Hello. Just going to test. Yeah, it looks fine. Hey, Alan. Nice to see you. Just going to see if everything works here. Think we should be go. I've done this a few days in a row now, so I think everything should be fine. Hey, I advised Nice to meet you. So yeah, today wanted to Hello Charlie. Wanted to um take a look at uh some clothes code changes. It's been a lot of talk about um rate limit changes. So I thought we can just go through that, see what people means, get people involved maybe in the chat. Uh I want to hear maybe what people think of the changes. That is kind of the main thing. And this morning I was testing different stuff in claw code. I was looking at different hooks I wanted to learn. uh a bit more about sub agents. Going to be looking at that and yeah, let's just see what happens. But let's just wait a few minutes and I'm going to dive a bit more into this uh rate limits here. So, I want to hear if people saw this coming or was this a surprise. Hello, depth. can go at the mug. So, I'm not 100% sure what this is going to mean for cloud code, but they are kind of saying they estimate this will apply to less than 5% So, it doesn't sound bad, but yeah, we'll see. And there's a lot of different posts on Reddit. I saw an X and stuff. You're going to just look at some comments, see what people are thinking about this. Thought it could be interesting. So, this post got uh 1.3 million views in one day. Pretty crazy. So, I personally haven't had any like big issues with rate limits uh before after I upgraded to the to the the $100 plan. But um what's going to happen now if um I have this it's going to be bad for those that use it seriously like yourself and better for casual users. Yeah, I'm not quite I don't think I'm kind of the biggest user of tokens, but like for developers that has kind of gone Hello Sunni, nice to see you again. Developers, software developers that have gone all in on this, they are probably maybe going to notice this. Hi English, nice to see you again. Nares, nice to see you. So, like I said in the beginning, we're just going to talk a bit about the changes to just want to hear what people think about the new rate limits and if people think it's going to be they are going to be affected by this. Hello Mova, nice to see you again. Kagas, nice to meet you. Haven't seen you here, so thanks for dropping by. Are uh Elvie, nice to meet you. You seem like you're Swedish, so hello. Hey. Yeah, like I was saying, uh I'm on the Hello, Benjamin. I'm on the 5x plan, so I haven't had any real issues yet. Hello, pitch and kicker. Thanks for that. I appreciate it. But um so you got the five hours limit yesterday. Yeah. Okay. So you are pretty uh you're pretty um I don't know English. What do you kind of think initially of the the new announcements on the rate limits? Do you think it's going to affect you? Yeah, it's going to be interesting to see. I know they state here that the estimate will only apply to less than 5%. So that's got to be kind of the the users that so 95% will not be affected. Uh so I assume it says it was kind of expected. I was asking questions about the limits and how they compare to my API usage. Yeah, that was what I was thinking too. like uh was this expected? I kind of wanted to do a poll on this. I started to use more set. Uh yeah. Yeah, that's what I've been doing, too. I kind of wanted to do a poll. What people kind of if this was a surprise. Uh so let's do that. Uh expected uh I'm going to do Hello Ali Risa. Do you think this AI companies charge based on the salaries of people who live in the US? It's Yeah, I guess if you're like from a a country like with less income, it's going to be expensive right? Yeah. See you soon, English. So, let's do a poll on this to see if the people are here what uh you think of this. We haven't really seen the effect yet, I guess, but uh let's see if we can I don't know. I just said expected seems fair or if you're disappointed in the changes. For me, I kind of when when we saw them people spending $5,000 of API cost and they paid $200. That kind of told me something is going to change. What are the new limits? Hello Martin Sarahana. Nice to meet you. One aspect cloud restricting usage and other aspect um and another aspect they are hook hook developer to use more sub agents to more automation. Yeah, could be that. Hello sir, nice to meet you. What I don't like about Antropic though is that they is that it's always about limits whether platform uh or the API like the RPM caps. Yeah, there could be like a lack of compute right? It has to be only a surprise because I didn't know it was coming, but it makes sense when affects only 5%. Uh yeah, I have some there's this kind of leaderboard. We're going to look at that that people spending like insane amounts of tokens. I don't even know how they can do this. Yeah, I think about the rate limits and stuff. It has to be compute issues. Thanks Cir. Nice to see you here. So Martin, we we're going to look at the new limits. I'm not 100% sure what's going to be changed. Like it's going to be like weekly now, I think, instead of 5 hours. Uh AI class says, "Thankfully, I canceled a few days ago when I realized it was getting dumber." Okay. I know the use case is different, but I never reach a limit using chat GBT. No, me neither. But of course, I spend more tokens in cloud code than I use in chat GBT. Yeah, it's going to be exciting to see GPT5s. I think I read someone that talked about it could be good for coding. Hello Tim. Uh Sims plan to minimize outage. Yeah, I have run into overloaded warnings on u cloud code. Thanks for covering it. Appreciate the live. Yeah, I'm interesting getting to coding. Brand new. Any recommendations uh on a total new beginner? Uh I have to jump but I'll be re-watching. Okay, cool. So, Charlie, uh, any recommendations on a total new beginner? Uh, maybe if you want to dive into coding, uh, I would just start with something simple. Maybe not start with Cloud Code, maybe you could start with, uh, some of the Lava Ball or some of the Vibe coding platforms. or if you want to build something and just try to take it from there, maybe learn a bit more about structure and stuff. I'm not quite sure what the best advice for a total new beginner. I would just get started getting familiar with the tools. That's kind of my uh usually recommendation if you want to get into this. Just start playing around with the tools and just take it from there, I guess. I would like them to do daily limits but reduce uh the 5hour weight down uh then at least when you when you are in the zone working reduce the 5 hour wait yeah okay that's an alternative I guess you can work better for a day uh or so to get the work done and not use it all our days I guess so the problem is that when I set daily weekly and monthly limits so if to use it heavily over a short period. Yeah, I see what you mean. You can easily get stuck. That would be helpful if uh how many tokens you have available before you run out. Yeah, I think they have to do that, right? So, you kind of know what where you are on your limit. Thanks, Charlie, for dropping by and have a great day. Hopefully, see you again soon. So let's take a look at the poll now. I don't know. Okay, we have a lot of votes here. So 44% says so far that is what's expected and almost 40% are disappointed. Uh yeah, of course you could be both. You could be disappointed, but also as expected when uh some of you guys spent like $2,000 on a $200 plan of API cost. But let's just go through the the the tweets and kind of the post. I also have the email. So, we're rolling out new weekly rate limits for Cl Pro and Max in late August. will estimate they uh will apply to less than 5% of subscribers based on current usage. Okay, so that's not going to be me. I think and I kind of follow up with uh Cloud Co has seen unprecedented demand especially as part of our max plan. Yeah. Uh because it was really good value if you used it a lot for things that was uh productive. We'll continue to support this growth while we work on making Cloud Code even better. But for now, we need to make some changes. Some of the biggest Cloud Code fans are re running it continuously in the background 24/7. These users are remarkable and we want to enable them, but uh a few outlying cases are very costly to support. For example, one user consumed tens of thousands of model usage on a $200 plan. Yeah, this was kind of the rumors, right? People spending thousands of dollars and just paying 200 in API cost. So, that's probably something uh anthropic. I should they should have expected that, but I don't know. Uh in other cases, a small uh number of users are violating our users policies by sharing and reselling accounts. Okay, that's not good. But that should be like an individual punishment shouldn't it be? Uh this impacts uh a capacity for all cloud users and we're taking um appropriate action to stop it. So from August 28th so it's a month uh we will introduce new weekly limits that will mitigate these problems while impacting as few customers as possible. We will also support max plan users buying additional usage at the standard API rates. Okay, so that's going to get expensive, right? Why do I keep clicking all that? And um we still uh we still we're still exploring the best ways to ensure uh as broad access as possible to cloud code. If you're a power user who wants feedback on how we can support your specific use of cloud code, we love to hear from you. Okay. So yeah they have some fair points, I guess, but shouldn't they kind of have expected that coming? So, Mabe Wrigley says some of the biggest Claw fans are ruining it the continuous running it continuously in the background 24/7 for example. Yeah. So, so he has done that, right? Uh yeah. Um have you used N8N? Have you made a video about it? I haven't yet, Death Wolf. But, uh it's on my plan. I'm going to try it. So, Tim is back to hating Antropic. Yeah. Yeah, I guess the only thing I have to say is like they should if they enable unlimited use. It's not fully unlimited, but if you can get this much value, some people are going to take advantage of that, right? So, here's one guy. I totally support this. It makes me feel less frustrated when not having cloud instances running during my working hours. I only use cloud code for personal projects. Was blocked by uh overloaded APIs over the past two weeks. Super annoying. Yeah, it happened to me a few times, too. Uh I got this overloaded message. Uh the correct sensible reaction. Thank you for limiting uh abusers and allocating more server space for normal users like myself. I guess I'm not going to be affected. I don't think I'm going to be because I don't spend. So, I can show you this. So, this is the leaderboard part I found. I don't know where this leaderboard is, but uh or this is kind of I don't know where this is, but here you can see it states that this user Jeffrey Huntley spent 458,000 13 billion tokens. Is that even possible? I don't understand this. Can this be correct? 13 billion tokens. 