They are looking to lower it further to $100 [0]. This is definitely concerning for frontier labs since Accenture also caps their employees' usage to $29 a month [1].
Yes. The frontier models are good, indeed, but they are just far too expensive to use much. Recently I was showing my executive leader that we could simply run a certain set of tasks on the top frontier model and we'd be good to go. Oh but it would cost over 15M/year. We're a relatively small company, that's over two magnitudes more than what makes sense for us. So, never going to happen.
What remains then is an optimization exercise to bring the cost down. For months now we've been optimizing usage so we can run nearly all tasks on very cheap models to good success. It's fun to optimize and keeps the CFO from having a heart attack.
But not good news for the AI companies relying in a firehose of cash coming in from their frontier models.
$100 day should be more than enough. To exceed that you have to coming up with creative ways to waste it. You couldn’t hit that in normal development using AI all day.
For the edge cases that might need more like running a security audit, it would make more sense to submit that as a project proposal and have the expense approved.
We don't know how low it will get yet. The first article I linked also has this: "[m]any groups outside of engineering already face a $100 daily cap, but Apple has considered reducing that limit to as little as $25 for some teams, the people said."
They could be tightening the purse strings further until they hit some magical ROI numbers.
> $100 day should be more than enough. To exceed that you have to coming up with creative ways to waste it.
You do not need much creativity for that. Just forget to put build artifacts or node_modules in your agent's ignore list and it will happily burn through the cap in thirty minutes
Why is it "concerning"? Personally, I get a lot of stuff done on the $200/month subs. +$100/day per employee seems like a fairly significant amount of money and if a lot of companies do that (even a lot less than that) that seems entirely fine for frontier labs, given all the other uses as well.
That is not to say that you can't spend a lot more. And I would be surprised if Apple did not allow for that too. Note that the text says "for some of its own engineers", which could amount to a total nothing-burger, given the overall vagueness.
Depends on the sub you're using, but the anthropic one is giving you ~$12k/month [0], or $400/day. So you're already working at 4x the $100/day you consider significant.
Accenture lol, in that case, considering the quality of the output previously released by their humans, maybe local LLMs are already more than enough for that use case.
For smaller companies they can get Claude team plan for $20/month/seat and the plans give a lot of usage. We rarely hit limits. Codex plans used to have more usage and those were the go-to. Before that, GitHub Copilot CLI was the service with insane usage limits for $10-$20/month.
So it's about staying nimble enough so you can easily switch over. Don't tie yourself down to a single product/harness.
The Claude Team plan is subsidized, and pricing for them is significantly less than what enterprises pay with API pricing. For example, your $20/month plan would cost the equivalent of $400/month if they charged you using API pricing.
That's why those plans are only available to small companies with 150 employees or less.
Accenture isn't a giant software company though. For the sort of things their employees do, that might be a reasonable average (think chat, PowerPoint, a bit of research).
Yes, considering their CEO said that AI usage was a requirement for promotion earlier this year [0]. Companies are backpedaling now when the AI bills come out and bite them.
$52K per engineer per year extra cost to the company. I think at an Apple scale that level of ROI should be easy. I wonder if they were not seeing the ROI they wanted? As an indie hacker AI has been amazing for me and has freed me up to do so much more.
I’ve been talking to a bunch of companies about how they use LLMs and I’m seeing dramatic differences in costs, with very few concrete measures of value. Mostly just PR counts. We’re talking order-of-magnitude differences in spending, from hundreds up to many thousands. (API pricing, not subscription prices.)
I’m not surprised companies are pulling back. That’s a decent chunk of money and CFOs are going to want to know what they’re getting for the money. “More PRs” isn't a good enough answer.
That is peanuts to Apple. It is possible engineering and programming is getting too much done and Apple doesn't have the means to push it all out live to customers.
Or, get this - figuring out what customers really want and need is super hard and churning out code doesn't solve the truly hard problems of software development.
I don't see any indication that LLMs help with requirements gathering or user studies, especially live ones.
Yes, it's a significant expense even on top of an Apple salary.
Any company, not just Apple, should be tracking whether the extra spending is worth it. There must be a point of diminishing returns where additional AI usage stops increasing productivity at a comparable rate. That point should be your target. Different teams can have vastly different thresholds.
