What is this, Claude for ants?
It brings me joy to know that corporations are trusting other corporations to not rip each other off. Wage theft is more lucrative than whatever product or service being sold. As if AI companies exists to help anyone other than themselves . The only way this works is if the largest paying companies work out some kind of kick back, pyramid, price fixing scheme. I really appreciate Claude for how it’s assisted me in fixing things but I know that someday AI will exist solely for the rich.
Accenture is a technology consulting company. Nobody needs technology consultants with AIs around, because AIs are more capable, much faster, and much cheaper than almost all consultants. Their standard hourly rates are typically in the same ballpark as AIs’ highest subscription monthly rates.
Therefore, as their sales start drying up, Accenture is starting to save money wherever it can before it needs to start firing people.
If AI makes you twice as productive, and there’s fewer taxes on spending money for tokens than there is for spending money on salaries, the rational amount for a corporation to spend on it is more than a human’s entire salary, per person working with it.
Naively assuming corporations approximate rational actors, and AI does measurably improve productivity, this implies a gain of roughly 2% if these employees are earning federal minimum wage, or well below 1% if they earn average software developer wages.
Maybe their contract deliverables are poorly suited to this use case or something, but if they are given a fixed amount of money to produce some software item, then it makes sense to prioritize productivity per dollar.
How many companies allow their employees to also be employed by their competitors?
Your idea of an AI being like an employee only if there’s an AI company that will agree to an exclusive contract or at the very least respect data ownership and respect privacy. Is there an AI company that’s like that?
Data ownership and privacy absolutely, within an enterprise license. They very deliberately make you pay for that.
While it’s hard to prove they’re not just breaching contract and training on it anyway, I don’t think it makes sense to assume they are.
@prime_number_314159
Maybe their business model does not depend on productivity. If the rate they charge for a developer with AI and without AI is approximately the same, they rather sell two people than making one more productive.
@FirmDistributionGenerally in the consulting business you’d just have your employees bill more hours than actually worked if they’re being more productive than average. Obviously without telling your clients.
For government contracts it’s not unusual that the agreed upon hours estimate is what is billed regardless of whether you actually spend more or less time.
“You must use AI!”
“No, no, no, no….not like that!”
Just trying to shoehorn profit into a system that was never going to deliver one in the first place.
I didn’t realize the business plans were token metered differently from the Pro plan I’ve been on for a while. $20/month, and it has session/weekly limits, but I rarely hit them, and I use it, a lot.
I do wonder if this will encourage businesses to branch out into running local models for simpler tasks.
You can get plans as a business too, but then it’s per user, so if your plan has $200 worth of “usage” per user and one uses $0, that does not mean another can use $400 and you still have to pay for the non-user’s $200, or you can use API pricing to pay for exactly what your users use, which for OpenAI and Anthropic is egregious. You get like 10x more per dollar with the Chinese models. And some of those providers ALSO have plans that make it much cheaper.
Why even bother at all? What can you do with those tokens? Ask for the weather twice?
The Haiku 5.5 model is actually quite inexpensive. I’ve used it to implement multiple features this month and have only spent $4 of my $20 monthly budget.
Are you using api or subscription?
Technically neither - I use it via Microsoft Co-pilot subscription, and it’s a la carte for model usage from every provider.
Ah co pilot was much worse than claude code when I used it. Subscription is far far better the api prices are outrageous
Microsoft’s copilot branding is all over the place. Copilot CLI and GitHub copilot (IDE plugin) both let you choose which model you want to use.
Yes but not using claude code as a harness results in worse use out of their models from my testing
Let’s not forget Copilot is also an office suite lol
Everything is copilot. No more windows, copilot is my operating system. Microsoft has changed its name to Copilot, Inc., and Satya Nadella has updated his CopilotedIn profile name to Copilot Copilot. Copilot is my copilot.
The legacy code I’m managing is a clusterfuck of the highest order. Even Sonnet struggles to analyze it sometimes 😑
I’m happy this is only a small part of my job, but oh my…
Same. AI has fallen down repeatedly trying to analyse both of my repos. And they were written by a bad human originally. The problems are all the same, except AI will get you spaghetti faster.
Messy brownfield code requires a lot more meta-analysis from previous sessions, just to get all of the spaghetti code into a manageable set of rules and documentation.
Don’t just send the agent blindly into the code. Every session, the agent is just going in fresh without any context on how the code works. Break the codebase down to a 10,000 foot view as a summary, and then get more documentation and agent rules for each of its major sections. Good tests help a great deal, too, and if there aren’t tests, start building them. Start creating skills just for the codebase, or any common tasks you’re asked to do against it.
Also remember that you don’t have to write any of this stuff. Have it write the docs, rules, skills, tests, and just review it to make sure it looks right.
That is a pitfall that is difficult to mitigate. When using these models your own code base is the primary input, and it will match the style. Garbage in, garbage out. It may be worthwhile to refactor prior using a model. Even having intelligible method and variable names can make a difference. Maybe do a sweep of simple renaming, and adding supplemental explanatory comments, then you might get better results?