89 billion tokens. This is just sad. They probably didn't even code something useful. So someone says to be fair this is an anthropic problem uh at its core not a user problem. They should have known better. They should have had better transparency and guard rates in place uh from the start so users knew exactly what they were getting for the money and what the actual usage limit is for the plants. So of course people start using a tool and realize there's no limit. They're going to keep pushing it right. Uh, I kind of agree a bit with this. They should have seen this coming. I think I think so. Uh, I was really happy with this 5 hour limit. It seemed quite fair. But um someone here yesterday uh mentioned something that they could have two accounts and they use this some kind of bridge. So when one account got limited they just switch to another one. I guess they're paying for both though. But it is pretty interesting. So, the email I got was uh I I guess we kind of went through this, but if you look at Oops. If you kind of look at the the limits, so hard to zoom on this. Uh my usage, I don't have the that installed now, but it's not high. Uh, I don't have a lot of uses for I don't spend a lot of tokens. So, what's changing? Starting August 28, we're introducing the weekly user limit alongside our existing 5hour limits. So, those will still uh prevail, I guess. So, the current is now uh limit resets every 5 hours. The new one is going to be an overall weekly limit that resets every every seven days. And Claude Opus 4 uh weekly limit uh that also resets every seven days. It's a bit confusing, isn't it? Hello, Mickb. In my case, the limit that I hit seems random. Yeah, I would like to see I would also like to see my um like an overview of kind of my usage limit. That would be nice. Yeah, this is seems a bit if you don't know how much left you have off your weekly limit, it could be limiting, I think. And I follow up with what what this means for you. most users won't notice any difference. The weekly limits are designed to support a typical uh daily use uh across your projects. So, most 5x users, this is kind of me, can expect 140 to 280 hours of Sonnet 4 and 15 to 35 hours of Opus 4 within the weekly late rate limits. So for me that seems fine. Heavy Opus users with large code bases and those running multiple cloud code instances in parallel we hit their limits sooner. You can manage or cancel your subscription anytime in settings. Okay nice. None of the companies ever give you info on how much your quot you have used uh until you reach it. Yeah, I guess uh chat GPT has something like you now have five messages left or something, I guess. But here it's a bit more urgent because if you're actually running this on your codebase and you're kind of set up your system, so you're kind of dependent on using uh maybe some agents and stuff to do stuff for you. If you suddenly reach your weekly limit, you could get some issues. But uh for me personally, I don't think I'm going to be affected. But I know some of you uh that has been here the last few last few streams uh spend a lot of tokens. So some of you could get affected I guess. So I kind of wanted to read a bit of the comments here what people are thinking about this just to get like um sentiment analysis. Yeah, I like this one maybe. So, uh, this Reddit user says, uh, okay, then add an always visible toggable percentage indicator of limits per model with time frames. This would be perfect, right? Then you can kind of see each day how you are kind of doing both on your weekly limits, maybe uh on your daily limits too, maybe on your 5hour limit too. or just give us a number that you can track yourself. The worst thing is when you just suddenly run out even you didn't see it coming. Hello Samuel. Unfortunately to see rate limits, but I see it as understandable. Yeah, it's it's expected. I'm not surprised when you kind of see people are spending $10,000 of API credits but only paying 200. So Martin says, "I think weekly is a bad idea unless you know how many." Yeah, exactly. But this toggle here, this is a good idea. This would be perfect or just some kind of slash command to see how you're doing. Uh IR Sunni says, "I hope GPT5 puts real pressure on them so just so they stop overdoing it." From my point of view, it feels like they have a strong position and just doing whatever they want. Yeah, they kind of have the if they if they have the best coding models, they need some um some competition. Right? Because I think I it feels like uh most developers now that are kind of serious are kind of even if they are on cursor they are picking the sonet models and I think especially Opus is like so expensive when you kind of compare them to other models. Yeah. Yeah. I can see that. Martin, that's a good one. Where are those leaderable guys flexing 24hour execution? Yeah, this is this is the guys. But I don't understand this. Is it possible to use 13 billion tokens? 89 billion How are they doing this? I don't know. Crazy. GPT5 will be a solution. I mainly use Opus 4 and I feel that uh $100 will not if if you are running cloud code on AP API cost, you're just going to get $200 in like a few hours. How did it get even get this information? I don't know. I know there's um there is this open-source tool that tracks your token spending so you can kind of see how many how much you have racked up in API cost. So Samuel says, "I also think AI uh as an assistant tool, amazing technology, yet uh I think it would be a big mistake to simply plug in all systems into it and autonomously run them without safeguards. Yeah, I don't even see if you're running this 24 uh 7. I don't even know what you get out of that. Do you even get I don't even know. You have to do some manual stuff, right? Yeah, I see that. Martin, are you on the $20 plan? It's weird that number one is exactly twice. Yeah, I see that now. Yeah, I'm not quite sure about this, but uh something I found. The leaderboard tells me Claude needs to tighten its rate limiting to reign in the users themsel the top 5%. Be interesting to see how they actually rack up this many tokens. Okay, you're on the $20 plan. That's a pretty good I kind of like that plan too if you have just you don't have any rush. Hello Dorups. Uh when you have unlimited data subscription uh with carriers uh you get onto a small leaderboard when hitting the you you get onto a small band when hitting the limit. Uh they could do the same for heavy users I guess. So uh they seem to have gone another way though with MVPs testing um eats up a huge amount of tokens before getting to the final solution. There's no choice. Uh, we have to run test. Yeah, I see what you mean. I spent uh I tried this test driven development workflow. Uh, the agents ran for six hours. I don't think they spent a lot. They probably spent like a million tokens, 1 million, something like that. But that was a small application. But they ran for six hours straight. Thanks Samuel. Thanks for dropping by. Yeah, a lot of people seem kind of angry about this. But uh again, I think maybe Anthropic should have seen this coming and it seems like it's just going to be um collective punishment. Um there might be 5% of users that are heavy users. I don't think it will affect users like this. I I I agree. I don't think I'm going to be affected at all. I'm quite sure. I think I'm going to be fine. Hello Alpha. But I think like if people has uh violated policies like sharing and reselling access uh they should be just banned, right? An advanced users pattern like running 24/7 in the background. If they broke the policy, I guess they could just ban them. It shouldn't be an issue. Anthropic are driving uh your most power users. Anyway, uh who also uh your most avid brand ambassadors, they will go to other providers. Uh that may well trigger a downfall of Antropic. We'll see. To be honest, I don't think it's going to be that big of a change. We'll see. You're working on HR project and imagine submitting 300 CVS to the model for a single test. That's 5 million tokens for one run. Yeah, I see what you mean. But um I think we'll be fine. Most of us Another thing to put do you need to use opus for for everything 100% I just uh the way I do it I go to my claude and I do slashmodels run npxcc usage on a terminal. Okay, I can try that. So what I do uh when I select my model if I don't have anything very specific I just uh I have set kind of as my default and if I know I have if I'm going to do something on my course platform or some that uh uh is a bit more serious stuff I'm just I'm I'm just switching it to opus. But just when I'm playing around, not doing anything specific, just learning about agents and stuff, I'm just going to use set. Uh, for me, that's been working fine. So, what did you say? Um, Mal, I haven't seen this. So much usage should be okay. Yeah, I think so. Okay. So, here you can see my usage. You can see I don't spend um I guess I've spent I got my value though. So since uh in the last month I have spent uh $276 and I I've been on vacation too. Uh, but I only paid uh I only paid $100, so I got some value. But uh this was in the middle of a vacation, too. You can see here from the third here. I didn't almost didn't use it because I was away. So uh yeah, still good value. Has anyone else run this? If you are on your computer now, just run npx. Uh, what was it? NPXC usage. Would be interesting to hear what your sum is the last uh, is it last month? I think so. So, that wasn't too bad. So, I'm going to be fine, but I'm going to use it a bit more going forward. That's why I upgraded to the $100 plan. I think so. Let's take a look at the poll again. So, we got 60 votes. That's pretty good. So, we have 45% is disappointed and 40% uh think of this as expected that uh they are changing the rate limits. Uh but again it could be both. Samuel says, "You spend much more than I do, sir. I stay within my limits." You have chat GPT plus. You have cursor pro. Okay. Yeah, I kind of I I kind of got rid of my cursor pro uh after I switched to the $100 uh cloud code. Hello, Night Spider. Nice to see you. So, you're not one of the people say Antropic is running it 24/7. Uh, no, that's not me. Um, so I'm uh I have some value I guess but uh not like crazy. I think they are still within the margins here, right? They have margins on the API cost at least for Opus. So you can see I haven't even used Opus lately. kind of I kind of should have done that maybe but uh I haven't really done any changes on my course platform, so I haven't used Opus this week. British has spent 1,600. Oh, you are on the $200 plan. I think you're going to be fine, though. How many tokens total? Use opus for concepts only. Yeah, Edgar, that's kind of what I do. I just go to models or model and I have set as my default and then I'm going to do I guess when you say concepts, do you use it for coming up with new concepts or planning plan mode? So I tried to when uh kind of the plan mode came out I tried to use uh opus for plan mode and I switched to set. Okay. So British upgraded to the $200 plan. 800 million tokens. Okay, that's a lot. Uh, but you might be fine. I'm not sure. Uh, Mikbab has also got great value. He spent uh on the $100 plan, he spent API cost of 1,200. That's pretty good. Hello, Sid. Can you configure custom instructions for sub agents also? Yeah. Uh if you go to agents, uh each of them have uh a system message. We're going to look at that soon. I have some things I want to check out. So you can do custom instructions for each sub agent. They even have their own context window. Now, Edgar says he has also developed my own his own prompting technique where I need uh the most precise instructions following. Okay. Is Ben still alive? Yeah, we're going to check him out soon. So, Edgar, do you mean uh uh is this because you have a lot of um tools and agents and custom commands so that uh Cloud Code picks the right tools? Do you have any advice on developing for iOS 26? I gave Claude VV uh DC25 transcripts. Oh, I I have no uh advice for iOS development. Uh maybe some other people do. If someone has any experience with uh is that Swift is it? Hello RS. Uh please correct me if I'm wrong, but you look like someone from Ethical Hacker I used to watch previously. Okay, that might be it. I don't know who that is. Let me know and I can check it out. Yeah. So, seems like people are uh if you look at the poll again. Okay. So, now we are at 4242. So, I guess kind of the conclusion is that a lot of people was expecting some changes but they are still kind of disappointed. That would be my takeaway. So you develop software iterately and don't plan. Okay, I see. So you don't plan ahead doing this. I work with uh state and try to leverage intelligence at the beginning of it each iterations. Okay, I see. Cool. So how has cloud code worked for you then? So, Night Spider, you tried the the YOLO mode yesterday. Is that um you mean like the what's this flag called? Dangerously skip a permissions. Just skip the confirmation steps. Uh yeah, I see what you mean. Where did I find the flags? Can I see them here? Can't remember. I I know what you mean. Those flags, right? Yeah, that one. Okay, I see what you mean. I haven't tried the dangerous skip permissions yet. Uh but I will. I guess I tried it, but not for anything uh anything special, right? Uh where are those flags? Is it slash commands or is it Yeah, I can't really remember. Uh yeah, but I've seen them. Okay so I think we kind of can conclude the rate limit part. Now