Companies also can't justify ever-increasing costs just because it increases productivity. It's business. You care about profits, not raw productivity. And Apple doesn't seem to be having a particularly great time these days.
It's only too low if the goal is to use AI as much as possible. If Apple are saying their engineers should not use AI for everything, and that they should still be working as they did but with a small assist from AI, the budget makes a lot of sense.
Giving people some constraints so they have to actually think about solving their problems properly rather than just throwing money (or tokens) at it is not a bad idea.
Apple is paying API prices, not subscription prices, though. $6000/month is half the engineer's monthly salary, it seems ridiculous to say that won't be enough, unless you're being very inefficient with token spend.
Yeah. I think people are reading $200 as "per month" and not "per day". I do a lot with unrationed Claude tokens and $6k/month is fine. Have I exceeded it? Yes. But it's very rare.
Things that drive up spend is mostly doing other people's work. Triage, "that doesn't seem quite right, let's see what happens if I prompt it", properly investigating incidents (I get a lot where the reporter's agent didn't read the code and just guesses at how it works, kind of making the report useless). Fixing those things is what would really get token spend down -- a consistent way of making bug reports against other teams that has the same context as the people that work on it everyday. I don't know if it's possible. But I'm anti-rationing because the first thing you do when your tokens are counted is to stop helping other people, right?
The fact that it's expensive has nothing to do with how useful it is though... I know plenty of people who max out their $200 plans every week, and they are working on stuff a lot less financially valuable than what Apple employees work on.
In my experience Claude Max x5 is the very reasonable amount of tokens for a work week.
But yeah, API pricing is crazy. Note that Apple might have negotiated a discount with Anthropic but is actually not allowed to share this discount details with its employees. So they communicate the budget based on public pricing.
It's kinda borderline. My dev workflow in my full-time job (Opus 5.5, 5 concurrent worktrees with human checking results and giving instructions) spend exactly $200/day on average. I imagine it could be more friendly if it's $300, as otherwise I might have to move some of my AI usage to weekends, which implicitly increases working hours :[
I’m not sure if the /usage in cc is directly comparable to this? It gives a $ value which I assume is API equivalent.
Most of the programming work I get opus5.5 to do, sometimes with fable coordinating. When I’m building a large feature it will often run over $500 and occasionally on longer sessions over 1k. Everyone else seems to think $200 should get by?
Just checked codexbar. I spend $1,000/mo in API credits across two $20/mo plans (claude and codex) just for side projects. I can't imagine having a API budget in this range for a day job, I'd run out every day
Do you work at a company that requires human code review for every change? Even if I tried to spend that much my work would just get logjammed in code review, so I intentionally don't generate output faster than I can get reviewers to read it.
My "AI subscription" is just buying 20$ openrouter credits when I run out and it's costing me less every month then a monthly Claude subscription. 200$ a day is insane
Your data is likely kept and used for training models.
For most businesses, that's a no go. At least at the company I work for, we have to wait for new models because they negotiate away the training and retention and the vendors are not too eager to give up their moat.
I also have the 20$ pro plan, I'm not reaching the limits of it... That said, I'm a one man company so I'm not just emptying a kanban board, I use Claude a lot as a thinking buddy, and my deliberations are the thing that limit the speed.
It seems the big AI companies are gouging enterprises with higher prices.
Claude tells me the retail price of the tokens I use on a personal plan outside of work. It’s way more than the plan costs me, by a factor of 50 or more. But at work I don’t see such discounts reflected in work’s estimates of my token spend. Something strange is going on.
200 daily means that you use 1 session in a turn-based setup. Hence likely you still know what is going on in the code.
Once you have multiple agents working in parallel, hands off the wheel becomes very attractive. And at that exactly point you lose completely touch of what the heck is going on in your ex-code.
Giving people an unlimited amount of anything means they have zero incentive to optimize using that thing. It's quite common that we complain here about Electron apps using 1GB for something as simple as a REST client. OTOH, many here keep calling for unlimited token budgets. I feel like capping will cause people to find better workflows, which we really need to in order to reduce (among others) the ecological impact of AI.
Finally companies are realizing that subsidizing unlimited AI for your employees is basically just paying for the tools for them to work 2hrs a day while you pay them for 8.
