{
  "text": "If you're watching this video, then I'm going to guess that you're somebody who's been running into plan limits on your Claude subscription.\nWell, the good news for you is that I'm about to show you 11 ways so you can make sure you never run into those usage limits again.\nAnd there's a key fundamental principle that most people totally overlook, and it will change the way that you interact with Claude code forever.\nSo the first thing which you need to understand is that Claude code does not actually remember anything at all.\nSo the way this actually works is for every new message you send away to CloudCode, it's not only receiving that latest message, it's also receiving everything prior to that as well.\nSo if you've had 50 interactions back and forth with CloudCode in your session, then on the 51st message, it's receiving all of that context again.\nAnd you might be thinking, well, that would be chewing through millions of tokens, and it would.\nAnd that brings us on to the idea of caching.\nSo caching is how Claude Code knows what information already existed in the session versus what is new.\nAnd this is what stops us from burning through our session limit after maybe only 20 or 30 minutes.\nSo on the first turn, so your first interaction in that session, Claude is going to receive the system prompt, the project context and your initial message.\nThen on turn two, Claude knows that all of this information from the first turn already existed.\nSo all you're going to be charged for on turn two is Claude's response and then your next prompt.\nAnd it's the same for turn three.\nIt then knows that all of the information from turn two was already there.\nSo you're only being billed for the new stuff at the end.\nAnd everything which happens before this reply, all of this is known as the prefix.\nYou'll see in a second why this is so important when we get into the tips.\nbut for now just understand that if something in the cash data or something in the prefix changes\nthen you're going to be paying for all of this a second time which means if you have 500 or 600\nthousand tokens sitting in your context window and something in the prefix has changed then you're\ngoing to be paying for all of that again. So all of that is to say that Claude does not remember\nwho you are between turns. Every new message it's sending everything back to Claude again,\nit's going to read through it, it knows what was already there, it knows what is new, and you're\nonly going to be billed full price for the new stuff. Now if you're using it on the API, we know\nthat cached inputs are 0.1x of the initial input cost. CLODs do not disclose publicly in their\ndocumentation what that multiple is for planned subscriptions, but we do know that we're not paying\nfull price for those tokens. Again it's going to be some form of fraction of the initial input cost.\nNow that we've cleared up that important distinction everything from here on will make much more sense\nso let's move into the first tip which is keeping all of your important context inside of your\nclod.md file. A clod.md is just a markdown file which lives inside of your project and it gets\nauto-injected into your context window at every new session start in that project. So the reason\nthis is so important is if you're not using a clod.md file then at the start of every new session\nyou need to explain who you are, what your tech stack is, where your files go, what your rules are\nand then you need to give it your task or your initial prompt. Whereas if you are using a clod.md\nall of this context is already stored in there so clod already has all of that information.\nYou're not wasting tokens going back and forth explaining the same thing every time. So it won't\njust save you tokens, it will also save you time. You can see my clod.md inside of my project here.\nNow if you want to create one, you can run slash command slash init and you can see here,\ninitialize a new clod.md file with code-based documentation or you can just describe in plain\nEnglish that you want it to create you a clod.md and it will go ahead and do that for you. Tip\nnumber two is keeping your context below 20 to 25%. If Opus 4.8 and these other new anthropic\nmodels all have a million tokens of context then why would we aim to keep things sub 25 percent\nwell there's a multitude of different reasons for this but i'll just cover a couple of them now\nand the next three tips after this one are all related back to this same issue so this is very\nvery important so the first one is as i mentioned in the intro when we send our prompts the first\ntime around to the clod api in clod code we're being charged 100 of the cost for those tokens\nSo we're being billed one token for every token. Now after that, once those tokens are\ncashed and they become part of the prefix, we're still paying for those tokens. And we\ndon't know what that cost is for Anthropic subscriptions or Claude plans, but we do know\nthat those tokens are not free. So even if it's 0.05x of the initial cost, so 1 20th of a token,\nif you have 500,000 tokens in your context window, then you're paying 25,000 tokens per turn,\nand that's going to be increasing with every new turn just in cached reads. So that's one reason.\nNow another reason that you would want to keep your context sub 20% is we know that these models\nstart degrading in performance the more you start to fill this context up. So what that means is if\nyou have 800,000 tokens, so 80% of your context window used, then the model is not going to be\nperforming optimally as it would if we only had 150,000 tokens in that context window. You can\nsee down here in my status line that I have my context usage as a bar and a percentage listed\nthere which means I can always see it at a glance. Now I would recommend that you do the same and I\nwill put a prompt on screen which you can copy and give to Cloud Code so that it can implement\nthe same thing for you. If you want to keep an eye on your context just run forward slash context\nand it will bring up this visual where you can see what you've used. You can see it broken down here\non the right hand side as well and it breaks it down below. When you start approaching that\n20 to 25% limit, I recommend that you ask Claude to generate a handoff file, which is just a markdown\nfile with a summary of everything you've done in that session, and also give you a session resume\nprompt. So you can then give that session resume prompt back to Claude in the new session, or after\nyou've cleared your context, have it read the handoff file, and you can pick up exactly where\nyou left off. And if you want to go one step further, I built this into a skill, I just called\nit handoff so all I need to do is run forward slash handoff and it will do all of that for me.