Anthropic Just Gave the Middle Finger to Devs

whilst Anthropic gave us extra usage or increased usage since the deal with SpaceX, they've also taken a tremendous amount of usage away by basically paywalling any sort of programmatic usage of your Claude Max subscription. Command AI If we draw If we just draw a line here and we get some text, you've got say when you go to claude.ai is the kind of ideal use case where you can use your Claude subscription. And then all the way over here, you've got API usage, okay? These are the extreme ends of what Claude Anthropic offer, you know, how you can pay them money. And and seemingly across the x-axis here is also the less preferential way to to to use your subscription. So, then further down the line, you've got Claude code here, which is you know, slightly less preferential because it's so flexible, but it's still within the realm of feasibility. But then you've got something like Claude - P, which uses Claude code, but allows you to wrap things around it. It's outside of the Claude code UI. Let's call it the UI, the TUI, the the the console where you know, that you probably all know and love. This enable things like T3 code, conductor, your own tools and all the rest of it. It starts to This is where it starts to sort of break down. And then you've got the Claude SDK, which is a similar sort of thing, allows you to build tools, but you write the code for it. Claude - P pipes into Claude code and uses everything around it. This uses It almost uses the code that Claude code uses. It's sort of like if you were to draw a tree, you've got Claude code here. Um you've got the let's say Claude uh core, let's call it rather than Claude code, and then the SDK will also share that that code as well. And but it's separate from Claude code. Hope that makes sense, but that that's like a little rough diagram there. Um what do we got after this? And then you start to get We have to move this along here. You start to get uh uh tools that are using Claude these these the Claude P or the Claude SDK. These are things like open Claude. These are things like um open code. Open code, right? And I think open code is probably a slightly more cuz it's still in the wheelhouse of code, it becomes more favorable on the left side as it is the right side. So, there's the scale of of what we're talking about here. And again, this is totally stolen from Theo, so maybe go watch his video, is that they drew a line a few weeks ago right here where they were like, "If you're using open code or using open Claude, then we're going to charge you API usage. It's going to rightly or wrongly steal from a you know, as soon as it detects the open code or you're using open code, it's going to take from like a bucket like a purse of cash um that you've up, and sometimes it will warn you, sometimes it won't. And there was a report that I think they got refunded, but um Hermes as an example, which is an open um open Claude competitor, the uh that there was a there was a Hermes.md written in a Git comment in their in their code base, and uh Claude code picked up on that and decided to decided to charge them 200 bucks and they didn't know anything about it. Um, I think that got refunded, but you see this kind of that was the kind of I think the point in this whole equation which triggered in Anthropic's mind that they can't accurately detect whether you're using open claw or whatever. So a new line needed to be drawn. And they've now drawn that line over here. This is where they draw the line. Basically, anywhere outside of any sort of Claude's UI is where you need to start paying for it. To sweeten the deal, they they're going to be introducing as of today and you'll probably get an email or have already gotten an email today saying that you get the equivalent of your subscription pocket of of token cash to use in these circumstances, right? And funny enough, I actually started to use P {dash} P in a couple of cases cuz it's quite nice just to send it off, do something, and then not have to worry about it cuz that's exactly what it does. It just it doesn't even enter Claude code. It just does its thing and then doesn't and it just exits straight away sort of thing, right? So you get a bit of money to spend in these areas. The moment you use this, it takes out of this extra purse. The problem is it charges you API usage. So if you've got a $200 a month plan, I've got the $100 or if you've got $20 a month, you know how expensive uh, Claude is, right? That doesn't last very long on the API usage. It's still nice, don't get me wrong. It's better than what we have, yeah, but or it's more clear what we have and we'll get into that, but it's not not necessarily going to last you that long. I mean, this wouldn't be the end of the world if they hadn't already had this, you know, had this line. That this was all if this never existed, it wouldn't be such a big deal, but now they've had to draw a line and and I forgotten his name, Matt Polkan, something like this. He said it really well in that it's such a relief that we have this clarity now. And this you know, again, Theo explained this and it was such a sigh of relief just to you can clearly see this transition now. I should probably remove this um line. Yeah. Um it's nice to have this clarity cuz it really like I bet when the Anthropic guys have come out with it it was like, yeah, this makes sense. And it does make sense. But it's still a kick in the nuts. You know, it's still a kick in the nuts that that it's almost anti, you know, a lot of people are building their own harnesses around Claude code and and all the rest of it. And at the end of the day you paid you're paying Anthropic money every single month. You should be able to use your tokens however you