Listen, I dont care if employees scam their bosses by offshoring their work to AI, but its absolutely the case that 90% of people irresponsibly use AI to be lazy, get by doing less work.
Unfortunately, as Ive personally witnessed when you do this and take away the cheating tools engineers just start copying and pasting IP into their own AI subs where you have less garuntees about ZDR. Employees will gladly spend an extra $200 a month so they can work 4hrs less a day and just get by.
Only 10%, maybe less, are actual super contributors with AI. The rest are literally getting a form of diabetes of the brain and the exercise required to reverse that disease is quite uncomfortable, especially when you've been hands off using a cheating tool for a year straight.
I wont argue with you there, I dont really care about the companies at all
Im just saying employers will notice that its basically being used as a cheating tool by a large subset of users and they 'll stop subsidizing that behavior.
What Im really trying to say is its not a producitivity enhancer for an enterprise, maybe for a small percentage of employees, so I dont really see the value in it as an enterprise wide investment.
Makes it a whole lot worse doesn't it?
My Fedora boots under a second in my extremely old and cheap HP. Meanwhile MacOS introducing new bugs and Liquid Glass that hung my m1 air
They are looking to lower it further to $100 [0]. This is definitely concerning for frontier labs since Accenture also caps their employees' usage to $29 a month [1].
[0] https://finance.yahoo.com/technology/ai/articles/apple-cuts-...
[1] https://finance.yahoo.com/technology/ai/articles/accenture-j...
> This is definitely concerning for frontier labs
Yes. The frontier models are good, indeed, but they are just far too expensive to use much. Recently I was showing my executive leader that we could simply run a certain set of tasks on the top frontier model and we'd be good to go. Oh but it would cost over 15M/year. We're a relatively small company, that's over two magnitudes more than what makes sense for us. So, never going to happen.
What remains then is an optimization exercise to bring the cost down. For months now we've been optimizing usage so we can run nearly all tasks on very cheap models to good success. It's fun to optimize and keeps the CFO from having a heart attack.
But not good news for the AI companies relying in a firehose of cash coming in from their frontier models.
$100 day should be more than enough. To exceed that you have to coming up with creative ways to waste it. You couldn’t hit that in normal development using AI all day.
For the edge cases that might need more like running a security audit, it would make more sense to submit that as a project proposal and have the expense approved.
Most days I use <$100, but very often will go to $200 with $1000 happening at least once or twice a month.
It's easy to spend a lot when writing an entire service, large features, or doing research loops.
Those special cases should be handled with overages and special approvals. Blanket approvals of massive spending doesn't make financial sense.
We don't know how low it will get yet. The first article I linked also has this: "[m]any groups outside of engineering already face a $100 daily cap, but Apple has considered reducing that limit to as little as $25 for some teams, the people said."
They could be tightening the purse strings further until they hit some magical ROI numbers.
> $100 day should be more than enough. To exceed that you have to coming up with creative ways to waste it.
You do not need much creativity for that. Just forget to put build artifacts or node_modules in your agent's ignore list and it will happily burn through the cap in thirty minutes
Why is it "concerning"? Personally, I get a lot of stuff done on the $200/month subs. +$100/day per employee seems like a fairly significant amount of money and if a lot of companies do that (even a lot less than that) that seems entirely fine for frontier labs, given all the other uses as well.
That is not to say that you can't spend a lot more. And I would be surprised if Apple did not allow for that too. Note that the text says "for some of its own engineers", which could amount to a total nothing-burger, given the overall vagueness.
Depends on the sub you're using, but the anthropic one is giving you ~$12k/month [0], or $400/day. So you're already working at 4x the $100/day you consider significant.
[0]: https://newsletter.semianalysis.com/p/anthropic-subscription...
Accenture lol, in that case, considering the quality of the output previously released by their humans, maybe local LLMs are already more than enough for that use case.
They should consider buying or renting their own GPUs to run one of the cheaper open models instead of paying the labs.
They also should have a smart model supervise the cheaper and less capable ones.
cheaper per token or by effect
29/month is offensively low
For smaller companies they can get Claude team plan for $20/month/seat and the plans give a lot of usage. We rarely hit limits. Codex plans used to have more usage and those were the go-to. Before that, GitHub Copilot CLI was the service with insane usage limits for $10-$20/month.