\nIf you want access to the skill you will find all of the resources from the video in the pinned\ncomment down below. Tip number three is making sure that you keep your prompts within one hour\nof the previous prompt. This is one of the simplest and most common mistakes that I see people making\ninside of Claude Code and it will chew through your five hour session limit and your weekly quota\non your Claude subscription. So we're looking at the Claude Code documentation here and you can see\nthat it says Claude Code requests the one hour TTL automatically. So we're talking about the\ncaching again here. So when you use Claude Code, every time you send a new message away, that\nresets the one hour cache. Now if you go out with that one hour, then the cache is then invalidated\nand your next prompt, you're going to be paying for all of those tokens again. So if you have\n500,000 tokens sitting in your context window, which you shouldn't have if you're following step\ntwo, then if you go away for an hour and you come back, you send your next prompt to Claude,\nyou're paying full price for those 500,000 tokens again. If you then go away for another hour and\ncome back and then send your next prompt, you've now paid for those 500,000 tokens three times.\nSo this is why it's super important that you keep it within an hour of your previous prompt\nand going back to the previous tip that you keep your contacts below 20%\nso if you do go out with that hour you're not burning 500,000 tokens every single time.\nmodel mid-session. How many times have you heard that you should be changing your model from Sonnet\nto Opus for more complex tasks and then back to Sonnet again for the easier stuff? And I'm going\nto tell you now that that is bad advice. We're staying inside of the Cloud Code docs and you can\nsee here actions that invalidate the cache. Number one on the list is switching models. So if we\nscroll down you can see that it says each model has its own cache. Switching with forward slash\nmodel means the next request reads the entire conversation history with no cache hits even\nknow the content is identical. So I see people making this mistake all the time. They're constantly\nswitching models between Opus, Sonnet, Sonnet, Opus, maybe Fable, and you're paying for all those\ntokens again every single time you make that switch. I'm not saying that you should not be\nswitching models at all because that is genuinely good advice to use the right model based on the\ntask at hand. What I am saying is that you should not be switching model mid-session. You should be\nputting together a plan using your higher power model, whether that's Opus or whether it's Fable,\nand then once you have a structured plan in place, switch your model down to whatever it is that\nyou're going to be using to execute that task, and do not switch again unless you absolutely\nhave to. Number five is almost identical, and that is avoiding changing effort level\nmid-session. Right below switching models and actions that invalidate the cash, we have changing\neffort level. So you can see here that it says that the cache is keyed by effort level as well\nas model so switching with forward slash effort means that the next request reads the entire\nhistory again with no cache hits. So this is the exact same as the previous tip if you can avoid\nchanging effort level mid-session then you should absolutely be doing that. Number six is building a\nproper memory system. I have a full video on my channel breaking down how you can set this system\nup for yourself. I'm not going to go too deep here but I will link that video on screen. But what\nwe're talking about doing here is building a proper memory system so that Claude Code has a\nworking map of our project and it always knows where to go to find the information that we're\nasking for. And we can do that by using something like Obsidian. And if you don't know what Obsidian\nis, it's just a software that allows us to view and edit marked-in files. And you can see here an\nexample of my obsidian knowledge graph you get this really cool graph we can kind of like pull\naround all these different nodes all these different lines that you see are connections\nbetween different documents so it can piece together different bits of information work\nbetween different documents to find the right information to give you a high level overview of\nhow this works we create a folder somewhere on our computer and we then point obsidian at this\nfolder and then that then becomes our vault now once we have our vault we can then point cloud code\nat this vault and inside of our vault we will have a bunch of different folders in there. Now these\nfolders can be whatever you like. For me I like to split this into three separate layers. So the\nfirst layer is my working folders. So for me that's things like agency, school, content and inside of\nthese folders I will have subfolders. So for agency that will be things like clients, it could be\ndeliverables, outreach, pipeline, for school, it might be frameworks, strategy, courses. So that is\nour first layer, all of the working files that you'll be in there messing around with on a day\nto day basis. The second layer here originates from Andre Carpathie's Obsidian RAG system. I'm\nnot going to go into that here so make sure you check out that video linked on screen if you're\ninterested in that. And the third layer is our clod.md. So in there as we covered in tip number\none. This is where we store our project map, our rules, our context, so that Claude knows everything\nwhich it needs to know about us at session start without us having to explain every time. So how\nthat relates to our memory system here is, first of all, we give Claude a map of the project folders,\nwhat's contained in each folder, so that it knows where things are. Then we need to give it a map\nof where it needs to put things when we ask for them to be saved. So you can see here, anything\nfor a client goes in agency clients and then the client's name. If it's an idea or a script or a\nhook it goes in content. So we give it the information which it needs so that things don't\nget lost when it saves them. And finally we give it the knowledge base rules. So this is how it\nknows how to operate our knowledge base system and this is arguably the most important part.\nYou can see here we have information on wiki links to make sure that it's always creating\nthose links between different documents and that is how we get our nice graph here inside of\nobsidian.md. Number seven is creating custom skills. This one is hands down the most powerful\ntip in this video, not only from a token saving perspective, but also in maximizing the value\nwhich you get out of cloud code for you specifically. So creating a custom skill is just packaging up\nsome amount of rich information on a specific topic that's specific to you so that it's repeatable\nand Claude can do it over and over again without us having to go back 20 different times, change\nthis, change that, because Claude already knows what a good output looks like. Now I would highly\nrecommend that you install the skill creator skill by Anthropics. This is an official Anthropics skill\nand it tells Claude Code how to build custom skills. So first of all you're going to explain\nwhat it is that you're trying to do. Then Cloud Code is going to ask you a bunch of questions,\nit's going to test it, it's going to run evaluations, it's going to make sure that it\ntriggers when it should, and it's going to make sure that you're happy with the output. So then\nall you need to do when you want to use it going forward is run a slash command and then that task\nis now repeatable on the mat. So the reason that this is so powerful is without a skill, you ask\nit to do something, you then have to go back saying no that's not right, you might need to go back\nagain and again and again and eventually you will have a product that you're happy with but by that\npoint you've burned a bunch of tokens and if you're doing this multiple times a week then I can\nguarantee you're wasting a huge amount of your plan limits on tasks that are repeatable. So what I\nwould suggest is that you copy the prompt which you see on screen and you have Claude go through\nyour last 10, 15, 20 sessions and pick out things which you've asked it to do multiple different\ntimes and tell it to build those into skills. To install the skill creator skill just run\nforward slash plugins and then on the discover tab here you will see it listed fourth from the top.