
"Do my job, make no mistakes"
That's basically gstack
digitaltrees 24 hours ago [-]
Why not buy hardware at that point?
TiredOfLife 22 hours ago [-]
That usage probably comes from a couple of 200$ a month subscriptions. Hardware to run something that approaches frontier at acceptable speed is at least a million.
wahnfrieden 24 hours ago [-]
I spend that much at API rates with a couple subscriptions
ex1fm3ta 1 days ago [-]
hmm isn't that the job of a concierge, or an assistant ?
xhkkffbf 1 days ago [-]
Didn't you mean to use past tense? "Wasn't that the job ...."
ShadowOfThePit 24 hours ago [-]
Touché.
Counter-point, they can't serve as a knowledgeable discussion partner, programmer, search engine, or your personal psychologist. Not saying that they should.
slopinthebag 23 hours ago [-]
he just needs to put the finishing touches on g-stack and then programming will be solved. just another $50k will do it i think!
lambdadelirium 1 days ago [-]
Website looks still like shit
ShadowOfThePit 1 days ago [-]
I think it looks neat and simple.
Also...
> Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting.
Just 200k a year? If this amount is for some folks news worthy get a reality check.
trog 19 hours ago [-]
The question thst matters is: is he more productive than someone with a standard $30/mo Claude subscription? Or with no AI?
It's pretty obvious $200k/yr in tokens is a ludicrously huge spend. Trying to pretend it's not to show how leet you are is weird. But the interesting topic buried here is what value he's creating with such a spend and how it benchmarks against alternatives.
Because that is the value prop of AI. If it doesn't stand up against options when measured clinically and objectively then what's the point of it
lrvick 4 hours ago [-]
I cannot even comprehend how someone spends that much without realizing you can just buy GPUs.
cromka 4 hours ago [-]
Presumably they spend this much on frontier models. You don't run frontier models on GPU you "just buy".
lrvick 40 minutes ago [-]
You can run GLM 5.3 or Kimi K3 on maybe 40k in hardware these days.
anoojb 23 hours ago [-]
The common trope is, "keep an eye on what the rich want."
Gary is rich. Let's assume now, he spends between $10-20K/month USD on tokens. Whether he gets overwhelming tangible value out of that token spend is immaterial.
What he's getting out of it is a mix of social signaling value and actual perceived productivity. I don't know, but I doubt he has measuring the ROI on the token investment.
Much like someone gets various kinds of value out of a a personal tutor for their child. Or a high powered executive who continues to request budget for a "Chief of Staff" and related personnel in 2026.
The question is whether this is signal, or noise. Will everyone who has the budget aspire to spend money on tokens as a subtle way of projecting value with some tangible benefit? How long of a runway does this have?
Much like a Chief of Staff office, there's more to manage...but you you get some tangible benefit by having people to whom you can [in theory] delegate tasks and trust they will be completed. To some degree none of this matters, value is in the eye of the beholder.
What a time to be alive. Rich people continue to spend money with abandon, just on different things.
Edit: I work at technology company. If I were spending $10-20K/month USD in tokens I would immediately get flagged by Finance and receive a note from my EVPs Chief of Staff/Business Admin. My usage and prompts would get audited and I would be asked to create a report justifying the business impact of the spend.
rcko 23 hours ago [-]
Note, however, that Garry also holds a much longer lever in his hands. It's very likely that he does get a huge ROI from the token spend because if it helps him make just one slightly better investment decision per year, that's millions in return.
anoojb 22 hours ago [-]
The rich always have leverage. The questions I'm asking are broader market questions...
Does it matter that he's spending this much money on tokens? What's the derived signal?
Why isn't he spending more?! Spend on intelligence seems to have no limit, and he's rich after all right?
What's the limiting rate of return on token spend for the rich?
Does Gary represent an archetype of other rich people who also want to spend money on tokens at that rate? ...if not more?!
bix6 22 hours ago [-]
He has a vested interest in hyping AI / spend because YC overwhelmingly invests in AI startups.
I wouldn't say this was the norm, but I have burned through £1k of tokens in a couple of days before. And, on that same gig, I realised a test I was considering running would burn £2.5k worth of tokens over the course of a few hours so spent a couple of hours of my own time re-engineering it to get that down to £150.
The point is that I don't necessarily know that spending a shitload of money on tokens really tells us anything useful. If he's getting a return on that investment, fine, whatever.