I just wanted to have a look at it. Seems like people was expecting this but a lot of uh people are uh a bit disappointed. I can see that. Yeah. So, I'm going to end this poll now. And we can see it here. So, I I also can't see the angle of it seems fair when people are like overspending a lot. But so, you always find yourself fighting against the chain of thought pre-training. Okay. How much uh was reduced? Do you mean the rate limits? Uh co hello Sean. So you started using the Windows version. Seemed like the Windows version wasn't as good as the VSSL version. So I switched back to Windows VSL. Okay. Yeah, there was someone here uh talking about um using it native in Windows yesterday. Hello, a spoonful of Joe. Most expensive prompt I had is 14.8 million tokens. Okay. So, you've been using the 20x uh plan code. I'm worried about the weekly limit. Yeah, if you just got here, we can take a look again. So, basically, they're going to make some changes. Uh, so now we kind of have the usage limits that resets every 5 hours. So, that's going to still be in. So, that's going to be no change. Uh, but they're going to add a new overall weekly limit that resets every 7 days. Uh, and they're going to add a new weekly limit to Opus 4 that resets every 7 days. So, it's kind of hard to say, but they have some estimates here that 5X users can expect uh up to 280 hours use of Sonet and up to 35 hours of Opus within the weekly rate limits. So, it's kind of hard to say. So, heavy Opus users might be more affected. So, not nice whether you are using uh Windows cloud code native. Okay. So, I thought I could ask uh I had two thing two more things I wanted to do today. Um the new limits will uh probably only affect you if you are exceeding the opus limit. Yeah, I think so. We'll see. Uh I had two more things I wanted to look at today. Uh I wanted to look at um sub agents and hooks. This is something I played around with this morning uh just my initial testing and maybe some of you know more about it. Uh, and I saw there's a guy on YouTube. He has this really good Claude code videos. Uh, Indie Dev Dan. Maybe a lot of you have seen his video, but I kind of wanted to take a look at it. I'm not going to publish it, but uh I think he has good videos on Cloud Code. It's a bit over my head some of the stuff he has, but uh I wanted to look at the sub agents video. So, English uh do you actually do you actually try um using son and adding the word ultra tink it switch to opus automatically? Yeah, someone talked about that yesterday to add the ultra tink something in the prompts. I haven't tried that yet. Indid. Yeah, he's great, but it's a bit could be a bit advanced for some people, I guess. So, code, you just ran the a flag dangerous mode felt sketchy. 35 hours uh a week of Opus 4. Opus failing the full 40hour work week. Yeah, in the dead is quite advanced. Yeah, I know. It's not so easy to follow along, but I had I watched like 10 minutes of his video yesterday and it doesn't didn't seem that complex, so I thought it could be interesting. The video from India about monitoring multiple yolo CC is marvelous. Okay. Yeah, I might take a look at that. To get Claw to think harder, think harder. Think hard as Ultra Think. Is that a thing? Sub agents use their own context. Yeah, that's true. Okay, so it looks like people uh wanted to take a look at uh we can give it some more time, but I just I had a look at more hooks and sub agents today because I saw if you go to slash hooks. I saw they added some new events here. They added uh they added an event called sub agent stop. So right before a sub agent task tool call conclude its response. So I don't know if this has been here. It could have been here couldn't have been here. So I haven't seen that before. Maybe it's not new but at least I tried to create a hook with the sub agent stop this morning just to get more knowledge about it and I thought I can just share what I learned. So it does people know if this new events here haven't been here before? Hello Chris killing jokes play I've been using cloud code pro subscription uh with a very good management context and everything uh I could easily manage up to four times filling of context uh but from yesterday you're lucky if done once so do you mean you get this um auto compact context issue Chris has been trying to run to get agents to run like an agency so you're the boss And the task goes to the manager and manager delegates task. Uh it kind of works uh with the new sub agents Chris then kind of claude main claude code is kind of delegating to the sub agents. We can have a look at that afterwards. So sonet with ultra tink was a good substitution for opus. Okay interesting. So don't multiple sessions of claude uh use their own context. I guess multiple session will use their own context. Yeah, that's true. But I think before we got the sub agents here, uh they kind of shared the context window with cloud code here. So you can run into issues if you had run long long run times. How do we check the API usage uh when using cloud plans? Uh you can check uh what was it? Uh SID you can go to your terminal and you can do npxcc usage. Then you can see how much you spent. Sub agents look interesting especially known code use cases. Yeah, I have created some workflows uh a lot of agents CC usage. Yeah, you can use flash compact. Yeah, so Martin uses cloud code to make micros apps. Still blowing my mind how good it is. Yeah, I think uh clo uh set or opus are good at UI UX, right? I will try set with ultra thinking. I can try it afterwards. I haven't tested that. I mean while uh using cloud uh pro uh on the same project, how many times do you get autoco compact popup? I see what you mean. Do you mean cloud cloud code now or do you mean I guess you mean cloud code uh auto compact? So I don't think uh the auto compact is affecting any plans you're on. But if you switch to using sub agents uh I don't think you're going to run into autoco compact anymore. Yeah, I see what you mean. Killing joke. So I I I understand what so before uh or now when you are running long sessions with cloud code. So what happens now if I do agent and I do Anna or I can do Ben. So when I run my agent now uh this is kind of the main context right for cloud code. You can see here are the tokens for cloud code. But um but Ben here this sub agent is going to have its own context window. So it doesn't really affect it doesn't affect this context window. So now we have kind of split this into two. Right? So you shouldn't if you use a lot of sub aent you shouldn't run into this compact issue at least that's my experience and you can use slashclear yeah and slashinit that's true >> I successfully retrieved and formatted the top 10 hacker news posts showing a diverse range of topics including AI tools file servers privacy concerns video codecs and development tools. The most popular post was about AI HUDs with 868 points, followed by a file server. Yeah, I'm just going to stop that. That is my hook. I was going to test that later. Okay. So, that was kind of my hook. I I'm going to show you that later. So, so what I did this morning I wanted to kind of see I wanted to create a hook that um so when the agent has completed a task it's going Um, it's going to call up uh my 11 labs and it's going to read out a summary of what is done. I think I saw that in a in the Dev Dan video. So, that is what I wanted to try out this morning and I kind of got it to work. It seems to be working fine. You're thinking of adding Gemini CLI for testing? Yeah, I think I'm going to check out more Gemini CLI too. So, kind of this simple setup I did this morning for testing is actually I wanted to create this simple command custom command. So I can do slash agent and you can see I can do like a agent name here to select what kind of agents I want to run or I can run two agents in parallel if I do like Anna uh comma and I can do my second agent. So I can run them in parallel. So that is kind of my custom command and kind of the way I set that up was uh I kind of have this description. So launch a specific agent by name or two agents if you want to run them in parallel. And we're going to do the agent name and the tool is going to be task. So this is going to use kind of arguments here. And for parallel execution we can do this example. And of course for my agents I have the name for the sub agents Anna and and Ben. And I have the description. I have the tools of course the color. Anna is going to fetch the top 10 post on hacker news. She has access to these tools. Web search, web fetch, write and read, and her color. So, I'm kind of going to show you kind of how I solve this. Can agents share information and state to come up with solutions and give the final solution back to main claude? uh or since they have their own context window uh I don't think agents can share information uh between them it has to go through the the main claw agent right so basically what I think is happening is like when one sub agent has done um has done this um has done its task I think it shared a context with the main claude, but I don't think it's it's not going to share it with the other sub agents. That's my uh that's what I think at least. I haven't seen anything about that. But it does share with the main cloud though. So you could chain them, right? You can do let's say I do So if my Ben agent has to uh calculate or roll a dice and it rolls six and Anna is uh going to use that roll uh to calculate something. So Claude can pass on the context from Ben to the next agent. You can chain them. Uh but I don't think they're going to share context among themsel right but you can chain and you can run in parallel. Uh but I might be wrong but that's kind of my first uh impression of trying this. So yeah, what I was struggling a bit was uh since I wanted to uh you might uh you might hack this into sub agents, create a file for a state and give every sub agent the hint. It needs to look at the file before starting. Yeah, that might be something. I did like a small hack this morning. I was going to tell you about that now because I wanted the so my initial thing was that my sub agent Anna here that's her name she's going to check out hacker news she's just going to read the top 10 post and return it that is her description and that is her task uh but I wanted 11 labs to uh kind of read out what Anna has on using the if you go to hooks we have something called sub aent sub aent um event so this is going to be right before the sub agent conclude its response so what happened is I couldn't get the summary uh from claude because this uh event was triggered after the sub agent was done. So I had to kind of work around that. So what I did is I gave the sub agent a new task. So instead of giving kind of passing along the the context to cloud code, I just uh instructed it to write its summary into a text file first. So we have a text file called Anna summary. So this is the eight sub agent is going to write this file and then we can uh use this text file here written by the sub agent and just place it into the prompt that is going to be sent to 11 labs that kind of includes our summary file. So we're just going to set the summary file to be this text file. Uh, and that seemed to work. So now I can actually get even though we have the sub agent hook, uh, we can, uh, play the summary of what the agent did. Thanks, Cole 63 watching. Yeah, smash the like button if you think this is interesting. Reminds me of crew AI. Yeah, it might be that. Uh, but the summary is from the latest task made. So what the agent done uh what the agent do is um so the idea behind the hook is that when the agent it's done it's going to say what it did right so if the agent has uh collected information about hacker news it's going to write a summary and read it out when it's done. Does 11 Labs uh pay as you go pricing for the API? I'm not quite sure. I have a monthly plan now. So, this was kind of my hack around this because I couldn't figure out another way to do it. But maybe there's another way. It's not a summary made every time an agent is finished. It's not a summary made every time an agent is finished. Yeah. So, this is kind of what I was struggling a bit with. Uh since I the problem is that when I