\nYou can see skill creator create new skills, improve existing skills and measure skills as\nwell. So all you need to do is enter this and install it. I would recommend that you go for\nuser scope which means you'll have access to across all of your different projects.\nNumber eight is using Caveman.\nCaveman is a plugin that makes Cloud Code less verbose.\nSo we know that these large language models love to talk.\nYou ask it a simple question and it gives you a huge reply,\nwhich will take you five minutes to read.\nAnd Caveman was designed to try and stop that from happening.\nAnd it actually does a very, very good job.\nI've been using it on and off now for around about three or four months.\nAnd it is very effective,\nespecially if you're not wanting a big walls of text back from the model.\nSo it says here, why use many token when few do trick?\nAnd that is kind of how it works.\nMake your AI coding agent talk like a caveman.\nSame answers, up to 65% fewer output tokens.\nNow for me personally, I probably haven't seen as much as 65% fewer output tokens,\nbut it does definitely reduce the token output.\nYou can see an example here.\nI said, explain how prompt caching reduces cost in LLMs.\nand this was the response that I got using Caveman.\nNow if we compare that to this other terminal,\nI gave it the exact same prompt,\nexplain how prompt caching reduces cost in LLMs,\nand...\nreduces cost in LLMs and the response which I got from this one was much much longer. You can see\nthe difference here in the size of the response. So this is definitely one that you should be trying\nout. Number nine is using ponytail. This one has been blown up over the past few weeks. You can see\nthat it's number one repository of the day and of the week and it was designed because these AI\ncoding agents love writing unnecessary amounts of code and ponytail is essentially designed to make\nyour AI coding agent as lazy as possible without making it complacent. So if we look at the\nbenchmarks here you can see the impact which this has. So this is versus a no skill baseline,\nminus 54% on lines of code, minus 22% on tokens, minus 20% on cost, and minus 27% on time. And if\nwe keep scrolling down here, you can see this hierarchy of how it works. So before writing any\ncode, the agent will stop at the first rung that holds. The first thing which it's going to do is\ncheck, does this even need to exist? Only if that is a yes, is it then going to see if there's\nsomething already in the code base that can handle it. And if that is a no, it's going to see,\ncan I handle this with one single line of code? And only if that is a no, is it then going to\nwrite the minimum amount of code to get the job done. Number 10 is using the Codex plugin for\nCloudCode. This is an amazing plugin if you're someone who's already paying for an OpenAI\nsubscription. So it allows us to use Codex inside of the CloudCode ecosystem, which means we can\noffload tasks from the Anthropic models onto the GPT models. Now we know that these GPT models are\nsubstantially cheaper than the Anthropic models and they're now very very competitive as well.\nGPT 5.6 has just been released so that is definitely something that you can take advantage of\nif you're already paying for an OpenAI plan. Once you have the plugin installed just do\nforward slash codex and you will see all the different slash commands which you now have\navailable. One which you should definitely check out is Codex Adversarial Review, which is going to\nhave the Codex agent do a deep dive into your code base to see if it can find any security\nvulnerabilities. This is definitely one that you should check out. Go in there, have a play around\nwith it, and this will save you a bunch of tokens. And finally, tip number 11 is offloading research\nheavy tasks to Google servers using Notebook LMPi. Notebook LMPi allows us to use Google servers to\ncarry out any kind of deep web research. We can also use it to generate infographics and slide\ndecks and a bunch of other cool stuff as well and we can do this all from the command line\ninside of Cloud Code. Now one of the main benefits of this is that we know that research flows,\nsub-agents, agent teams, all these things are super token heavy and we don't really want to\nbe chewing through our Cloud subscription on those tasks and that is when we can offload those tasks\nto Google servers essentially for free using Notebook LMPi and one of the massive benefits\nto this is because it's Gemini under the hood it has access to all of the YouTube content\ntranscripts and all of that rich information which you normally would not be able to surface\njust using Claude Code alone. So I asked Claude Code to use the Notebook LMPi CLI to carry out\nresearch on GPT 5.6 SOL versus Claude Fable 5 and then turn that into an infographic. So it went\nthrough the process here and this is what we got at the end. Now bear in mind I did not need to do\nanything for this. I did not use any of my clod tokens or my plan limits and this is what we got\nat the end. So this is an awesome tool and you should definitely be leveraging this to save\nyourself a bunch of tokens. That is everything for this video today. If you got some kind of\nvalue out of this then please hit subscribe. I would massively appreciate it and make sure that\nyou check out the pinned comment for all of the free resources. You'll also find a full clod code\nmasterclass in the school which will take you from complete beginner to proficient\ncloud code user so thanks for watching i'll see you in the next one",
  "language": "en",
  "duration": 1199.194563,
  "segments": [
    {
      "start": 0.0,
      "end": 6.28,
      "text": "If you're watching this video, then I'm going to guess that you're somebody who's been running into plan limits on your Claude subscription.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07765346716258152,
      "no_speech_prob": 1.279362791921368e-12,
      "compression_ratio": 1.7207792207792207
    },
    {
      "start": 6.88,
      "end": 13.6,
      "text": "Well, the good news for you is that I'm about to show you 11 ways so you can make sure you never run into those usage limits again.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07765346716258152,
      "no_speech_prob": 1.279362791921368e-12,
      "compression_ratio": 1.7207792207792207