So it's about staying nimble enough so you can easily switch over. Don't tie yourself down to a single product/harness.
The Claude Team plan is subsidized, and pricing for them is significantly less than what enterprises pay with API pricing. For example, your $20/month plan would cost the equivalent of $400/month if they charged you using API pricing.
That's why those plans are only available to small companies with 150 employees or less.
$420 on the Enterprise plan, you pay 20 bucks just for the seat.
Accenture isn't a giant software company though. For the sort of things their employees do, that might be a reasonable average (think chat, PowerPoint, a bit of research).
Yes, considering their CEO said that AI usage was a requirement for promotion earlier this year [0]. Companies are backpedaling now when the AI bills come out and bite them.
[0] https://finance.yahoo.com/news/accentures-ceo-says-using-ai-...
$5600 per 28-day period is quite a lot!
Could be $4,400–about 22 weekdays in a month. Still a lot.
$52K per engineer per year extra cost to the company. I think at an Apple scale that level of ROI should be easy. I wonder if they were not seeing the ROI they wanted? As an indie hacker AI has been amazing for me and has freed me up to do so much more.
I’ve been talking to a bunch of companies about how they use LLMs and I’m seeing dramatic differences in costs, with very few concrete measures of value. Mostly just PR counts. We’re talking order-of-magnitude differences in spending, from hundreds up to many thousands. (API pricing, not subscription prices.)
I’m not surprised companies are pulling back. That’s a decent chunk of money and CFOs are going to want to know what they’re getting for the money. “More PRs” isn't a good enough answer.
> That’s a decent chunk of money and CFOs are going to want to know what they’re getting for the money.
Finally! Gravity and sanity are starting to win.
That is peanuts to Apple. It is possible engineering and programming is getting too much done and Apple doesn't have the means to push it all out live to customers.
Most likely they see no meaningful return on all the money spent on tokens.
Why is that most likely?
LOL, I appreciate the boundless optimism.
Or, get this - figuring out what customers really want and need is super hard and churning out code doesn't solve the truly hard problems of software development.
I don't see any indication that LLMs help with requirements gathering or user studies, especially live ones.
Yes, it's a significant expense even on top of an Apple salary.
Any company, not just Apple, should be tracking whether the extra spending is worth it. There must be a point of diminishing returns where additional AI usage stops increasing productivity at a comparable rate. That point should be your target. Different teams can have vastly different thresholds.
Companies also can't justify ever-increasing costs just because it increases productivity. It's business. You care about profits, not raw productivity. And Apple doesn't seem to be having a particularly great time these days.
People here are completely clueless if they think an expense of ~50k/y per employee is peanuts.
They vibe coded some slopware mobile app on a $200 Claude subscription and roleplay as that being productive.
In reality those companies are measuring what the extra cost brought in as a return, and are likely deciding that it was not worth it.
I fully expect this to become the norm, figuring out ways to extract usefulness out of LLMs while spending as little as possible.
If $200 in API price equivalent it's too low. That's less than Claude Max x5 plan gives. So with $100 plan you'd have more tokens than apple engineer
It's only too low if the goal is to use AI as much as possible. If Apple are saying their engineers should not use AI for everything, and that they should still be working as they did but with a small assist from AI, the budget makes a lot of sense.
Giving people some constraints so they have to actually think about solving their problems properly rather than just throwing money (or tokens) at it is not a bad idea.
Apple is paying API prices, not subscription prices, though. $6000/month is half the engineer's monthly salary, it seems ridiculous to say that won't be enough, unless you're being very inefficient with token spend.
Yeah. I think people are reading $200 as "per month" and not "per day". I do a lot with unrationed Claude tokens and $6k/month is fine. Have I exceeded it? Yes. But it's very rare.
Things that drive up spend is mostly doing other people's work. Triage, "that doesn't seem quite right, let's see what happens if I prompt it", properly investigating incidents (I get a lot where the reporter's agent didn't read the code and just guesses at how it works, kind of making the report useless). Fixing those things is what would really get token spend down -- a consistent way of making bug reports against other teams that has the same context as the people that work on it everyday. I don't know if it's possible. But I'm anti-rationing because the first thing you do when your tokens are counted is to stop helping other people, right?