have two agents, since this hook uh is for all the agents, uh I kind of have to include the summary part on all agents. So, I can kind of show you how it works now. So if I start my Anna agent, I just do the custom command and I do Anna. So this agent now is going to launch the Anna agent and this has uh uh like a set a set uh task it's going to do. is going to scrape the hacker news top stories and it's going to write a summary about it and when the agent uh sub agent is finished it's going to read out what it did. That was my what I wanted to learn because a hook will trigger every time that happens. Yeah. So if I want this hook, I have to all agents have to write the a new summary. If not, it's just going to read a another summary. So it's not really viable, but uh it was a nice way for me to learn more about hooks since I haven't really dived so much into that. So, so I got it to work and now I kind of understand how I can build more maybe useful hooks, right? So, but this hook isn't really anything useful. It's just going to read out the summary of what the agent did. So it's not very helpful but I think it was a nice way of uh so now I haven't given this permission every time so I just have to do this if I don't remember babbly you can specify settings for specific events in certain conditions yeah that might be true in this system you would exit the hook based on the scent parameters should work for sub agents as Well, uh, in this system, you would exit the hook based on the sent parameters. >> I successfully retrieved and formatted the top 10 hacker news posts showing a diverse range of topics, including AI interface design, file servers, rate limiting discussions, LLM security concepts, and programming languages. The most popular post was about AI HUDs with 889 points, followed by a file server project with 854 points. Notably, there's an active discussion about Claude Code weekly rate limits with 561 points, indicating ongoing community interest in AI development tools and their usage constraints. Did you hear that? So, that is kind of the hook. Uh, sorry, Edgar. Uh, in this system, uh, you would exit the hook based on the scent parameters. Should work for sub aents as well or am I missing something? I'm not quite sure. I'm quite new to hooks, so I haven't really Oh, you mean it systems. You would send a source, for example, and only execute for valid sources. Yeah, that might be true. I I'm not quite I'm quite new to this. Uh you can make a hook every time you use uh BCP. Yeah. Okay. Yeah, I understand that. But what I don't 100% understand is okay. So I'm not going to remove this, but uh I want to learn a bit more about this. I also created like a log for my hook. So just so I can do some debugging so I can kind of see what it does here. Um, but I think I need to spend a bit more time on the hook part because in your case, you want to trigger uh that agent when agent Y finish or something similar right? Yeah. So, we can change this up. Uh so let's say you can basically the event is that when this sub agent is done or right before it's done it's going to of course run this hook here that's going to run this Python code use npx daddy cloud in a new repo as a template. What is that? I haven't heard of that, Tron. So another thing I tried to do was uh if I switch this now to just read agent complete. So this is just a string right? But I can do this. And so if I change this to agent summary or the string, uh I can have if I do agent and I do Anna and Ben. These are going to be parallel not found. Okay. So we can run those in parallel. Hello Carl. The newest indie dev then was went a bit over my head too. Yeah he has very advanced videos. Um so the idea behind this hook now when we kind of changed it up to just reading out the string uh agent completed task. So hopefully now when let's say Anna is finished she's going to say agent complete the task and the same with Ben. So I would like to hear this two times now. So someone was asking about the context window earlier. So of course now Anna has her own context window but she's still writing the summary. I don't know why, but uh I have her instructed to do that. And of course, Ben agent here has his own context window, but the English >> successfully retrieved and formatted the top 10 hacker news posts showing a diverse range of topics including AI interface design, file. >> Okay, so I forgot. I think you have to reset this. I forgot about that. I think you have to go out and in to get the new hooks. This is a settings example not tested. Uh, what do you mean English? In the dead down works very quickly. It's more for pro. Yeah. This is a setting example not tested. Okay. I'm not I'm not 100% following. Sorry. Yeah, I had a look at his video yesterday. I watched like 10 10 minutes or something. Uh it was pretty interesting, but uh I think it's for more professionals. I kind of agree with that. But I learned something and I think we all can if um config agents name scrape name summary. Ah okay I see was a bit I I I think I see what you mean. But the idea is that you can trigger an agent once the other is finished and pass the context. So you you're using hooks for that. I guess that could be interesting. I think I have to look into that. Do you have like Could you send that to me in like a text bin or something? Paste bin. Yeah, try that. that could be interesting to test. So if we point to this and change the hook to only say uh agent is dumb. Yeah, try to do that. And I can look at it uh later today. Ask cloud desktop or cloud web to create a hook and be able to and setting be able to trigger once an agent is finished successfully. So you mean like the the sub agent event Well, I say what you mean. Agent is done to uh 11 laps. I don't want to print that. So, one thing I did this morning was that I set instead of only putting up uh sub agent stop, I had stop agents and sub agent stop. And every time the agent uh completed like a tool, it spammed uh 11 labs. So I found out that I only had to run the sub agent stop event. So let's see how this looks now. Yeah, this looks a bit better. uh remove the summary part. We don't need that. So we can remove this. Okay. And I want to see now if it says agent is done when my uh agents are on. If you want an agent to be able to create a summary uh of another agent results, this could be the way to create a hook. Yeah, it's quite complex, but it's an interesting workflow. And another thing I heard uh in the dev talked about in his video was that he he was talking about if a sub agent can call another sub agent, right? So let's say Anna here could could launch other sub agents. >> Agent is done. >> Okay. So that was one agent done. That was probably Ben, was it? Yeah. And I want to see now if Anna is uh when she is done if we get the same message. >> Agent is done. >> Okay. So now we kind of understand that we have both agents are done. It should be. Yeah. That's good. So that hooks works pretty good, I guess. But I guess we could create what could be an interesting hook we could do when one agent is uh finished. This is a bit boring. What do you think, English? Do you have any idea of what interesting hook we could do when one sub agent is done? Like if I had if I had my email MCP, I did once create a hook that sent me an text message when the agent was done. But I guess in theory, we could create a hook that sends me an email when the agent is done. You can hooks for tests. Yeah, that's pretty good. I guess you could launch some tests on when one agent is done. I guess that's interesting. I only have my one MCP server here now. So it isn't it's quite easy to understand but um it's coming up with like um interesting use case for it. So Carl says summary in one to two sentences of what I've been done. Yeah, that is why I wanted to try uh try this this morning uh to kind of use the hooks here on the sub agents. I want to do this lab integration too. Yeah, it's quite simple. So what I have is um I have a MCP server uh 11 Labs and I just used the so we're just going to use we're just going to use the Level Labs MCP server here to send the the string agent is on. And then we have like a temporary file here. That's going to play. >> Agent is done. Agent is done. >> I have some old files here, but I have like a temp folder here that plays this sound. What about having to write a Twitter post, create a YouTube short script, or create a Google Docs file, or Yeah, we could do that. Of course, you could of course have any hook you wanted here at the end, but the issue is that we can't pass along the the context. I think we have to get the agent to write the context we create. Let's say we get we use one agents to fetch uh some kind of information but I don't think then it has to save the information somewhere before we can use it as a hook I think. So we could do we could do something like uh so if we do we can read this and we could list um Hey, modern wizard. Nice to see you. And we can ask, can you create a hook that uh writes a summary from each agent uh when they have completed a task. Uh if that is impossible, you could of course instruct instruct each agent to write a very short summary and save it to a text file and use that text file um as a context for the hook. I'm not quite sure how you could do this. What is the best way and I'm not quite sure about where is it? Uh where are the events? So the pre-tool use events, we have the post tool use events, notification events. This is the sub agent stop precompact. Okay. session start. How much do you use CL code uh in your current $100 plan? Uh I can show you. We just um if you do NPX um what was it? I think it was NPX CC usage. You can kind of see how much you spent. So I have spent $277 worth of tokens API cost in 30 days. So, I've gotten some value from it. You can also do this if you are in your on your terminal. Hey guys, will the limits uh come into place uh 28th of August? Yeah, Michael, if you look at the announcement here, you can see uh it's going to be rolled out in late August. So, from August 28th, we're going to introduce the new limits, right? Uh I think we only want like uh one text file with a summary from both agents. Uh we don't need any text to speech uh for this hook. Do you you do a lot of you use uh tools like whisper flow like uh speech to text? I use it sometimes. It's not all the time. Yeah, it's good for I like it for when I'm doing a plan mode and just I'm kind of rambling about the application I want. Whisper is actually one of the good ones. Yeah, Whisp. What is it called? The one I have. Is it super whisper? Whisper flow. Okay, so the plan is now that the boat agents will write a shared file. Okay, but that doesn't really that's not a hook, is it? Okay, so you made your own one, Carl. That's pretty cool. Using Grock and Whisper. Okay, so the latency is good, but how do you solve kind of the Okay, so you're on Windows, the clipboard and the paste that works good. Let's say we run the the agents in a chain or in parallel. I guess it's not possible in parallel, but let's say we run the agents in a chain. Is it possible to the first agent in a chain uh writes a summary and we could get the second agent in a chain to read the summary? But that's not using hooks, is it? Have you checked out the various agents that's available in Clo? You mean the sub agents? I need to play more with sub agents, hooks, and MCP. Yeah, that's kind of what um I want to learn more about this. I'm going to see how this works. Yeah, I'm kind of playing around with that too, Mickbub. I kind of want to learn a bit more about hooks. Uh, I tried it when it came out, but I haven't really implemented any good hooks for my workflows, but uh I want to learn more about that. So now the agents are instructed to Okay, I guess I need to stop this again because now Ben needs to be able to write and read. So we need to give him that tools. Anna has write and read. So that's good. Just want to see how this turns out. So we can do resume. We can do uh test. Now before starting your task, check if there is an existing agent summaries. txt. Read it to see what the previous agents has done. Use this context to inform your work. After completing your task, write a brief summary what you found into a file called agent summaries