Obviously most of us couldn't afford to spend that much ourselves, no matter how much value it might deliver, and many of our employers wouldn't be too happy if we spent it on their dime either. Equally, if you can, and it's delivering real value, then I don't see why you wouldn't?
details please, if it's interesting
You can do it if your employer has endless pockets.
Example: $30/1M input tokens (I don’t count output or cache, for simplification), you have several daily agents for everything => 3M tokens daily per agent * 10 agents ~> $900/day => $328k/year (API spending)
And, in fact, I got it from £2500 to £150 by careful model selection (I ran a trial over a small dataset with different models until I found the least expensive/capable model that would reliably do what I needed), using a local model for one step, and paring down the inputs as well so the token count plummeted.
They feel like they're super smart but it's just as smart as hiring a $500k/yr PhD as a PA to schedule your meetings and answer the phone. Sure it does a good job but you're just wasting money and abilities.
They feel super edgy and bleeding edge but all they really do differently is switching the AI chat from auto to the most expensive option all the time. When I see what they're using it for they're only grasping the basics of its abilities.
They're also people sending 2 page AI generated emails that make no sense for literally everything where a "Yeah sure let's do that" would have sufficed.
Unfortunately most of these are business leaders who just want to look edgy and "all in on AI" so we can't really limit their tokens. Yet they're also the ones saying we spend too much and want to limit some of our dev teams actually putting it to good use.
Management succumbed to AI psychosis, and so we were directed to show that we were adopting and using AI, but what we were using it for was not considered important. So we went out of our way to burn as many tokens as possible.
If you tell Claude to be inefficient and burn as many tokens as it can, it’ll happily oblige. It will also offer useful advice as to how to use even more tokens.
Once you factor in cost of electricity and depreciation of hardware, it is more expensive to run locally.
There are many arguments for running local but cost isn't one of them.
Let's do quick back of the napkin math.
The Studio M5 Ultra has a worklife of at least 3 years.Once you make the jump to agents managing agents the only limit is your budget and your harness's ability to coordinate. We'll all be using far more tokens than this in a year or two.
It sounds like a lot when I type it like that, but it's not actually that much if you delegate brutally. Like I said, I'm constantly token starved and could do a lot more!
Or are people paying $1K/mo (subsidized) just because AI is fun? Can you imagine a video game charging $1K/mo?
In Tan's case, I think the "fun" thing is the right way to look at it. He gets to play at being a big shot developer and show off on Twitter, so clearly he feels he's getting some value from his spending even if no-one else does.
There is a more serious question about to what degree the AI vendors have a duty of care to prevent addicts from getting themselves into financial trouble - perhaps they might in future face the sort of regulation applied to high street bookmakers. I doubt Tan would fail any putative affordability test, though, so it would still be up to him to seek help for his problem if he feels its getting out of control.
But you might have hit the nail on the head - yes, it's extremely fun and maybe that's the main reason after all. And FYI many, many people pay a lot more than $1k/m for video games.
This is my daily intrusive thought. "Why the hell am I doing any of this? Why am I not yelling louder at people that have so much and give so little?"
This website is full of people that could save millions of human lives in aggregate, but choose not to and actively argue against the idea because, and I do swear it's a damned curse at this point, "there is no incentive to do it."
I'm not even remotely rich. I have no "savings" or "investments". Yet I have a comfortable life in a house with access to food and care as I need. Sometimes I really hate it.
There are many people in this world who have more, view people as dirt, and dedicate their lives to keeping people down. Good, kind people who are disempowered by a system that betrayed us all are not at fault. Feel better man.
You can get access to DeepSeek 4.1 Flash for almost nothing, on a $1, $10, or $100 subscription. That can complement your 5 Claude accounts, or replace some. DS Flash 4.1 is so fast and cheap it’s ridiculous
I occasionally get my agents to do sensitive things; I'm not comfortable having my transcripts collected and sold, so I don't use them. They say they anonymize them but who can ever tell? Not for me, thanks.
With AI spend - especially at Anthropic/OpenAI API prices - spending 0.5-1 macbook a month worth of tokens per employee seems normal (from what I've seen, heard, what happens in this forum and thread and what my employer does).
Tokenmaxxxing (followed by commit/PR-maxxxing) is celebrated. Don't think - just launch orchestrator that will create dozen child agent sessions to do useless work to impress your superiors. If buy the end of the day you will ask to increase your token allowance - you're a superhero.
That's hardware.
https://en.wikipedia.org/wiki/I_Am_Rich
> “It’s too expensive right now,” he said, adding that he knows the amount of money he spends on AI agents is “a crazy privilege that not everyone can do.”