    },
    {
      "start": 13.86,
      "end": 21.12,
      "text": "And there's a key fundamental principle that most people totally overlook, and it will change the way that you interact with Claude code forever.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07765346716258152,
      "no_speech_prob": 1.279362791921368e-12,
      "compression_ratio": 1.7207792207792207
    },
    {
      "start": 21.76,
      "end": 27.26,
      "text": "So the first thing which you need to understand is that Claude code does not actually remember anything at all.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07765346716258152,
      "no_speech_prob": 1.279362791921368e-12,
      "compression_ratio": 1.7207792207792207
    },
    {
      "start": 27.26,
      "end": 37.44,
      "text": "So the way this actually works is for every new message you send away to CloudCode, it's not only receiving that latest message, it's also receiving everything prior to that as well.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09532206633995319,
      "no_speech_prob": 1.5920301101729484e-12,
      "compression_ratio": 1.7326007326007327
    },
    {
      "start": 37.9,
      "end": 46.72,
      "text": "So if you've had 50 interactions back and forth with CloudCode in your session, then on the 51st message, it's receiving all of that context again.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09532206633995319,
      "no_speech_prob": 1.5920301101729484e-12,
      "compression_ratio": 1.7326007326007327
    },
    {
      "start": 47.22,
      "end": 51.94,
      "text": "And you might be thinking, well, that would be chewing through millions of tokens, and it would.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09532206633995319,
      "no_speech_prob": 1.5920301101729484e-12,
      "compression_ratio": 1.7326007326007327
    },
    {
      "start": 51.94,
      "end": 54.3,
      "text": "And that brings us on to the idea of caching.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09532206633995319,
      "no_speech_prob": 1.5920301101729484e-12,
      "compression_ratio": 1.7326007326007327
    },
    {
      "start": 54.3,
      "end": 61.6,
      "text": "So caching is how Claude Code knows what information already existed in the session versus what is new.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12127772499533261,
      "no_speech_prob": 8.489604960866537e-13,
      "compression_ratio": 1.7509578544061302
    },
    {
      "start": 62.02,
      "end": 67.68,
      "text": "And this is what stops us from burning through our session limit after maybe only 20 or 30 minutes.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12127772499533261,
      "no_speech_prob": 8.489604960866537e-13,
      "compression_ratio": 1.7509578544061302
    },
    {
      "start": 68.26,
      "end": 77.46,
      "text": "So on the first turn, so your first interaction in that session, Claude is going to receive the system prompt, the project context and your initial message.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12127772499533261,
      "no_speech_prob": 8.489604960866537e-13,
      "compression_ratio": 1.7509578544061302
    },
    {
      "start": 77.78,
      "end": 84.0,
      "text": "Then on turn two, Claude knows that all of this information from the first turn already existed.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12127772499533261,
      "no_speech_prob": 8.489604960866537e-13,
      "compression_ratio": 1.7509578544061302
    },
    {
      "start": 84.3,
      "end": 90.24,
      "text": "So all you're going to be charged for on turn two is Claude's response and then your next prompt.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.106499500102825,
      "no_speech_prob": 1.0523894039202175e-12,
      "compression_ratio": 1.65234375
    },
    {
      "start": 90.7,
      "end": 91.96,
      "text": "And it's the same for turn three.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.106499500102825,
      "no_speech_prob": 1.0523894039202175e-12,
      "compression_ratio": 1.65234375
    },
    {
      "start": 92.32,
      "end": 95.84,
      "text": "It then knows that all of the information from turn two was already there.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.106499500102825,
      "no_speech_prob": 1.0523894039202175e-12,
      "compression_ratio": 1.65234375
    },
    {
      "start": 96.2,
      "end": 99.36,
      "text": "So you're only being billed for the new stuff at the end.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.106499500102825,
      "no_speech_prob": 1.0523894039202175e-12,
      "compression_ratio": 1.65234375
    },
    {
      "start": 99.36,
      "end": 104.5,
      "text": "And everything which happens before this reply, all of this is known as the prefix.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.106499500102825,
      "no_speech_prob": 1.0523894039202175e-12,
      "compression_ratio": 1.65234375
    },
    {
      "start": 104.84,
      "end": 109.12,
      "text": "You'll see in a second why this is so important when we get into the tips.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.106499500102825,
      "no_speech_prob": 1.0523894039202175e-12,
      "compression_ratio": 1.65234375
    },
    {
      "start": 109.12,
      "end": 114.8,
      "text": "but for now just understand that if something in the cash data or something in the prefix changes",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07775270616686022,
      "no_speech_prob": 1.2944811238549403e-12,
      "compression_ratio": 1.867704280155642
    },
    {
      "start": 114.8,
      "end": 121.06,
      "text": "then you're going to be paying for all of this a second time which means if you have 500 or 600",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07775270616686022,
      "no_speech_prob": 1.2944811238549403e-12,
      "compression_ratio": 1.867704280155642
    },
    {
      "start": 121.06,
      "end": 126.32,
      "text": "thousand tokens sitting in your context window and something in the prefix has changed then you're",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07775270616686022,
      "no_speech_prob": 1.2944811238549403e-12,
      "compression_ratio": 1.867704280155642
    },
    {
      "start": 126.32,
      "end": 130.86,
      "text": "going to be paying for all of that again. So all of that is to say that Claude does not remember",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07775270616686022,
      "no_speech_prob": 1.2944811238549403e-12,
      "compression_ratio": 1.867704280155642
    },
    {
      "start": 130.86,
      "end": 136.52,
      "text": "who you are between turns. Every new message it's sending everything back to Claude again,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07775270616686022,
      "no_speech_prob": 1.2944811238549403e-12,
      "compression_ratio": 1.867704280155642
    },
    {
      "start": 136.52,
      "end": 141.16,
      "text": "it's going to read through it, it knows what was already there, it knows what is new, and you're",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08132633360305636,
      "no_speech_prob": 1.1649218915887882e-12,
      "compression_ratio": 1.6099585062240664