The fact that it's expensive has nothing to do with how useful it is though... I know plenty of people who max out their $200 plans every week, and they are working on stuff a lot less financially valuable than what Apple employees work on.
But does the spending on the ai result in more revenue for the company?
I know people who spend that on lottery tickets. Obviously rational.
I don't think you actually know anyone spending $10,000/yr on lotto tickets.
But it's telling that you claim to while treating this as some sort of equivalence.
I think engineers in the Bay area are paid more than 12k / month. Maybe it's 1/3 of their salary :)
Imagine if they let you take home half of whatever money you save from not abusing AI tokens. That'd make people more efficient.
"Apple is paying API prices, not subscription prices, though. $6000/month"
Smart guy showed up. Yes. Don't do this.
In my experience Claude Max x5 is the very reasonable amount of tokens for a work week.
But yeah, API pricing is crazy. Note that Apple might have negotiated a discount with Anthropic but is actually not allowed to share this discount details with its employees. So they communicate the budget based on public pricing.
It's kinda borderline. My dev workflow in my full-time job (Opus 5.5, 5 concurrent worktrees with human checking results and giving instructions) spend exactly $200/day on average. I imagine it could be more friendly if it's $300, as otherwise I might have to move some of my AI usage to weekends, which implicitly increases working hours :[
It's 200 per day.
$200 a day is not that bad. probably more than the majority of companies?
Yeah, mine is $500/mo currently, but I have the ability to request more if I can justify it.
Yeah, my cap is higher but I don’t use it.
I like the term “brute-force inference” because it describes the state of the art accurately.
I’m not sure if the /usage in cc is directly comparable to this? It gives a $ value which I assume is API equivalent.
Most of the programming work I get opus5.5 to do, sometimes with fable coordinating. When I’m building a large feature it will often run over $500 and occasionally on longer sessions over 1k. Everyone else seems to think $200 should get by?
I’m looking for a point of reference here.
Just checked codexbar. I spend $1,000/mo in API credits across two $20/mo plans (claude and codex) just for side projects. I can't imagine having a API budget in this range for a day job, I'd run out every day
Do you work at a company that requires human code review for every change? Even if I tried to spend that much my work would just get logjammed in code review, so I intentionally don't generate output faster than I can get reviewers to read it.
I use claude code $20 plan for the whole month, $200 is a whole year of that.
Enterprises don't get to pay that plan. That's the consumer option. Businesses pay extra.
My "AI subscription" is just buying 20$ openrouter credits when I run out and it's costing me less every month then a monthly Claude subscription. 200$ a day is insane
Your data is likely kept and used for training models.
For most businesses, that's a no go. At least at the company I work for, we have to wait for new models because they negotiate away the training and retention and the vendors are not too eager to give up their moat.
That's their moat: your data.
Factually incorrect https://openrouter.ai/docs/guides/features/zdr
It’s difficult to be sure here, though. Can happen either accidentally through a rogue CSAM filter or intentionally w/o OpenRouter knowing.
No, legally incorrect, or perhaps better: "based on my reading of the tos..."
factually would imply it's impossible for your data to ever be used inappropriately.
As long as you have under 150 users you can get the cheaper 25 dollar per seat plan.
My company pays Claude half than consumer price because we are not for profit.
Subscriptions are subsidized, so they give you a lot more than what you get with API. It is at least an order of magnitude, if not more.
But if you got access to all the latest models, through your org, without limits to spend, you'd likely not stay at $20 for even a day.
I also have the 20$ pro plan, I'm not reaching the limits of it... That said, I'm a one man company so I'm not just emptying a kanban board, I use Claude a lot as a thinking buddy, and my deliberations are the thing that limit the speed.
Good for you but $20 is too limited, a review of a code base and it’s gone.
Oh… my sweet summer child. Hang onto that non-enterprise pricing for as long as you can.
I don't use that much to get my job done, but I know some of my friends working in a China's AI company and they use more than $300 per day
FTA (emphasis added: Apple has reportedly slashed the daily AI spending limit for SOME of its own engineers, dropping the cap from $300 down to $200
TIL "FTA" means "from the article"
Do we know if Apple is working on their own SWE LLM? IIIRC they bought the people behind https://github.com/exo-explore/exo at the very least?
