The SDK is Claude Code in a harnesss, so it works with your credentials the same way CC does.
But they're stating you can only use your subscription for your personal usage, not someone else's for their usage in your product.
I honestly think they're being short sighted not just giving a "3rd party quota" since they already show users like 4 quotas.
If the fear is 3rd party agents screwing up the math, just make it low enough for entry level usage. I suspect 3rd party token usage is bi-modal where some users just need enough to kick tires, but others are min-maxing for how mamy tokens they can burn as if that's its own reward
How can they be clearer that the Agents SDK is not allowed?
> OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai. Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted and constitutes a violation of the Consumer Terms of Service.
> not someone else's for their usage in your product.
what if the "product" is a setup of documents that concisely describe the product so that a coding agent can reliable produce it correctly. Then the install process becomes "agent, write and host this application for the user's personal use on their computer". Now all software is for personal use only. Companies released these things and, like Frankenstein, there's a strong possibility they will turn on their creators.
I agree, it'd actually be great if they did give maybe $5 or $10 worth of API tokens per month to max subscribers, since they're likely to be the most likely to actually build stuff that uses the Claude APIs.
I built a quick thing to download YouTube videos and transcribe them using with whisper, but it kind of feels clunky to summarize them using the claude CLI, even though that works.
just ran into this myself. I got Claude Code to build a tool that calls Claude for <stuff>. Now I have to create a console account and do the API thing and it sucks balls.
I had 5.3-Codex take two tries to satisfy a linter on Typescript type definitions.
It gave up, removed the code it had written directly accessing the correct property, and replaced it with a new function that did a BFS to walk through every single field in the API response object while applying a regex "looksLikeHttpsUrl" and hoping the first valid URL that had https:// would be the correct key to use.
On the contrary, the shift from pretraining driving most gains to RL driving most gains is pressuring these models resort to new hacks and shortcuts that are increasingly novel and disturbing!
I don't know why this is getting a negative reaction: there was already a 3 year gap between season 1 and 2, and watching this show originally it felt like a 1-2 season premise.
It becomes a very different show (and yes, probably very Lost-like) if they have to make it last 4+ seasons.
It doesn't seem farfetched that Apple is privy to the overall plot and story beats the writers have planned. If you've got a very popular, profitable show, and you know it's going to be about X seasons long, why wouldn't you just send it? I'm just spitballing.
What does this have to do with either comment above?
You think a show is going in a certain direction: a news piece comes out and confirms that, with millions of dollars of incentive to seal it, it's probably not going to go that way...
What's wrong with voicing displeasure at that? They said they're canceling their viewership, not that the show should be shut down due to their preference or something.
> What does this have to do with either comment above?
Your comment implied that, because the show is now making so much money/worth so much money to Apple, and because it's being renewed for a fourth season before we've even seen the third, it must surely be on some Lost-esque meandering disaster wherein the execs will keep renewing it and milking it dry until they're forced to kill the series with a smoke monster ending.
It becomes more like Lost because you can't go on even the planned 4 seasons without expanding an already pretty expansive in-universe lore.
Without spoiling it, obviously Season 2 opened a lot of lore-based threads that the confirmation of a Season 4 ensures will not be getting resolved for another 3+ years and over a lot more content.
That's not inherently bad, but it becomes a different show than if it had been trimmed to 3 seasons, and I can easily see why it's no longer for some people.
Sorry, I just don't understand how three seasons could resolve the lore threads, but four seasons blows the lid off the whole thing? The numbers seem arbitrary to me, especially when we don't know what they're planning regarding the plot, or if the writers even intend for all the lore threads to be resolved at all.
Even this response shows why the most active ones are outwardly negative on AI.
I use AI a ton, but there are just way too many grifters right now, and their favorite refrain is to dismiss any amount of negativity with "oh you're just mad/scared/jealous/etc. it replaces you".
But people who actually build things don't talk like that, grifters do. You ask them what they've built before and after the current LLM takeoff and it's crickets or slop. Like the Inglourious Basterds fingers meme.