    },
    {
      "start": 141.16,
      "end": 147.06,
      "text": "only going to be billed full price for the new stuff. Now if you're using it on the API, we know",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08132633360305636,
      "no_speech_prob": 1.1649218915887882e-12,
      "compression_ratio": 1.6099585062240664
    },
    {
      "start": 147.06,
      "end": 154.7,
      "text": "that cached inputs are 0.1x of the initial input cost. CLODs do not disclose publicly in their",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08132633360305636,
      "no_speech_prob": 1.1649218915887882e-12,
      "compression_ratio": 1.6099585062240664
    },
    {
      "start": 154.7,
      "end": 160.92,
      "text": "documentation what that multiple is for planned subscriptions, but we do know that we're not paying",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08132633360305636,
      "no_speech_prob": 1.1649218915887882e-12,
      "compression_ratio": 1.6099585062240664
    },
    {
      "start": 160.92,
      "end": 166.98,
      "text": "full price for those tokens. Again it's going to be some form of fraction of the initial input cost.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07042324542999268,
      "no_speech_prob": 1.0772408355966268e-12,
      "compression_ratio": 1.7314487632508835
    },
    {
      "start": 167.1,
      "end": 171.5,
      "text": "Now that we've cleared up that important distinction everything from here on will make much more sense",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07042324542999268,
      "no_speech_prob": 1.0772408355966268e-12,
      "compression_ratio": 1.7314487632508835
    },
    {
      "start": 171.5,
      "end": 176.92,
      "text": "so let's move into the first tip which is keeping all of your important context inside of your",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07042324542999268,
      "no_speech_prob": 1.0772408355966268e-12,
      "compression_ratio": 1.7314487632508835
    },
    {
      "start": 176.92,
      "end": 183.42,
      "text": "clod.md file. A clod.md is just a markdown file which lives inside of your project and it gets",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07042324542999268,
      "no_speech_prob": 1.0772408355966268e-12,
      "compression_ratio": 1.7314487632508835
    },
    {
      "start": 183.42,
      "end": 189.46,
      "text": "auto-injected into your context window at every new session start in that project. So the reason",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07042324542999268,
      "no_speech_prob": 1.0772408355966268e-12,
      "compression_ratio": 1.7314487632508835
    },
    {
      "start": 189.46,
      "end": 194.66,
      "text": "this is so important is if you're not using a clod.md file then at the start of every new session",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.06481707480645948,
      "no_speech_prob": 1.405132629174466e-12,
      "compression_ratio": 1.864864864864865
    },
    {
      "start": 194.66,
      "end": 200.34,
      "text": "you need to explain who you are, what your tech stack is, where your files go, what your rules are",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.06481707480645948,
      "no_speech_prob": 1.405132629174466e-12,
      "compression_ratio": 1.864864864864865
    },
    {
      "start": 200.34,
      "end": 205.98,
      "text": "and then you need to give it your task or your initial prompt. Whereas if you are using a clod.md",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.06481707480645948,
      "no_speech_prob": 1.405132629174466e-12,
      "compression_ratio": 1.864864864864865
    },
    {
      "start": 205.98,
      "end": 210.96,
      "text": "all of this context is already stored in there so clod already has all of that information.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.06481707480645948,
      "no_speech_prob": 1.405132629174466e-12,
      "compression_ratio": 1.864864864864865
    },
    {
      "start": 211.52,
      "end": 216.16,
      "text": "You're not wasting tokens going back and forth explaining the same thing every time. So it won't",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.06481707480645948,
      "no_speech_prob": 1.405132629174466e-12,
      "compression_ratio": 1.864864864864865
    },
    {
      "start": 216.16,
      "end": 221.48,
      "text": "just save you tokens, it will also save you time. You can see my clod.md inside of my project here.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08469312973604858,
      "no_speech_prob": 1.0320223408494256e-12,
      "compression_ratio": 1.6797153024911031
    },
    {
      "start": 221.88,
      "end": 226.62,
      "text": "Now if you want to create one, you can run slash command slash init and you can see here,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08469312973604858,
      "no_speech_prob": 1.0320223408494256e-12,
      "compression_ratio": 1.6797153024911031
    },
    {
      "start": 227.06,
      "end": 232.56,
      "text": "initialize a new clod.md file with code-based documentation or you can just describe in plain",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08469312973604858,
      "no_speech_prob": 1.0320223408494256e-12,
      "compression_ratio": 1.6797153024911031
    },
    {
      "start": 232.56,
      "end": 237.4,
      "text": "English that you want it to create you a clod.md and it will go ahead and do that for you. Tip",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08469312973604858,
      "no_speech_prob": 1.0320223408494256e-12,
      "compression_ratio": 1.6797153024911031
    },
    {
      "start": 237.4,
      "end": 244.4,
      "text": "number two is keeping your context below 20 to 25%. If Opus 4.8 and these other new anthropic",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08469312973604858,
      "no_speech_prob": 1.0320223408494256e-12,
      "compression_ratio": 1.6797153024911031
    },
    {
      "start": 244.4,
      "end": 250.34,
      "text": "models all have a million tokens of context then why would we aim to keep things sub 25 percent",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05030380215561181,
      "no_speech_prob": 1.1115400050040747e-12,
      "compression_ratio": 1.707142857142857
    },
    {
      "start": 250.34,
      "end": 254.74,
      "text": "well there's a multitude of different reasons for this but i'll just cover a couple of them now",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05030380215561181,
      "no_speech_prob": 1.1115400050040747e-12,
      "compression_ratio": 1.707142857142857
    },
    {
      "start": 254.74,
      "end": 260.28,
      "text": "and the next three tips after this one are all related back to this same issue so this is very",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05030380215561181,
      "no_speech_prob": 1.1115400050040747e-12,
      "compression_ratio": 1.707142857142857
    },
    {
      "start": 260.28,
      "end": 266.08,
      "text": "very important so the first one is as i mentioned in the intro when we send our prompts the first",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05030380215561181,