Sleep is one of the most important things in the world. Why not have the agents work while you sleep?
Even better, why not have the agents sleep for you?
EDIT: for those who say this is not realistic, I've had Claude sleep for 10 hours in just 30 seconds. Saved me an entire night.
[1] https://www.amazon.com/Million-Plus-Random-Digits-Symbols-Gu...
People in psychosis don't exactly weigh all of the options with a level head. He may not recognize that he has "cyber psychosis" or that his lack of sleep is a bad thing.
Don’t make me defend Garry Tan…
It was productive, but as you say, it's dangerous to gamble with sleep.
This is a difficult question to answer ! I would ask a related question which is equally hard. Has AI-use increased your human cognitive capability or decreased it ?
I accept this loss because the productivity is there. & hope to retire in a few years if things don't go south. For people entering the field I worry
what has he produced with that?
I don't think he has produced anything.
Tokenmaxxing was a dumb trend he has not figured out he's out of the loop yet.
Gary spending 250K/year on nothing is not strongly related to pros using $200/plan to do much better.
He's gonna hit the halting problem
2020s, techbros: "Here's why it's a huge problem for non-billionaires to have things."
Garry Tan is fully aligned with Elon Musk in going after MacKenzie Scott's charitable giving, making spurious arguments that she's doing it in the wrong way when they are all just running cover for the fact that she's one of the few at that level of wealth being charitable at all these days.
Silly woman, she should give her money away the right way, which I guess is by attempting to buy political power.
https://garryslist.org/posts/mackenzie-scott-s-26-billion-su... - The rant itself.
> This isn’t generosity. It’s negligence at scale.
Is when I realized I was wasting my time reading the article
Software? I'm not aware of anything.
"Garry, this is a text file".
https://www.youtube.com/watch?v=Q6nem-F8AG8
Evaluate the original post and response and build a high level learning plan with citations and references that covers each of the topics in the original post. Start the learning plan with an overview of effective prompting techniques with specific examples to research individual topics.
Large language models are usually priced on MTok or “million tokens”. You’ll often see LLMs priced at $X/MTok for input and $Y/MTok for output.
A harness is a program that interfaces with a large language model and provides it tools and context to achieve a goal. Codex and Claude Code are good examples of harnesses.
What exactly all of this accomplishes is not something I can answer for you :)
Tokens are what you get when you have to map a chunk of text to specific numbers, so that the LLM can actually work with it.
https://platform.openai.com/tokenizer
Reducetively speaking:
[Your text] --> [List of numbers (tokens)] --> [List of [List of numbers describing a concept/term/thing/job/item/etc] (embedding)]
That final thing (List of embeddings) is what gets fed into the LLM, so that it can produce (List of embeddings most likely to follow afterwards), which is then turned back into text.
> and their worth
Tokens are priced based on the model you're using. In general, expect to pay more when you need to use larger models. (Note: Exceptions exist, but treat them as exceptions)
The space as it is now prices tokens based on:
- how many tokens it needs to process (input),
- how many it needs to generate (output), and
- how often you send the same chunks of text that their system already has on file (cached).
> and harnesses
A harness is something that adds in tools & directions for the LLM, so that it can do its tasks. In a dumbed-down semi-visual summary:
---
[Text about what the LLM should be & what they're allowed/not allowed to do]
[Text about available tools & how to call them]
[Your text here]
---
The harness, if its part of a program/app, then looks at what the models generate in response, and does the (file writing/reading, tool calls, web searches, subagents, etc.), based on the formatted output that the LLMs should produce.
> and what exactly all of those things accomplish
Depends on the goal of the user. Some use them as assistants, most use it for coding.
https://openrouter.ai/rankings?utm_source=share&utm_medium=c...
This is insane. He's turning the failure of his AI assistants to do what he wants every single time and to do it in a safe manner into a positive thing. Failure as a positive!
That said, I approach AI with deep skepticism. I try to learn it, including its strengths and weaknesses. But I don't blindly trust it.
Coincidentally, my stock brokerage just yesterday sent an email reminding users that anything their agents do using the user's credentials is the responsibility of the user.
Since failure is such a likely outcome in a startup, "failure as a positive" is a mindset that a startup accelerator must propagate to keep founders going. And when you sell a message for long enough, you start to believe it yourself.
(Figures vary, see e.g. https://www.linkedin.com/posts/christham_4-weeks-on-claude-2... for $13,700/mo.)
If you hit your limits on two Max 20x plans, you could very well be exceeding $300k/yr in token use at current API rates.