in the working directory. Include your name and key findings. Append to the file if it already exists. But again, I don't think this is any good solution of passing on context. Let's see what happens. Might as well just uh use cloth for this, I think. So Ben ran first fetch the cloud connection. Okay, let me read that file. Okay. Yeah, that's not so interesting, I think. Yeah. Um, I think I want to watch a bit of this video. So, if you want to hang around and watch some of the video with me, you can. Uh if not that's fine. No problem. Let me see if the sound here. Okay, back engineers indie dev Dan here. Imagine starting your day. You of course open up the terminal. You fire up cloud code. Then you kick off a single prompt slashcook that does >> how does the sound? Is it too loud or is it too low? It's kind of hard for me to know. Yeah, stay around if you want to. too low. Okay, I can put it up here. The work it used to take you hours in minutes. You're able to accomplish this with claw code sub aents. You've created workflows of specialized agents that do one thing and do it extraordinarily well. You can see I have this agent called meta agent. My agents are building my agents. Code is a commodity. Your fine-tuned prompts can be valuable. And now your cloud code sub agents can yield extreme value for your engineering if you know how to avoid the big two mistakes engineers are making with sub aents. In this video, we're going to break down how to build effective cloud code sub aents. We'll of course use this powerful meta agent to build new agents. But before we get there, sub agents have serious trade-offs and pitfalls you should know about so you don't waste your engineering time and tokens. Let's understand how to take our agentic coding to the next level with Claude code sub agents. Matia says, "What about creating a slash command like morning? Then claude uh should call sub agents to tell you the weather, summarize your emails." Yeah, that should be that you can create. That should work fine. You just need your MCP servers. Hello, calling Larry. Nice to see you again. Yeah, Matias, that should be possible. So, first things first, what are claw code sub agents? I can almost guarantee you sub agents don't work like you think they work. Let me explain. Here's what the flow of claw code agents look like end to end. Starts out with your prompt. Your primary agent then prompts your sub agents. Your sub agents then do their work autonomously and then this is important. They report back to your primary agent and your primary agent reports back to you. The flow of information here is absolutely critical. You prompt your primary agent and then your primary agent prompts individual sub aents based on your original prompt. Your sub agents respond. >> Yeah, that's true. Because these sub agents they don't have anything of the context right before if you had a chat with your primary agent let's say this is cloud code this sub aent doesn't have anything of that context right so kind of cloud code primary agent we can call it that like he does has to pass on that information to the sub agent >> and not to do they respond to your primary agent? Okay, many engineers are going to miss this fact and this changes the way you write your sub agent prompts. Let's break down exactly what claude code sub agents look like. So inside this codebase, we have a simple hello world agent prompt. You can of course see this is operating inside of the brand new agents directory. We have a new agentic directory to focus on. So, we'll get to the important meta agent in a moment. Let's start simple and understand what sub agent prompts look like. We'll open up a new shell here. Fire up Claude in YOLO mode. And you'll notice here in the description, >> is it yolo mode? I guess that's the the dangerous um permissions flag. >> If they say high cloud or high CC or high cloud code, use this agent high CC. And now claw code immediately finds this agent, right? It finds the description for this agent and kicks it off. You have your agent name. This is its unique ID, the description, which is very important. This communicates to your primary agent when it should call this agent tool. >> Yeah, we kind of talked about that yesterday because if you have a lot of different agents, your description becomes more and more important. That is precise, right? So um kind of lowers your chances that, um, Claude code primary agent is going to miss miss what do you call it? Miss call or pick the wrong agent. I heard yolo, but then I saw yellow on the screen. On the $20 plan, I can't use opus in cold code. Really? just do slashmodels. So you can specify specific tools available >> subgent complete >> and the color which you can see right here. You can see we have a nice formatted response there that natural language text to speech by the way that's set up with cloud code hooks on the stop event. But if we look at this format there's something really important here and we're coming up on the first big mistake engineers are making when using cloud code sub aents. If we open up this prompt here, you can see we have a classic markdown format purpose and >> okay, I didn't know there was no opus in 20. That's pretty interesting, but uh like I have almost only used uh set and report. The first mistake engineers make is not understanding that what you're writing here is the system prompt of your sub agent. Okay, this is not the prompt for your agent. Okay, it's the system prompt. This is important. It might not seem like an important detail, but it changes the way you write the prompt and it changes what information. >> Yeah, of course, it's the the system prompt. That is why you can create different agents. So, one of your agents can be a back-end developer. One of your agents can be like a front-end developer, a QA agent. So, it's a system prompt, right? Like he says, I'm loving yolo mode. I think I have to try it. >> Is available. That's a big mistake. But even bigger is this. Notice how I have this section here. And remember in our diagram here, remember who your sub aents are responding to. It's not you, it's your primary agent. Okay, this report, this response format is going to be really important. You can see here, I'm explicitly having the sub agent communicate to the primary agent. I'm saying Claude, respond to the user with this message. And guess what happened? Look at the response. Hi there. How can I help you? Did you know Nvidia blah blah blah blah blah. Right? We had it research some random uh tech news, right? This is the sub agent prompt format. We have variable. So co you said uh you only use opus. The usage monitor on GitHub shows you uh opus versus sonet. I used all opus 4 tokens but still has messages. Okay. Uh I'm on the 100 plan now. Not the max 20, max five, I think it's called. Even with a $100 Opus limits are significant after 30 minutes. Yeah, I think the Opus model is I'm just going to save that for I'm saving that for uh things. But uh yeah, like someone said you can try set with ultra think crew AI. Yeah, that format. That's true. >> Declaration at the top. The cracked cloud code team is going to tweak, improve, add to these as time goes on. They're probably going to add the model here at some point. We have our system prompt, right? And you can see this right in the docs. If you don't believe me, you can always just open up the doc, search system, and you can see what we're actually writing here is the system prompt. To be super clear, this is not the user prompt. Okay, when you're writing uh let's open up the prime command. When we have something like this slash prime, this runs right into our primary agent. This is a prompt, right? A user prompt that goes right into the context window of our primary agent. Okay, our agents directory is very different. This is a system prompt. We're defining the top level functionality. These prompts are going to look similar, but it's important to delineate, right? You can pass variables into this. This is a system prompt. This is not what triggers what's actually done. And so, you know, again, just to bring this up, if you understand this, you're going to perform very well with your sub agents and you're going to be avoiding the top two mistakes engineers make with cloud code sub aents. >> Yeah. You just have to you just have to understand that the primary agent is what is sending the prompt to the sub agent. So you're not sending the prompt to the sub agent that is going to be clawed code that is kind of going to pass on the prompt to the sub agent. Sonet is enough. I use opus only for planning and yeah me too. >> You don't prompt your sub agents. You can write a prompt for your primary agent to prompt your sub agents. But you're communicating with your primary cloud code agent, right? The top level agent. We'll call this the primary agent. It's cloud code that prompts your sub aents, right? It is delegating. You really want to be thinking about your cloud code sub aents as tools of delegation for your primary agent. And this is why the big three is so important. Okay? We we cover this all the time on the channel. This isn't going away anytime soon. In fact, it's only going to become more important as we scale up our agent coding to multi- aent systems. Right. Context. >> Yeah, Carl. Uh the sub agent doesn't have any context, right? So the primary agent has that's called code has to pass on this context to the sub agents. You could of course give them context in the system prompt. I have an sub agent that has like a template email and that context comes from the system prompt. Right? So you could give context in the system prompt to the sub agents model prompt and specifically the flow of the context model and prompt between different agents. Okay, this is super important. Okay, so I think you get the point. I don't need to keep harping on this. It's just really important. If you make these mistakes, this will eventually go ary, right? Especially as you start scaling up what you can do with chains of sub agents, right? So, we're just getting started here. You can chain the call and responses and call and responses, right? It's funny that we started out prompt chaining years ago and now we're still prompt chaining. We're just prompt chaining with bigger compositional units. You and I, the user, prompt the primary agent. They can then run tasks and then based on the prompt, they come back into the primary agent and you can have so I don't know if anyone has an opinion on this, but I was thinking when the sub agent passes on the context, is there any loss here? like do we have to trust that the sub agent passes on I guess it doesn't pass on all the context it's just going to pass on uh what it thinks is relevant I guess so there could be some loss here in context when we kind of chain this because cloud code is going to get the context from this uh let's say planner agent sub agent but there could be loss in context here. I'm not quite sure how that works. >> Have your agent keep cooking, right? It can keep doing important work for you. So, you can fire off another set of sub agents. And so, this is the true flow. Your sub agents are responding to your primary agent. And if you're doing powerful multi- aent orchestration, which you will be pretty soon here, if you're watching the Andy Devdan channel, make sure you're subscribed and you comment to stay plugged in to key engineering information like this. You can tell um I don't just copy and paste the documentation. Um you know, everyone's using this. I think cool, that's good. Use this. But um read the documentation, guys. Like really read through this stuff. It's important. Okay? This is the most important technology of the year, probably over the next few years, right? Agents are how you win as an engineer. Understand what they really do, right? Don't offload all the >> He's really bullish on agents. At least >> cognitive work to your agents, right? Understand the most important technology. Okay? And to be super clear, there's nothing more important right now than agents. And the leading agent is Claude Code. If you've been following the channel, you already know this. You've been listening to me glaze the team and glaze the the product. You know, really communicate the value of