      "no_speech_prob": 1.1115400050040747e-12,
      "compression_ratio": 1.707142857142857
    },
    {
      "start": 266.08,
      "end": 271.88,
      "text": "time around to the clod api in clod code we're being charged 100 of the cost for those tokens",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05030380215561181,
      "no_speech_prob": 1.1115400050040747e-12,
      "compression_ratio": 1.707142857142857
    },
    {
      "start": 271.88,
      "end": 277.88,
      "text": "So we're being billed one token for every token. Now after that, once those tokens are",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10061746258889476,
      "no_speech_prob": 1.309704731819461e-12,
      "compression_ratio": 1.696629213483146
    },
    {
      "start": 277.88,
      "end": 282.24,
      "text": "cashed and they become part of the prefix, we're still paying for those tokens. And we",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10061746258889476,
      "no_speech_prob": 1.309704731819461e-12,
      "compression_ratio": 1.696629213483146
    },
    {
      "start": 282.24,
      "end": 287.16,
      "text": "don't know what that cost is for Anthropic subscriptions or Claude plans, but we do know",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10061746258889476,
      "no_speech_prob": 1.309704731819461e-12,
      "compression_ratio": 1.696629213483146
    },
    {
      "start": 287.16,
      "end": 294.82,
      "text": "that those tokens are not free. So even if it's 0.05x of the initial cost, so 1 20th of a token,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10061746258889476,
      "no_speech_prob": 1.309704731819461e-12,
      "compression_ratio": 1.696629213483146
    },
    {
      "start": 295.24,
      "end": 301.4,
      "text": "if you have 500,000 tokens in your context window, then you're paying 25,000 tokens per turn,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10061746258889476,
      "no_speech_prob": 1.309704731819461e-12,
      "compression_ratio": 1.696629213483146
    },
    {
      "start": 301.4,
      "end": 307.16,
      "text": "and that's going to be increasing with every new turn just in cached reads. So that's one reason.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05084528804810579,
      "no_speech_prob": 1.1201571354507722e-12,
      "compression_ratio": 1.7536231884057971
    },
    {
      "start": 307.6,
      "end": 313.0,
      "text": "Now another reason that you would want to keep your context sub 20% is we know that these models",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05084528804810579,
      "no_speech_prob": 1.1201571354507722e-12,
      "compression_ratio": 1.7536231884057971
    },
    {
      "start": 313.0,
      "end": 318.72,
      "text": "start degrading in performance the more you start to fill this context up. So what that means is if",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05084528804810579,
      "no_speech_prob": 1.1201571354507722e-12,
      "compression_ratio": 1.7536231884057971
    },
    {
      "start": 318.72,
      "end": 324.6,
      "text": "you have 800,000 tokens, so 80% of your context window used, then the model is not going to be",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05084528804810579,
      "no_speech_prob": 1.1201571354507722e-12,
      "compression_ratio": 1.7536231884057971
    },
    {
      "start": 324.6,
      "end": 330.84,
      "text": "performing optimally as it would if we only had 150,000 tokens in that context window. You can",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05084528804810579,
      "no_speech_prob": 1.1201571354507722e-12,
      "compression_ratio": 1.7536231884057971
    },
    {
      "start": 330.84,
      "end": 336.28,
      "text": "see down here in my status line that I have my context usage as a bar and a percentage listed",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.060480594635009766,
      "no_speech_prob": 1.5250754218515405e-12,
      "compression_ratio": 1.7652439024390243
    },
    {
      "start": 336.28,
      "end": 341.7,
      "text": "there which means I can always see it at a glance. Now I would recommend that you do the same and I",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.060480594635009766,
      "no_speech_prob": 1.5250754218515405e-12,
      "compression_ratio": 1.7652439024390243
    },
    {
      "start": 341.7,
      "end": 346.4,
      "text": "will put a prompt on screen which you can copy and give to Cloud Code so that it can implement",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.060480594635009766,
      "no_speech_prob": 1.5250754218515405e-12,
      "compression_ratio": 1.7652439024390243
    },
    {
      "start": 346.4,
      "end": 351.96,
      "text": "the same thing for you. If you want to keep an eye on your context just run forward slash context",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.060480594635009766,
      "no_speech_prob": 1.5250754218515405e-12,
      "compression_ratio": 1.7652439024390243
    },
    {
      "start": 351.96,
      "end": 356.52,
      "text": "and it will bring up this visual where you can see what you've used. You can see it broken down here",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.060480594635009766,
      "no_speech_prob": 1.5250754218515405e-12,
      "compression_ratio": 1.7652439024390243
    },
    {
      "start": 356.52,
      "end": 360.82,
      "text": "on the right hand side as well and it breaks it down below. When you start approaching that",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.060480594635009766,
      "no_speech_prob": 1.5250754218515405e-12,
      "compression_ratio": 1.7652439024390243
    },
    {
      "start": 360.82,
      "end": 367.52,
      "text": "20 to 25% limit, I recommend that you ask Claude to generate a handoff file, which is just a markdown",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.054484535217285156,
      "no_speech_prob": 1.1971592337248205e-12,
      "compression_ratio": 1.7366548042704626
    },
    {
      "start": 367.52,
      "end": 373.02,
      "text": "file with a summary of everything you've done in that session, and also give you a session resume",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.054484535217285156,
      "no_speech_prob": 1.1971592337248205e-12,
      "compression_ratio": 1.7366548042704626
    },
    {
      "start": 373.02,
      "end": 378.88,
      "text": "prompt. So you can then give that session resume prompt back to Claude in the new session, or after",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.054484535217285156,
      "no_speech_prob": 1.1971592337248205e-12,
      "compression_ratio": 1.7366548042704626
    },
    {
      "start": 378.88,
      "end": 383.78,
      "text": "you've cleared your context, have it read the handoff file, and you can pick up exactly where",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.054484535217285156,
      "no_speech_prob": 1.1971592337248205e-12,
      "compression_ratio": 1.7366548042704626
    },
    {
      "start": 383.78,
      "end": 389.3,
      "text": "you left off. And if you want to go one step further, I built this into a skill, I just called",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.054484535217285156,