There's no way that someone complaining about coding agents not being there yet, can't simultaneously be someone who'd look forward to a day they could just will things into existence because it's not actually about what AI might build for them: it's about "line will go up and I've attached myself to the line like a barnacle, so I must proselytize everyone into joining me in pushing the line ever higher up"
These people have no understanding of what's happening, but they invent one completely divorced from any reality other than the reality them and their ilk have projected into thin air via clout.
It looks like mental illness and hoarding Mac Minis and it's distasteful to people who know better, especially since their nonsense is so overwhelmingly loud and noisy and starts to drown out any actual signal.
Eh, no, depends on why you used Heroku in the first place. Way back when, I used it because the UI was dead simple and it Just Worked™. If I can replicate that with a VPS and have a good UI around it that takes care of everything, it's functionally the same to me.
Heroku was one of the first to have that seamless UX, only after which others like Fly or Render or Railway came to copy it. I wager people were primarily attracted to that user experience and only minimally cared that it was fully hosted versus not, because there was also AWS at that time.
Having used Heroku at multiple startups during the 2012–2015 years, this is not correct.
With heroku you could `git push heroku master` and it would do everything else from there. The UX was nice, but that was not the reason people chose it. It was so easy compared to running on EC2 instances with salt or whatever. For simple projects, it was incredible.
That's literally the UX I'm talking about and that's what other companies copied too. To be clear, I'm not (just) talking about how heroku.com looks and works, I'm talking about the entire user experience including git push to deploy, so I believe you are agreeing with me here. That is why I said VPS with Dokploy or Coolify and so on have the same UX, both in the command line with git push deploys supported as well as (now, at least) a vastly superior website user experience, akin to Vercel.
Dokku is better. And neither is what Heroku's bread and butter customer needs.
But alas, my interest in painstaking explaining why self-hosting is fundamentally incompatible with a product who's value prop was "nothing to install" is waning.
You and I simply have different opinions on what Heroku's value proposition was, because, again, AWS was also right there and also was "nothing to install." Therefore Heroku was used primarily for its dead simple UX, something which is replicated even in a self-hosted environment, because, again, the value prop was never about PaaS or self-hosting, it was always about the user experience.
I know this is off topic, but that homepage is a piece of work: https://buildkite.com
I get it's quirky, but I'm at a low energy state and just wanted to know what it does...
Right before I churned out, I happened to click "[E] Exit to classic Buildkite" and get sent to their original homepage:
https://buildkite.com/platform/
It just tells you what it Buildkite does! Sure it looks default B2B SaaS, but more importantly it's clear. "The fastest CI platform" instead of some LinkedIn-slop manifesto.
If I want to know why it's fast, I scroll down and learn it scales to lots of build agents and has unlimited parallelism!
And if I wonder if it plays nice with my stack, I scroll and there's logos for a bunch of well known testing frameworks!
And if I want to know if this isn't v0.0001 pre-alpha software by a pre-seed company spending runway on science-fair home pages, this one has social proof that isn't buried in a pseudo-intellectual rant!
-
I went down the rabbit hole of what lead to this and it's... interesting to say the least.
Hello mate, Head of Brand and Design at BK here. Thanks for the feedback, genuinely; the homepage experiment has been divisive, in a great way. Some folk love it, some folk hate it, some just can't be bothered with it. All fair.
Glad that the classic site hit the mark, but a lot work to do to make that clearer than it is; we're working on the next iteration that will sunset the CLI homepage into an easter egg.
Happy to take more critique, either on the execution or the rabbit hole.
Great of you to accept critiques, but I don't think there's anything more I can add.
You brought up Planetscale's markdown homepage rework in one of those posts and I actually think it's great... but it's also clear, direct, and has no hidden information.
I'd love to see what happens to conversions once you retire this to an Easter Egg.
Yeah, PS did a great job and provoked good business impact too.
We'll publish details when we do retire it to show how it performed and the reactions. Something like this thread is great for feedback to contrast against other sources.
I did a BK search earlier in the article and ended on the same page, decided I couldn't be bothered to play those sort of games and clicked away. The GPs link actually looks rather interesting so I'll investigate, so take this a hate-it-folk vote.
Understandable; let me ask a a question. You don't want to play these sort of games (read a paragraph, enter a word). For you, browsing to find a compelling devtool, what makes you say, this is legit? Can you share examples of a couple of sites that do exactly what you are after?