You'd think they must be investing heavily into "secret" agentic engineering dogfooding at this point.
It seems the big AI companies are gouging enterprises with higher prices.
Claude tells me the retail price of the tokens I use on a personal plan outside of work. It’s way more than the plan costs me, by a factor of 50 or more. But at work I don’t see such discounts reflected in work’s estimates of my token spend. Something strange is going on.
How's this part even possible? "Non-English prompts use up to 179 times more energy."
Does anyone know that study? I mean, I can understand other languages takes more tokens but that amount of energy? ...
It can't be true. Most of the energy goes for reasoning and all my reasoning prompts are translated and done in English then translated back.
I got ~3.5 billion a year napkin calculation (if internet stats of a 66,000 engineer count is remotely true).
Yet $200 a day per engineer still sounds like a pretty generous budget to me
if you buy 15 subs each to claude and codex that's a lot of code. not so much if it's an API per token limit.
I have been able to soak up a weeks worth of tokens in 6 hours with astra ultra if I'm not very explicit and bounded in my requests.
Brute force machines are expensive...
200 daily means that you use 1 session in a turn-based setup. Hence likely you still know what is going on in the code.
Once you have multiple agents working in parallel, hands off the wheel becomes very attractive. And at that exactly point you lose completely touch of what the heck is going on in your ex-code.
what's the source? ok i know everything is nowadays ai generated but come on?
They could always use Siri for coding, it's an idea?
$200 per workday? That’s still a lot
That's a lot of tokens. I burn Opus about half the day and reach maybe half of that at most.
So if you burned Opus the whole day…
A push in the right direction.
Tokens are a resource. We should use as few as possible (Tokengolfing?).
Giving people an unlimited amount of anything means they have zero incentive to optimize using that thing. It's quite common that we complain here about Electron apps using 1GB for something as simple as a REST client. OTOH, many here keep calling for unlimited token budgets. I feel like capping will cause people to find better workflows, which we really need to in order to reduce (among others) the ecological impact of AI.
I'm curious, what laptops do apple engineers get and are they encouraged to try local LLMs?
at this scale, it is totally worth for them to self-host models for part of the work.
Finally companies are realizing that subsidizing unlimited AI for your employees is basically just paying for the tools for them to work 2hrs a day while you pay them for 8.
Listen, I dont care if employees scam their bosses by offshoring their work to AI, but its absolutely the case that 90% of people irresponsibly use AI to be lazy, get by doing less work.
Unfortunately, as Ive personally witnessed when you do this and take away the cheating tools engineers just start copying and pasting IP into their own AI subs where you have less garuntees about ZDR. Employees will gladly spend an extra $200 a month so they can work 4hrs less a day and just get by.
Only 10%, maybe less, are actual super contributors with AI. The rest are literally getting a form of diabetes of the brain and the exercise required to reverse that disease is quite uncomfortable, especially when you've been hands off using a cheating tool for a year straight.
Companies deserve it and then some.
I wont argue with you there, I dont really care about the companies at all
Im just saying employers will notice that its basically being used as a cheating tool by a large subset of users and they 'll stop subsidizing that behavior.
What Im really trying to say is its not a producitivity enhancer for an enterprise, maybe for a small percentage of employees, so I dont really see the value in it as an enterprise wide investment.
So they are paying approximately $6k/engineer-month.
That's a maximum/cap, not an average.
Who said it was an average?
So that’s a yearly budget cut from $110k down to $73k, correct? I could live with that. I’d have a hard time finding ways to burn all these tokens.
Presumably they mean work days rather than all days of the year
How many engineers at Apple are on-call?
And that's assuming they hit the limit every single day
If you work 365 days a year which hopefully you don’t…
$ 200 a month and yet new releases of Mac are not upto the mark. Very strange
$ 200 daily, not month.
Crap. 200 a month!!! They better rewrite MacOS in Rust then lol
$200 a day
Makes it a whole lot worse doesn't it? My Fedora boots under a second in my extremely old and cheap HP. Meanwhile MacOS introducing new bugs and Liquid Glass that hung my m1 air