      "no_speech_prob": 1.1971592337248205e-12,
      "compression_ratio": 1.7366548042704626
    },
    {
      "start": 389.3,
      "end": 394.66,
      "text": "it handoff so all I need to do is run forward slash handoff and it will do all of that for me.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0721766655905205,
      "no_speech_prob": 2.0928953015086904e-12,
      "compression_ratio": 1.7765567765567765
    },
    {
      "start": 394.92,
      "end": 399.56,
      "text": "If you want access to the skill you will find all of the resources from the video in the pinned",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0721766655905205,
      "no_speech_prob": 2.0928953015086904e-12,
      "compression_ratio": 1.7765567765567765
    },
    {
      "start": 399.56,
      "end": 404.64,
      "text": "comment down below. Tip number three is making sure that you keep your prompts within one hour",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0721766655905205,
      "no_speech_prob": 2.0928953015086904e-12,
      "compression_ratio": 1.7765567765567765
    },
    {
      "start": 404.64,
      "end": 409.9,
      "text": "of the previous prompt. This is one of the simplest and most common mistakes that I see people making",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0721766655905205,
      "no_speech_prob": 2.0928953015086904e-12,
      "compression_ratio": 1.7765567765567765
    },
    {
      "start": 409.9,
      "end": 415.48,
      "text": "inside of Claude Code and it will chew through your five hour session limit and your weekly quota",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0721766655905205,
      "no_speech_prob": 2.0928953015086904e-12,
      "compression_ratio": 1.7765567765567765
    },
    {
      "start": 415.48,
      "end": 420.88,
      "text": "on your Claude subscription. So we're looking at the Claude Code documentation here and you can see",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05863757993354172,
      "no_speech_prob": 1.0482829881919287e-12,
      "compression_ratio": 1.8250950570342206
    },
    {
      "start": 420.88,
      "end": 426.44,
      "text": "that it says Claude Code requests the one hour TTL automatically. So we're talking about the",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05863757993354172,
      "no_speech_prob": 1.0482829881919287e-12,
      "compression_ratio": 1.8250950570342206
    },
    {
      "start": 426.44,
      "end": 432.14,
      "text": "caching again here. So when you use Claude Code, every time you send a new message away, that",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05863757993354172,
      "no_speech_prob": 1.0482829881919287e-12,
      "compression_ratio": 1.8250950570342206
    },
    {
      "start": 432.14,
      "end": 438.98,
      "text": "resets the one hour cache. Now if you go out with that one hour, then the cache is then invalidated",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05863757993354172,
      "no_speech_prob": 1.0482829881919287e-12,
      "compression_ratio": 1.8250950570342206
    },
    {
      "start": 438.98,
      "end": 443.62,
      "text": "and your next prompt, you're going to be paying for all of those tokens again. So if you have",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05863757993354172,
      "no_speech_prob": 1.0482829881919287e-12,
      "compression_ratio": 1.8250950570342206
    },
    {
      "start": 443.62,
      "end": 448.28,
      "text": "500,000 tokens sitting in your context window, which you shouldn't have if you're following step",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061567859489376806,
      "no_speech_prob": 2.0358792776620227e-12,
      "compression_ratio": 1.85546875
    },
    {
      "start": 448.28,
      "end": 453.82,
      "text": "two, then if you go away for an hour and you come back, you send your next prompt to Claude,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061567859489376806,
      "no_speech_prob": 2.0358792776620227e-12,
      "compression_ratio": 1.85546875
    },
    {
      "start": 454.16,
      "end": 460.18,
      "text": "you're paying full price for those 500,000 tokens again. If you then go away for another hour and",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061567859489376806,
      "no_speech_prob": 2.0358792776620227e-12,
      "compression_ratio": 1.85546875
    },
    {
      "start": 460.18,
      "end": 465.64,
      "text": "come back and then send your next prompt, you've now paid for those 500,000 tokens three times.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061567859489376806,
      "no_speech_prob": 2.0358792776620227e-12,
      "compression_ratio": 1.85546875
    },
    {
      "start": 466.04,
      "end": 469.98,
      "text": "So this is why it's super important that you keep it within an hour of your previous prompt",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061567859489376806,
      "no_speech_prob": 2.0358792776620227e-12,
      "compression_ratio": 1.85546875
    },
    {
      "start": 469.98,
      "end": 474.38,
      "text": "and going back to the previous tip that you keep your contacts below 20%",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.11090186747109018,
      "no_speech_prob": 8.589555390743631e-13,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 474.38,
      "end": 479.86,
      "text": "so if you do go out with that hour you're not burning 500,000 tokens every single time.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.11090186747109018,
      "no_speech_prob": 8.589555390743631e-13,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 482.02,
      "end": 487.32,
      "text": "model mid-session. How many times have you heard that you should be changing your model from Sonnet",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.046265263202761815,
      "no_speech_prob": 1.3941638642156473e-12,
      "compression_ratio": 1.6791808873720135
    },
    {
      "start": 487.32,
      "end": 493.14,
      "text": "to Opus for more complex tasks and then back to Sonnet again for the easier stuff? And I'm going",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.046265263202761815,
      "no_speech_prob": 1.3941638642156473e-12,
      "compression_ratio": 1.6791808873720135
    },
    {
      "start": 493.14,
      "end": 498.2,
      "text": "to tell you now that that is bad advice. We're staying inside of the Cloud Code docs and you can",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05388860269026323,
      "no_speech_prob": 8.423490312388371e-13,
      "compression_ratio": 1.6964285714285714
    },
    {
      "start": 498.2,
      "end": 506.02,
      "text": "see here actions that invalidate the cache. Number one on the list is switching models. So if we",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05388860269026323,
      "no_speech_prob": 8.423490312388371e-13,
      "compression_ratio": 1.6964285714285714
    },
    {
      "start": 506.02,
      "end": 511.32,
      "text": "scroll down you can see that it says each model has its own cache. Switching with forward slash",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05388860269026323,