> He describes using AI agents as getting help with “almost anything that you do in your day-to-day life. (...) Almost anything you want to do, it knows you, and it’s there to help. And it will actually sort of run ahead and say, ‘What about this?’ Or it’ll help you see around corners.”
Wow, impressive. Both the capabilities and how much he's willing to share.
Is it worth $17k per month? For me, absolutely not. But people spend more on luxurious cars and such, so if you've got money to spare...
> Y Combinator is also using an agent that is “rebuilding the way we operate at YC. We now have an internal AI agent, for instance, named Stray Light that kind of has a personality, and then it has a company brain. It actually has the total context of every founder we’ve ever backed, anyone who’s ever emailed us, like all the applications.”
Holy anthropomorphization. But besides that, what's the goal of doing this? Keeping a handy overview of when and why someone said something?
The only way I can imagine to burn that many tokens is just to run every task fully through the most expensive LLM with no real strategy. Like indexing emails should be a combo of scripts and small summarizing prompts I'd think.
Because they can. It’s as simple as that. To find meaning, to feel in control, to win. To dominate.
You'd be surprised how important such an overview can be in relationship-based businesses. People spend a lot of time trying to track down a vague memory they have of someone telling them something. CRM platforms are big for a reason, after all.
Not cloud costs, though; consumer costs. These agents must run on-device for privacy protections, because anything less will be a global non-starter. That means we need models that excel on phones, tablets, laptops, and desktops; we need home compute modules to run larger, “frontier”-class models for bigger tasks but without guzzling electricity (the Mac Mini’s power envelope is an ideal target); we need smart harnesses that can foist off inference to bigger models elsewhere on your private network of devices as needed without the user prompting for specific thinking modes or model names. All of this has to happen automatically, privately, securely, without leaking data to advertisers or surveillance companies.
It’s hard work, but it’s definitely the future of AI. I would argue the business case is solved with being able to build bespoke stacks of kit and run models through neocloud providers with ZDR policies. It’s the consumer angle that’s hard to crack, because everyone wants a repeat of Web 2.0 for ad-dollars instead of a privacy-centric modern tech stack that forces companies into lower revenue streams.
The future of personal assistants is most likely akin to Muse Spark, and yes I expect Meta is training on all of those interactions and using them to target ads.
I beg to differ.
> he’s suffered from “cyber psychosis” and sometimes barely sleeps
If your business model is taking money from gullible investors, maybe it's reasonable to appear (or actually be) incompetent so that you don't accidentally waste your time on people who haven't drunk the Kool-Aid.
That's basically gstack
Counter-point, they can't serve as a knowledgeable discussion partner, programmer, search engine, or your personal psychologist. Not saying that they should.
Also...
> Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting.
https://news.ycombinator.com/newsguidelines.html
It's pretty obvious $200k/yr in tokens is a ludicrously huge spend. Trying to pretend it's not to show how leet you are is weird. But the interesting topic buried here is what value he's creating with such a spend and how it benchmarks against alternatives.
Because that is the value prop of AI. If it doesn't stand up against options when measured clinically and objectively then what's the point of it
Gary is rich. Let's assume now, he spends between $10-20K/month USD on tokens. Whether he gets overwhelming tangible value out of that token spend is immaterial.
What he's getting out of it is a mix of social signaling value and actual perceived productivity. I don't know, but I doubt he has measuring the ROI on the token investment.
Much like someone gets various kinds of value out of a a personal tutor for their child. Or a high powered executive who continues to request budget for a "Chief of Staff" and related personnel in 2026.
The question is whether this is signal, or noise. Will everyone who has the budget aspire to spend money on tokens as a subtle way of projecting value with some tangible benefit? How long of a runway does this have?
Much like a Chief of Staff office, there's more to manage...but you you get some tangible benefit by having people to whom you can [in theory] delegate tasks and trust they will be completed. To some degree none of this matters, value is in the eye of the beholder.
What a time to be alive. Rich people continue to spend money with abandon, just on different things.
Edit: I work at technology company. If I were spending $10-20K/month USD in tokens I would immediately get flagged by Finance and receive a note from my EVPs Chief of Staff/Business Admin. My usage and prompts would get audited and I would be asked to create a report justifying the business impact of the spend.
Does it matter that he's spending this much money on tokens? What's the derived signal? Why isn't he spending more?! Spend on intelligence seems to have no limit, and he's rich after all right? What's the limiting rate of return on token spend for the rich? Does Gary represent an archetype of other rich people who also want to spend money on tokens at that rate? ...if not more?!