this, right? But it but it's all for a reason. I focus on the signal that every engineer has a superpower. Every engineer is super super good. They're cracked at something. One of my key abilities is to focus on the signal and cancel out the noise. Okay? I can focus I can focus directly on the signal of where the valuable information is. Truly, there's no other way I could show up here for you every week with something valuable if I wasn't constantly just obsessed and passionate about finding the signal. Okay, this, you know, again, this is super powerful. Copy it, throw it into an LLM, but don't offload deep understanding, right? You're basically just vibe coding then. Okay, deeply understand your tools so that you can do more, better, faster, cheaper, so on and so forth. Okay, any >> Yeah, I guess that's a good point. That's why I'm kind of kind of doubling down now on um looking into hooks. I kind of want to learn that a bit more and trying to combine hooks with sub agents and MCP servers. Uh I think that's a good combo. Hooks, MCP servers and sub agents. That is going to be my focus going forward. Um yeah, at least when it comes to my workflows. >> Anyway, where were we? Um we were talking about chaining. Yes. So this is super important, right? If you want to scale up, you need to understand the information flow. Okay. We're not just prompting a chat interface anymore. We're not just dealing with one context model prompt agent system. We're dealing with multi- aent systems. Okay, there's a reason I put up that multi- aent observability video. I knew this was coming. It's here. Okay, multi- aent systems are here. The best engineers, the most cracked seniors. We're scaling up hard. Okay, compute, compute, compute, compute. And you can see it here. This is just a simple flow, right? You prompt your agent, hit fire sub agents, collect results. Okay, then the next step in that custom slash command is going to be great. Collect the results. Fire off more sub aents that you know create a concrete solution. This is all fant. I don't know. I'm not quite sure, but would be nice if he can comment on kind of the context loss here. Uh I'm not quite sure. There's no way to monitor, is it? What the agent is doing? We can't really see the full workflow of the planner. I guess we can kind of see the output, but we can't monitor it. Uh, this is indie dev Dan. He's he's great on cloud code. I think he only does one video a week, but uh I really recommend them if you want to dive deep into cloud code. So, I'm trying to kind of up my game a bit on Cloud Code. So, I've been watching his videos and I recommend them if you want to spend a lot of time on this part of your workflow. I'm going to link the video after the results. Fire off more sub agents that you know create a concrete solution. This is all fantastic. Let's actually use this. I did mention I'm going to share my meta agent with you. I've built uh 50 probably a 100 plus agents already which is kind of crazy to say but it's all thanks to this meta agent right as soon as you get access to a new feature figure out how you can scale it up oftent times you know with Genai you build a meta version of that right the thing that builds the thing >> scares me every time >> what are we going to do here let's just delete a bunch of stuff okay if this causes you pain you know if deleting causes you pain you probably aren't deleting enough. Uh so let's get rid of this. Let's get rid of this. And uh let's get rid of this as well. Let's create new agents with our meta agent. And so to cue this up, you know, a big issue I see in the engineering space in the Gen AI ecosystem is that I'm seeing a lot of engineers uh using technology to create solutions for problems that don't exist. >> Yeah, it is very dramatic. That's true. Can you create a sub agent that writes to a file writes to a file the input it received? Uh, so you mean it writes to a file what the primary uh or cloud code told it to do? Is that what you mean? I work a lot of I work a lot of meta nowadays. Totally feel him. Okay, >> noobs and beginners, they start with the technology and work backward. If you want to become a real engineer, if you want to continue being a valuable engineer in the generative AI age, work the other way. Work the right way. Work the way that product builders think. You have a problem first. Next comes a solution and then the tech comes third. Okay? Problem, solution, tech. Okay? So, what do I mean by that? Let me show you a concrete example. Let's use our meta agent to solve a real problem. Let's start at the top, right? What's the problem? When I'm a magenta coding at scale, I lose track of what some agents have done. Okay. Problem statement. Solution. Add text to speech to my agents so they notify me when they're done and more importantly what they've done. Okay. So now we have a problem and a solution. Now let's use the technology. Okay. Only after you have a problem solution. >> Yeah, that's the hook we created earlier. Uh I kind of saw him do that in another video. I think it was in a hooks video or something. He had Ev11 Labs read out what the agent or I guess it wasn't the agent uh has done or that it was done. But yeah, he's very dramatic. But one thing I never seen him build anything in his videos. That would be pretty interesting to see how he uses this for like a a small project or something. Maybe has done that, but I'm not quite sure. >> Should you move the technology? So technology, we can use cloud code sub agents. I have a meta agent here that I can use to build up a new texttospech agent that communicates what was done. Okay, fantastic. Order matters problem solution technology. So inside of MCP.json.sample, we have a setup for the 11 labs MCP server. I have this configured in my.json here. This is get ignored because of course we have environment variable information here. First I want to understand what my agent can do. So I'm going to run all tools. All tools looks like this. It's a simple reusable prompt. List all available tools in your system prompt. I want bullet points. I want TypeScript function signature format. This is going to allow us to understand the tools available to us. And specifically, I want to see the 11 Labs tools. Great. So, here are all the available tools. I'm just going to copy these out so I can quickly search. I need two things. Text speech and we probably need some way to play, right? So, I'm going to search text forwardlooking speech. text to speech. So I'm just going to paste that here. And we also want to find a play. There it is. Play audio. Great. Okay. So we have these two methods that we can use. And we actually want the entire definition so that we can know the exact parameters that we're playing with. Right. Let's paste this all in mode back to text format real quick. Great. Now let's go ahead and just run these. Right. So this is in our primary agents context window. I'm just going to say 11 labs. I've completed. Next, we can and I think we have voice ID here. Let me quickly copy one of my favorite 11 Labs voice IDs. Voice ID. Just going to be super clear in the prompt. Output directory. Right. Output. Great. Output directory. Output to pwd, which is current working directory/output. I'm going to have cloud code opus fire off this tool here. And all I'm doing here is validating the workflow. I'm going to encode into an agent. So, text to speech. Great. Save the file. And now we want to run. All set and ready for your next step. >> Great. Now we're going to run play. >> Where did he get that list of MCP servers? I think it just has uh a custom command that lists all tools. I'm going to link you the video afterwards, but I think he has a custom commands that just lists all his tools on all his uh uh MCP servers and other tools. >> Audio. And I should just be able to say play audio because all the context is loaded. Um, our agent is prime. Okay, that was good. I was talking over it. So, I'm gonna run it again. Run again. >> I've completed XYZ. Next weekend, ABC. >> Great. So, we can see that we have text to speech working. We now have full capability to have our agents communicate with us via a sub agent. Okay. So, fantastic. So, now we can crank open our powerful metaprompt. Okay. So now that we know the workflow and the work that we want done, we can build a new agent that encapsulates this work. So we can open this up and we can see exactly what our meta agent is doing. It has a system prompt that details how to create a new agent, right? It's doing all this work in the isolated context window. Okay, so let me see. I guess he's going to use his meta agent as like a a builder for this 11 Labs agent, I guess. Yeah, it has to be the way agent. >> I guess that's pretty cool, right? It's doing all this work in the isolated context window. And so I'll fast forward through this. I'm going to type up the prompt right here. So this is what it's going to look like, right? Build a new sub agent. So I'm asking Cloud Code to build the agent. And you can see here uh generate a new complete clock sub aent configuration file you from a user's description. Okay. So use this to create new agents proactively. So we're using the information dense keyword encoded by anthropic and that we should probably make this uppercase just to make it more dramatic. When a user asks you to create a new sub aent. Okay. So we want to make sure that we have that language inside of our prompt. Build a new sub aent. Great. We're defining when it's proactively triggered based on our conversation so far. Our agents will have blah blah blah blah blah. I'm just detailing exactly the flow that we went through. Okay. Get the current working directory. Text of speech. Play. Great. So again, copy this. Let's fire this off. And now you're going to see the description of our meta agent here. Get activated. This is very important. You need your description to properly set up when your agent should call a specific sub aent. So you can >> Yeah. Uh, I just used I can see why he wants to use it this way if he needs kind of a very structured way, but uh, let me see here. Completes code sub agent confiration for a file. Yeah, I guess the subscription use this proactively when the user asks you to create a new sub agent. Yeah, I guess that's fine. But uh you could you could just use the slash agent too. I guess but I guess you can use I I see >> cloud code is thinking about this. It knows and now it's going to use the sub agent. So couple key pieces inside of this prompt. I'm not just having this build on zero information. I'm repooling the cloud code documentation live. Okay. I want the most recent updated AI documentation. Viewers of the channel you know that I like to place this in AI docs. But there's a more powerful way to do this where inside of AI docs you just place a >> Yeah, that's something I've been doing for months now. Has to be a year. I always start my project with an AI docs folder. Uh but lately I've been kind of using cloud codes to use web fetch. But uh when I have docs I don't have to fetch it every single time when it's not in context. read me and then on the fly you have you know a prompt or now a sub agent pull in live documentation with some type of refresh command and so you can see we have a brand new agent generated there let's go ahead and understand what's happened great so three tool uses our meta a hello Toro okay so proactively is an anthropic keyword I didn't know that uh could you elaborate a bit do you know what what exactly it means >> agent >> success uccessfully created. Now it wants to read and verify. So this is great. So our reasoning model is double-checking the work. It's read the file and now it exists. >> All set and ready for the next step. >> Let's go ahead and check it out. This is a great summary. Voice proactive output autoplay. Looks good. Let's go ahead and look at our new agent. You can see it's in the exact format as we asked. Fantastic. So we can see here, you know, always review every word must add value. That's great. I like that line. No pleasant trees. Yep. Exactly. I'm going to add a couple of things. Right. So, variables, username, and what else we're going to do here? I want just one sentence. Uh, these can be quite long. A concise one sentence looks great. I want