The $29 figure is doing a lot of quiet work in this story, because that’s exactly what a Pro plan costs. So the real message isn’t really a budget cap, it’s ‘stop expensing API tokens and expense a subscription instead’. I’d bet a chunk of that $1,000 spend was people wiring API access into internal tooling and prototypes, which feels productive right up until finance sees the invoice and has no idea what any of it maps to. The part that’ll actually sting is that this is how a lot of shadow automation dies: not because it didn’t work, but because nobody could explain the line item. At least a hard cap forces the conversation that should’ve happened months earlier.
But anthropic and the others don’t give subscriptions to enterprises, only tokens at full API prices
Not quite true, for low seat organizations there’s a fixed rate Teams plan
Yeah, but Accenture is anything but a low seat org.
If you’re spending $1K per month on LLMs – putting aside whether this is productive or ethical for a moment – wouldn’t it make more sense to just pony up the $5K-$10K or so that it would cost to provision a private, local model? Might that not actually be superior in terms of corporate security?
Not at the moment. It’s definitely cheaper to buy LLMs as a service than running an in-house server hall. Anthropic and OpenAI are running at a loss, you know.
Now, risking to lose your data to the service provider is another thing.
Look at it this way, if I pay Open AI $1k a month, their people make sure it works they month. If I let Joe from finance spend $10k on computer parts, Joe needs to ensure that it runs or I need to hire someone who is knowledgeable on local models. If I’m a simple firm, $12k a year might be cheaper because someone else makes sure it works and I can fire Joe when the fancy autocorrect loses a bunch of clients by falsifying work. Then I don’t have to spend even bigger bucks hiring a new Joe because no one knows how to work the damn server he replaced the break room with.
It depends, sometimes the private API models are better at a specific task to the point it wouldn’t be worth using one that is worse. (Really it’s the harnesses around them tbh)
Overall sovereign AI is where it’s at. If you have good enough tooling in place
Definitely. We do this at my job. But a server big enough to provide a comparable service would cost at least $30k. And you likely need more than one. And you still need someone to maintain it.
I’ve heard that GPU’s for this work are exceedingly hard to get ahold of as well. Glm 5, which i think is near frontier model capabilities, takes around 1.5 Tb of vram un-quantized.
Not hard. Just really fucking expensive and might have a long lead time.
So hard, then.
Amateurs.
I’m spending 4000$ per month.
The title is misleading. They didn’t cap it, they just declared that if you are spending more than that, you have to be able to explain what you’re spending it on and why, giving it a business reason and a project to associate the costs to, which I expect almost all Claude users at Accenture probably can. This is simply an accounting change, not a company-wide shutdown of Claude.
Heavy use remains possible, but it now needs a budget owner.
explain what you’re spending it on and why, giving it a business reason and a project to associate the costs to
I love these chilling effects
So only if it can be recharged to a customer
I mean, that’s literally Accenture’s default business model (same as almost any consulting/professional services company), so it shouldn’t be surprising. My point is that it’s not an indictment of AI, and their AI usage isn’t actually going to go down; it’s just Accenture doing Accenture things. They’re not a charity, they’re not using AI for fun, they’re using it to make profit, and they make profit by charging their customers for every cent of AI they use on their behalf, usually with a hefty markup, in addition to everything and anything else they can justify charging against their customer’s accounts. It’s a very profitable business model, and I’m sure they love using AI for it and will continue using AI just as heavily if not even more in the future.
Ten times over.
$29 is enough for one of them pro plans. That’s plenty if you’re not wasteful.
But do companies like Accenture get that subsidized plan at scale? At API rates that can easily be thousands of dollars.
Their website is a little unclear there.
There’s a team plan for 2-150 person orgs, above that you have to go with Enterprise, where apparently the seats are 20$/month billed annually, 20 seat minimum + API usage costs, and it says “All Team plan features, plus:” (a bunch of enterprise-focused things), so that to a reasonable person would suggest that the users would still get the same subsidized usage until they’ve used up their rates and after that it’s API pricing. Of course what it suggests to a reasonable person may not be what it actually offers because it’s an American company and good fucking luck making those adhere to rules about misleading advertisement and whatnot
Well duh? I doubt $29 get you anywhere at the moment and that nonsense is only going to get more expensive.
For now, that’s not crazy. I have a company-provided api key and subscription. Now maybe they bumped me to the a $100/mo tier without telling me, but as far as I know the lowest tier keeps me in tokens heat about all month and I use it a lot. I switch over to API for a few hours per month and probably spend more in that time than my sub costs all month.
I spent about $2k developing a set of tools and AI orchestrates them for me, and I can add new tools to orchestrate as needed. Caveat: I do more production support and my teams do the production development work. We are all on subscription plans, but they do more hand development than I do.
Point being, you can easily spend >$1k/mo and me and several people on my teams did before we got subscriptions, but the subs give some AI assistance without going all-in on AI coding, which is the sweet spot IMO.
That’s basically saying that they shouldn’t use Claude at all.
Good.