want. However, they're so subsidized that they're probably you know, they're they're losing so many tokens. OpenAI taking the complete opposite stance. They are you get your OpenAI subscription, you can use whatever you want with it. You've bought those tokens and whatever. I don't know how long that's going to last because we've reported in the past that OpenAI are taking the inshitification not the inshitification model but the the startup model of just giving you everything just like Uber did, just like Netflix did, just like all of these other companies. They just give you everything for a short price until you're hooked. And then they take it away. So that's and and that's all Sam Altman knows. We've we've reported on this in the past. So that's the long and short of what's going on. That's how it kind of, you know, the landscape of things. Discuss. [laughter] Yeah, well, no, you explained everything perfectly and it is Yeah, it's quite sad because I was enjoying T3 code. Yeah, so now T3 has been updated to basically run Claude code inside of its UI, which doesn't Yeah. It's Why would you do that? It defeats the purpose, right? Yeah, it it looks like an iframe. It defeats the purpose. Uh you can still use it just for clarity, but it's like you hit the limits 40 times faster or something like that. And and then, yeah. So, it's pretty much unusable. I think this is This speaks to the reality of how expensive these AI models are, how expensive they are to run, uh to train and to run. And we've been getting the subsidization from Claude, but they just can't do so much because they cannot offer more compute for free. Um and they have to give it to their enterprise clients, and that's what they are doing. And Mhm. It It sucks. It sucks, but they've been giving away way too much basically for free, right? Like you can you can draw like what? 3, 5K worth of tokens, 200 bucks, but you use as much as 1,500 No, 15,000, right? Not 1,500. 15,000, yeah. Yeah. That That That's crazy. So, they subsidize that by for example, me not buying the 200 bucks plan, but not using it all that much, but you going nuts and using all of it. But they can't do this forever. And the point is AI is more expensive than uh these companies wish it to be and more than what we wish it to be. And we are facing this reality right now, and everybody is kind of waiting on maybe another model coming up you know being much more token efficient maybe these companies are working on that they know that they can get it much more efficient because we likely need way more tokens way more automations like Open Claw was just a start of it like everything will become Open Claw if if we would have enough tokens we just simply don't have nearly enough tokens as much as like we we would like to have so that is going to happen and this kind of like uh speaks nice to uh the the next topic like the next video which we have about the 12 million context uh model. You can do all of this with everything that Claude has offered you Anthropic has offered you right you've got your routines you've got um you've got Claude code for writing code you've got schedules you've got loads of things managed agents right loads of things they offer you. I think the biggest I mean you know Theo was really angry but unfortunately he's he's kind of biased because he the wind got taken away from his sale of his products that he was pumping a lot of time and energy into right whether he's got this secret master plan of actually wanting to sell it in the end even though it is open source or whether he's got you know whether there's something along those lines or whether the fact that he literally has just pumped a load of time and money into building something that's now again But they they there are better ways to do things the the Claude app the desktop app Claude code they they are a little bit buggy they aren't the best and people were building clean alternatives you know [music] that that people prefer to use right and they would do that with their Claude Max subscriptions, which was a win-win. It really fueled and um inspired the developer community, right? And with these new changes, they've basically shunted the developer community. They even though they provide core code, the real power users were using core-p or the or the agent SDK. Um and those people are the ones most hurt by this. Um Yeah. But it is what it is and I do I do see why they've done it. I I they have every right to say why you know, we'll give you we'll give you subsidization as long as you use our tools. There's probably an analogy out there in some some way, shape, or form. It's like, "Okay, why would you go buy a car but the moment you put in an engine a different engine in it, why should that come car company help you out if you break the car, you know, or crash the car or something like that, you know?" Uh that makes sense that they are taking control of the products that they offer. Um and if they'd never done it in the first place it was if it was never there in the first place, I think this would be a a much easier pill to swallow. But because it was given to us and kind of taken away and then given to us and then taken away again. Yeah. No, but the clarity is there which Matt Pocock said. He used a term called away from keyboard. Anytime you're away from the keyboard um is where they're going to punish or punish I shouldn't use that term is where they're going to charge you. Yeah. And that makes sense because open core made it so you could have agents running pretty much 24/7, you know? Yeah. Um so anyway, that's the end of that one. In good news from uh Claude is the agent This is part of a larger conversation on my show Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter. Command AI.