to make a couple tweaks here specifically to the description. This determines when your primary agent is going to call your sub agent. Cloud code has this IDK, this information keyword that they mention um, proactively, right? They even have a little section for it. Use proactively or must be used. >> Okay, so here it is. So to encourage more proactive sub agent use uh okay include face phrases like proactively or must be used in your description field. Ah when I see it now I kind of remember that to encourage more proactive sub agent use. So that is kind of for the primary agent then I guess so cloud code should always try to delegate more to the sub aents. What we can do here is something a little bit extra, right? You noticed in my hello agent, hello world agent, I like to have these concrete tags or these concrete triggers for my sub agents. So if they say X, Y, or Z, use this agent. So I'm basically just going to reuse this pattern. Frankly, I should encode this into my meta agent. I'm just going to add this here, right? If they say TTS, TTS summary, use this agent. review user prompt given to you a concise summary of what it does. By the way, um, one of the big reasons I'm using cursor still is for writing prompts. It's great for that tab completion. I also want to add one thing here. So, coming full circle here, remember that if we go back to our basic flow here, it's your primary agent that prompts your sub agent. What your primary agent can see and has access to is this description, right? This is how it knows when to call any given sub agent you have. So you >> So he's trying to use keywords. So he has like TTS, TTS summary, auto summary. Yeah. The proper detail instructions for agents become more and more important uh the more agents you have, right? Because you don't want the primary agent that is called code to call the wrong agent for a task. If you have like five different front end agent, that's going to be difficult to for cloud code to pick what kind of agent is correct. So I guess the more precise you are in your agent description, the better cloud code is going to be to delegate tasks to the correct sub agents. really want to leverage the description and tell your top level agent how to prompt this agent. When you prompt this agent, describe exactly what you want them to communicate to the user. And I'm going to add even more detail here. I'm going to say, remember, and again, this is something that I really just need to encode in my meta agent. Um remember this agent has no context any questions conversations between you and the user. Yeah. So of course the description this is for cloud code primary agent right? This is not for the sub agent. So this description part is of course only for cloud code kind of the primary delegation main agent. So yeah, that's I kind of like that inclusion here. >> Any questions, conversations between you and the user and something I like to do to increase prompt adherence. Um, Anthropic has another information dense keyword that you can use it is this. So important colon. I want to be absolutely clear with this agent. Do these following things, right? We want a concise summary of what was done. And also I want to add one more here. Best practices. important run only bash pwd and the 11 labs mcp tools don't use any other tools base your response exactly okay let's fire up another claw >> I guess that part you can just with sub agents you can uh also use the kind of the permission if you only want to give them certain tools but he's trying to do it this way, I guess. >> Base your response. Exactly. Okay, let's fire up another claw instance. We'll go ahead and just use Sonnet for this one. What does this codebase do? Use TTS summary. After we add details to when this should be called, after we add instructions to the system prompt for a work completion summary sub agent, we now have an operational agent that can quickly summarize in natural language for us anywhere, anytime on any piece of work. Right? This is the beauty of agents and reusable prompts. Now we have this problem solved for good and we can reuse it and improve it when we need to. codebase has been analyzed and now we're getting text to speech structure, text to speech response. There it is. There's an input prompt. >> This codebase demonstrates Claude code hooks mastery with all six life cycle events implemented for deterministic control featuring security filtering, intelligent TTS feedback, automatic JSON logging and UV single file architecture. Next step, explore specific hook implementations or test the security filtering functionality. >> Nice. Okay, so that looked good or that sounded good rather. You can see here there is work to be done here. That was still longer than I wanted it to be. If we open up logs, thanks to cloud code hooks, we are recording all the conversation that just happened. So, we can do something really cool. If we want to, we can dive in and actually see our top level agent prompting our sub agent and our sub agent prompting the response back to our top level agent. I'll leave that up to you after you finish watching this video. Run this. fire off the hello world agent and the meta agent so you can really understand what the flow of information is between multi- aent systems inside of cloud code. Yeah, I kind of want to check that out. I kind of want to see how how I wanted to see how cloud code actually prompts the sub agent and I kind of want to see uh what the sub agent gives uh its response to the primary agent. So the benefits are pretty straightforward, right? Cloud code engineers said it themselves. Context preservation. This is both a benefit and an issue. We'll talk about how this is an issue in a moment, but each sub agent operates in its own context. Preventing pollution of the main conversation. So this >> yeah that is kind of what I talked about. I think when each agent has kind of their own context window, this means that we can run the uh we can keep it running very long kind of the if we have a lot of sub agents we don't have to stop because of uh a full context window like I said many times now I ran like a 6h hour sub agent test I ran 6 hours >> this is powerful we are booting up fresh agent instances for every one of these tasks that we have to our natural language text of speech for our meta agent. They all have their own isolated context window. This is very powerful. On the channel, we're going to talk more about how you can use this to scale agents across your large complex code bases. You could already do this with sub aents, but now you can do it even better with specialized sub aents. All right, so you get to save your context window. This is a big idea that's going to come back up over and over again until we get those two 5 million plus context windows. You know, I can tell you those aren't coming uh anytime soon as far as I can see. I would love to be wrong about that. Specialized agent expertise. We can of course fine-tune the instructions and the tools. So, you know, you can see this here for our text to speech agent. You can see we wrote a nice rich description on not only when to trigger this, but um we're giving our top level agent instructions on how to prompt this. Okay. A lot of engineers are going to miss this. Don't be one. Yeah, this is what he meant by we're giving instructions to cloud code how to prompt the sub agent I guess of them you can instruct how exactly you want the prompt to flow in here in the description reusability this is a classic one by storing this inside your repository you can build agents for your codebase there's a powerful way we're going to discuss on the channel to use a meta agent to build specialized localized agents that excel at operating specific parts of your large codebase. Again, subscribe, comment, all that good stuff so you don't miss out on future valuable information like this. Flexible permissions, you can lock down the tools your agent can call. That's fantastic. Obviously, if you're in YOLO mode, you have to be more explicit about what tools can run, which is why I have this I added this important best practice here. But that's that. So, these are the kind of four key benefits that Enthropic lists directly here. I think these are all true. There are a couple hidden benefits as well. You can focus agents. All right. And you know the >> Yeah, try that. Matias, if you finish before we're done with this, let me know. >> Team does mention this, but you can really take this far, right? Just like booting up a fresh agent with a single prompt, when you use cloud code sub agents, the agent is fresh. It only knows what your primary agent tells it. That means it's less likely to make mistakes given that you designed a good system prompt. Why is that? It's because your agent is focused on one thing. Just like a focus engineer, when you're focused on one thing, you perform better. Full stop. Another hidden benefit of cloud code sub aents is simple multi- aent orchestration. And this is hello rapid. I was wondering how I can build an agent based on full documentation of a library. Just saying agent uh for this library. Uh, I want to use the best practices plan in the docs. Have you looked at uh the MCP server context 7? I can probably link it to take a look at context 7. Uh, I'm going to link it. So, if I go here, check this out. Could help you. This is something that I'm most excited for. With custom/comands combined with cloud code hooks combined with sub aents, right? just kind of stacking up these powerful features. You can build powerful yet simple multi- aent systems. So you know as a concrete example we have this in our commands we have this classic prime command and we can easily improve this right. So now we can have prime tts. So say I want to kick off prime I can say when you finish run the TTS summary agent the user know you're ready to build. I already have uh cloud code hooks with text speech. We did this in a previous video. I'll link that in the description if you're interested. But >> yeah, that's the one I saw. Uh just want to say I'm glad you're streaming when I got into the flow this morning. Yeah, thanks co. Thanks for hanging out. I like it. Cool. Uh for better efficiency, uh creator share memory so agents track each other. Cuts content size since they uh read instead of using uh keging prompts. Okay, that's pretty interesting. go to context 7, copy the whole docs of our library, put it into a file, and then the agent actually works with it locally instead of going to the MCP. Yeah, you could do that, but then I would really try to do to use the sub agents rapid so you don't fill up your context window, right? need to test more uh of the possibility to give sub agents proper context. Yeah, you could instruct the sub agents to use MCP servers too. So they could use the context 7 MCP. Yeah, Rapid I will look at sub agents and just make them read from uh your file. create a sub agents and just instruct them to read from your let's say you have a big file or something or like a part of it library I'm not quite sure we can of course you know run this now in a fresh instance we can run prime tts and now when this finishes we're going to run right we're going to chain text to speech summary agent at the end there it is and so here's the summary it wrote the text to speech so we have the audio file code hooks mastery project with complete coverage of all six hook life cycle events. >> There you go. So, it's ready to go. So, if you open up that that summary there, control E, you can see cloud code analyze hooks mastery blah blah blah blah blah. So, it's letting us know that it is complete. You know, we were able to guide our primary agents prompt to our sub agent right here. Right? When you finish, run the TTS summary agent and let the user know you're ready to build. So, this is another hidden benefit, right? We improve our multi- aent orchestration. Okay? And so another pro and con is prompt delegation. So again, this is kind of on the line. You're delegating your prompts to the primary agent. Oftent times, just like you saw here and just like you saw in the work completion summary agent, right, in the description, this means you have to do a little bit more work guiding your primary agent to call your sub agent