I say that not because we wanted the CLI homepage to be 'legit', the light context there is we needed a way to quickly change direction from a previous failed initiative that added stark category marketing across the classic site... so took the opportunity to do purposefully do something very different from conventions, rightly or wrongly.
I'd never heard of BK before and I see some positive opinions on HN; I manage a small company's CI, we're really rather happy with GitLab CI, but I'm always on the lookout for something better. Clicking through to your page I'm looking to quickly find out what are the features, why it's different, how much it costs ... and for that a boring, routine website is what I'm hoping for. I'm very much not against the command-line (most places I work people complain that I use the command line when "there's this really good GUI"), but command-lines are hard, they need to be learned -- when I'm just looking for the outline as to whether it's worth digging further I really don't want to have to learn your command-line in order to get it, boring and routine is better. Just one grunt's personal opinion -- best of luck with the business!
I have a lot of engagement data on LLMs from running a creative writing oriented consumer AI app and spending s lot of time on quality improvements and post training
Consumer can eat all the GPUs they have and more if we stop trying to force B2B
Right we have a loop where AI is so expensive (because it's priced to feast on B2B margins) that the best consumer experiences aren't affordable, and they're not affordable so they don't go mainstream, and they're not mainstream so no one is willing to take less money and bank on the incredible volume that would emerge if it went mainstream.
If we can get model pricing cheaper AI entertainment alone will probably carry things (I'm 99% sure NovelAI is already one of their largest customers outside of major AI labs)
Even if consumer can eat all the gpus, it cannot have the margins (as you say), and thus won’t sustain the current valuations all these companies have and which fuel the (necessary) investments.
NVIDIA is sitting on 74% gross margin. If we reach a place where "all" these companies have to do to unlock nearly unbounded demand is take lower margins, they will find capital.
If anything I'm more worried about the consumers than the companies.
It is not nvidia who has problems here, but the ai companies which depend on huge valuations for their continuous funding and survival, thus also those who invest on those valuations. I am not worried about them though, I hope they all burn to the ground, but alas it will not be them who pay for it in the end.
> Consumer can eat all the GPUs they have and more if we stop trying to force B2B
You should really crunch the numbers on buying and then running enough compute to run a leading edge model. The economics of buying it (never mind running it) just dont add up.
You still haven't factored in "training", the major problem right now that every one remains head in sand about.
I dont need a model to know who Tom Cruise is or how to write SQL if I am asking it "set up my amazon refund" or "cancel xyz service". The moment someone figures out how to build targeted and small it will take off.
And as for training, well having to make ongoing investment into re-training is what killed expert systems, it's what killed all past AI efforts. Just because it's much more "automated" doesn't mean it isnt the same "problem". Till a model learns (and can become a useful digital twin) the consumer market is going to remain "out of reach".
That doesn't mean we dont have an amazing tool at hand, because we do. But the way it's being sold is only going to lead to confusion and disappointment.
Consumer, as in B2C, not consumers buying directly. B2C companies will happily buy (or rent from people who are buying today) GPUs, because a huge part of the game is managing margins to a degree B2B typically doesn't need to concern itself with.
> I dont need a model to know who Tom Cruise is or how to write SQL if I am asking it "set up my amazon refund" or "cancel xyz service". The moment someone figures out how to build targeted and small it will take off.
I think people got a lot of ideas when dense models were in vogue that don't hold up today. Kimi K2.5 maybe be a "1T parameter model" but it only has 32B active parameters and still easily trounces any prior dense model, including Llama 405B...
Small models need to make sense in terms of actual UX since beating these higher sparsity MoEs on raw efficiency is harder than people realize.
I disagree with their reasoning and would say it's more for strategic benefits.
Giving firms that they get along well with (like Sequoia) allocation feels like a mix between a favor and possibly a way to signal that the valuation has some external buy-in too.
But they're stating you can only use your subscription for your personal usage, not someone else's for their usage in your product.
I honestly think they're being short sighted not just giving a "3rd party quota" since they already show users like 4 quotas.
If the fear is 3rd party agents screwing up the math, just make it low enough for entry level usage. I suspect 3rd party token usage is bi-modal where some users just need enough to kick tires, but others are min-maxing for how mamy tokens they can burn as if that's its own reward
reply