properly. This is powerful though. It's still good. We are offloading work, right? We're encoding powerful engineering practices into our prompts and now into our sub aents. So all in all, it's good. So what are the issues of this? Right? The opposite of context preservation is that and and you know this was a great example because you kind of saw that you need to be super clear to your sub agents and to your primary agent what you're passing into. >> Yeah, I see what it means because the the sub agent doesn't have any context what happened earlier in the conversation since every sub agent comes in fresh with a clear context window. uh they don't know what's going on. So they need all that description or information from the primary agent, right? Has to be that sub agent because it doesn't have any previous context. By definition, every sub agent having its own context means that there is no context history. It doesn't have the rich context that your primary agent has. It has only what your primary agent prompts it with. This is the equivalent just to make it ultra clear. This is like firing up claude in print mode, you know, whatever. You can run yolo or dangerous blah blah blah. And don't run this, by the way, if you don't know what you're doing. Uh, this will run any command. We're starting to see in the industry, uh, yolo mode coders are getting cooked by this command. So, >> wasn't there something I guess that maybe that was replet, but some kind of vibe coing tool that deleted a full database? I think so. Uh I think it's okay for sub agents not to have all the context uh is optimal but the main agent should not lose context when getting sub agent responses. Yeah. So that is kind of I think what's pretty good with the main agent now is that it's not going to run out of context if you run the sub agents. I think used sub agents over a year ago with the MCPS are part of a research project. The real issue was the loss of efficiency because the agents uh yeah, I see what you mean. Uh are you are you still creating local doc references or fully swapped over to just context 7? It kind of depends. Night spider. I sometimes set up my MCP construct 7 uh uh yeah context 7 MCP but for sometimes I'm just creating local references if I just have a small project alpha says do you think the primary agent has all the sub agent context? Yeah, that was what I was wondering about. uh is there any loss there on the sub agent because it has to pass on its context to the primary agent right so I was kind of hoping he talked about maybe some loss on the the sub agents when it's passing on its context because it's not going to give the full context to the primary agent again hello 88 duski so the context 7 API is currently in the preview yeah I think it is but I really like what they are doing. And you can also if you don't want to use their MCP server, you could just copy their documentation like in a raw text file. Be really careful. Um, but this runs the following, right? So it's basically saying call work completion summary agent and then you're passing in one prompt. Okay, this is what your primary agent is doing when it's calling the sub agents. It's it's literally like it's you calling a oneshot prompt to the sub agent. Okay, there's no context. This is a problem, right? It's a problem and it's a benefit, right? It's the it's the opposite side of the coin of context preservation. Another big issue, these sub agents are hard to debug. You can see every time we've run Yeah, that was what I was thinking about. We can't really see everything that's going on, I guess, in the sub agent. At least I haven't seen a way to do that. Maybe he he knows something here. >> One of these, right? If we run high CC, even with this simple prompt, we have no idea what's going on here. We'll get the tool calls, which is super super nice. But the actual workflows, the prompts, the full parameters for every tool call, we don't get these. And this is by design, but it also makes these harder to debug and understand what's going on, right? So, another big issue here is decision overload. As you start scaling up the number of agents you have, it's going to be harder and harder for your agent to know which one to call. >> Sub agent complete. >> Okay. Commands are a bit different. Although, you know, you the engineer might forget all the commands that you have, right? You just might start forgetting all these commands, right? All the powerful >> Hello, Alexander. How how long does the sub agent retain the context once completion? Uh or does it loop uh back into or does it loop back into no context? I think to be honest I just think when one sub agent is uh done when you start it again it's just going to refresh right I don't know if any other think that I think that's what's going to happen when one sub agent has completed and it says done uh when we started again it's just going to be fresh again so we're going to rely on the main agent to keep hold of the context Next commands we've built. Agents are a little different because based on their description and based on the number of agents you have, your primary agent might get confused here. Okay, so this is important. You really want to be keeping track of your agents. Otherwise, it'll come back to bite you and you're going to kick off these agents when you don't want to. As a solution, you really want to be clear about when to call your sub agents with the description variable. This is the most important variable right next to the name. Two remaining issues here. Dependency coupling is going to be a problem here. So once you start setting up prompts, primary agent, reusable prompts or custom slash commands and once these are calling out your sub agents, which this is how you, you know, start scaling up into multi- aent systems. Once you get that going, it's going to be hard again back to debugging, it's going to be hard to understand what's going on. You're going to have dependency coupling. Inevitably, you're going to have an agent depend on output from another agent, depend on the format of the specific response from another agent, on and on and on. And then one day you're going to need to change something to improve it or to ship something new, whatever. You're going to change the planner agent right here and it's going to ruin everything else. >> Yeah, I see what it means. Because if this agent, let's say the reporter agent is dependent on kind of the the output from the planner agent. And if this change and this is expecting an output in this format, uh that is going to break, right? Uh Benjamin says, "Yeah, I think the sub edges just passes the output to the primary agent." Yeah, I think that's it. And when it's done, it's uh it's just you start it again. I think it's just going to be a new context window because one thing changed. We're already operating in non-eterministic systems. When that one thing changes, it could blow up everything else. Okay, so this is another problem with sub aents. As you start scaling them up, keep your eye on this. Try to keep them separate. Try to keep them in isolated workflows. Don't overload your sub aent chain. Okay. And then last thing, and this is just kind of a request on my side, even though it is literally counterintuitive to what I was just saying, you cannot call sub aents in your sub agents. Okay? Probably for all the reasons that I just mentioned. You know, it would be cool if we had, you know, kind of like dangerous um sub agent. True, right? So that >> Yeah, that's that that is a cool idea, right? But dude, wouldn't that how would that work? If your sub agents could call other sub agents, then it starts to be really hard to control that, right? inside of this agent or you know it's probably like sub sub aent true. So that inside of this agent we enable this dangerous powerful setting of calling sub aents inside of sub aents. So this is me just being picky. That's not actually a real issue but it's important to note that you know you can't call sub aents in your sub aents at least not yet. That's me being hopeful. So this is a very powerful feature. We have quite a few more ideas to explore here with the combination of agents and custom slash commands aka reusable prompts. Remember that perspective matters as you start scaling up your multi- aent system, right? The flow of the big three, context, model, and prompt, matter more than ever, and more and more as you scale up the number of agents who have shipping work on your behalf. There's a reason why it's a principle of AI coding. As usual, link in the description if you want to learn the other principles of AI coding that will help you differentiate what you can do with agentic coding. And I'll go ahead and bring back the other uh prompts that I had here just for fun. But the meta agent is going to be here available to you. Link in the description. I'm going to add these to the cloud code hooks mastery codebase. If you made it to the end, be sure to like, subscribe, and comment your thoughts. No matter what, stay focused and keep building. Yeah, I think it's a good video. I think it's good if I'm going to link it. I've been using Claude code for a month. >> Sorry about that. Uh, I'm going to link the video. Just give me a second. And if you I think this is a great channel if you kind of want to dive deep into cloud code. I think it mostly focuses on cloud code. So yeah, that's my alarm. I have to I have a meeting in 30 minutes. Staying focused. That's the challenge. Yeah. Yeah, I agree. Uh let's ask clone primary agent if he can access one of your sub agents context. Okay, I see what you mean. Uh each model has its strength and weaknesses. Giving sub aents their own MCPs or sub sub aents broadens their view and make responses more complete and efficient. Yeah, you can there's no problem giving uh the MCPS at least they can use that but sub sub agents it would be cool to try it out. I would if uh it was possible. Uh but yeah, pretty good video. I liked it. Uh I think my takeaway is that uh I really like what he said about the description part. So if we look at my my simple agent here, Anna, we could kind of give some instructions to to the primary agent. Always uh remember that the agent does not have any context of previous conversation. update the agent with the latest information. So, if I do that and I go to my agents here, uh I can do like a restart Claude. And I'm just going to say if I say to Claude now, hello. Uh, my name uh is Chris and I do agent Anna just going to see. Okay, so I can't really see what the agent is putting in here. Uh, but that's fine. I I might look into that how we can do that. Uh, but yeah. Uh today we had a look at um the new the new um rate limits for cloud code. Pretty interesting conversation about that. Uh a lot of people were a bit disappointed but not surprised. Uh we had a look at some hooks and some agents. Uh I'm still trying to learn uh more about sub agents and hooks. So I'm going to continue a bit with that. I also have some plans at looking at claude flow uh when I have time and we're going to watch the new video. Uh I thought it was pretty interesting. I think I learned something from it. So yeah, thanks Spider. Thank you for tuning in. Uh probably wouldn't have done much live testing on CL code while watching. That's good. Yeah, I appreciate everyone tuning in. Uh it's always nice to hear what other people really thought about this new weekly limits and yeah, thanks for all the chats and the engagement. Uh like I said, this week uh I'm uh uh probably going to do a few more streams. Uh yeah, I think it's just fun. Okay, Benjamin, thanks for tuning in. Uh good night, I guess. Thank you, Sunni, as always for everyone tuning in. Appreciate it. So, yeah, exciting times ahead. We're looking out for GD5 soon. Hopefully we can give Anthropic some uh more competition on the coding side so maybe we can get the prices down. Alexander, thanks for tuning in. Appreciate it. Bye Carl. Good night. So yeah, uh I might do a new stream tomorrow. Let's see if I have time. Other than that, enjoy the rest of your day or night or morning, I guess, if you're on the west side of the world. And hopefully I see you again pretty soon. So, yeah, thank you everyone for tuning in. Have a great day and we speak soon.