      "no_speech_prob": 8.423490312388371e-13,
      "compression_ratio": 1.6964285714285714
    },
    {
      "start": 511.32,
      "end": 518.04,
      "text": "model means the next request reads the entire conversation history with no cache hits even",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05388860269026323,
      "no_speech_prob": 8.423490312388371e-13,
      "compression_ratio": 1.6964285714285714
    },
    {
      "start": 518.04,
      "end": 523.54,
      "text": "know the content is identical. So I see people making this mistake all the time. They're constantly",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03835346318092667,
      "no_speech_prob": 1.501614913142213e-12,
      "compression_ratio": 1.8270676691729324
    },
    {
      "start": 523.54,
      "end": 529.82,
      "text": "switching models between Opus, Sonnet, Sonnet, Opus, maybe Fable, and you're paying for all those",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03835346318092667,
      "no_speech_prob": 1.501614913142213e-12,
      "compression_ratio": 1.8270676691729324
    },
    {
      "start": 529.82,
      "end": 533.82,
      "text": "tokens again every single time you make that switch. I'm not saying that you should not be",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03835346318092667,
      "no_speech_prob": 1.501614913142213e-12,
      "compression_ratio": 1.8270676691729324
    },
    {
      "start": 533.82,
      "end": 538.46,
      "text": "switching models at all because that is genuinely good advice to use the right model based on the",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03835346318092667,
      "no_speech_prob": 1.501614913142213e-12,
      "compression_ratio": 1.8270676691729324
    },
    {
      "start": 538.46,
      "end": 544.7,
      "text": "task at hand. What I am saying is that you should not be switching model mid-session. You should be",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03835346318092667,
      "no_speech_prob": 1.501614913142213e-12,
      "compression_ratio": 1.8270676691729324
    },
    {
      "start": 544.7,
      "end": 548.98,
      "text": "putting together a plan using your higher power model, whether that's Opus or whether it's Fable,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0823155149407343,
      "no_speech_prob": 9.582434672139595e-13,
      "compression_ratio": 1.75
    },
    {
      "start": 549.44,
      "end": 553.84,
      "text": "and then once you have a structured plan in place, switch your model down to whatever it is that",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0823155149407343,
      "no_speech_prob": 9.582434672139595e-13,
      "compression_ratio": 1.75
    },
    {
      "start": 553.84,
      "end": 558.14,
      "text": "you're going to be using to execute that task, and do not switch again unless you absolutely",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0823155149407343,
      "no_speech_prob": 9.582434672139595e-13,
      "compression_ratio": 1.75
    },
    {
      "start": 558.14,
      "end": 563.64,
      "text": "have to. Number five is almost identical, and that is avoiding changing effort level",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0823155149407343,
      "no_speech_prob": 9.582434672139595e-13,
      "compression_ratio": 1.75
    },
    {
      "start": 563.64,
      "end": 569.9,
      "text": "mid-session. Right below switching models and actions that invalidate the cash, we have changing",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0823155149407343,
      "no_speech_prob": 9.582434672139595e-13,
      "compression_ratio": 1.75
    },
    {
      "start": 569.9,
      "end": 575.44,
      "text": "effort level. So you can see here that it says that the cache is keyed by effort level as well",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03878362329156549,
      "no_speech_prob": 1.6362917965029311e-12,
      "compression_ratio": 1.7627737226277371
    },
    {
      "start": 575.44,
      "end": 581.12,
      "text": "as model so switching with forward slash effort means that the next request reads the entire",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03878362329156549,
      "no_speech_prob": 1.6362917965029311e-12,
      "compression_ratio": 1.7627737226277371
    },
    {
      "start": 581.12,
      "end": 587.64,
      "text": "history again with no cache hits. So this is the exact same as the previous tip if you can avoid",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03878362329156549,
      "no_speech_prob": 1.6362917965029311e-12,
      "compression_ratio": 1.7627737226277371
    },
    {
      "start": 587.64,
      "end": 593.08,
      "text": "changing effort level mid-session then you should absolutely be doing that. Number six is building a",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03878362329156549,
      "no_speech_prob": 1.6362917965029311e-12,
      "compression_ratio": 1.7627737226277371
    },
    {
      "start": 593.08,
      "end": 598.04,
      "text": "proper memory system. I have a full video on my channel breaking down how you can set this system",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.03878362329156549,
      "no_speech_prob": 1.6362917965029311e-12,
      "compression_ratio": 1.7627737226277371
    },
    {
      "start": 598.04,
      "end": 603.68,
      "text": "up for yourself. I'm not going to go too deep here but I will link that video on screen. But what",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05218262401053576,
      "no_speech_prob": 1.427225850524072e-12,
      "compression_ratio": 1.6794425087108014
    },
    {
      "start": 603.68,
      "end": 608.88,
      "text": "we're talking about doing here is building a proper memory system so that Claude Code has a",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05218262401053576,
      "no_speech_prob": 1.427225850524072e-12,
      "compression_ratio": 1.6794425087108014
    },
    {
      "start": 608.88,
      "end": 613.96,
      "text": "working map of our project and it always knows where to go to find the information that we're",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05218262401053576,
      "no_speech_prob": 1.427225850524072e-12,
      "compression_ratio": 1.6794425087108014
    },
    {
      "start": 613.96,
      "end": 618.82,
      "text": "asking for. And we can do that by using something like Obsidian. And if you don't know what Obsidian",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05218262401053576,
      "no_speech_prob": 1.427225850524072e-12,
      "compression_ratio": 1.6794425087108014
    },
    {
      "start": 618.82,
      "end": 624.4,
      "text": "is, it's just a software that allows us to view and edit marked-in files. And you can see here an",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05218262401053576,
      "no_speech_prob": 1.427225850524072e-12,
      "compression_ratio": 1.6794425087108014
    },
    {
      "start": 624.4,
      "end": 629.28,
      "text": "example of my obsidian knowledge graph you get this really cool graph we can kind of like pull",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.046571267169454825,
      "no_speech_prob": 9.657287990127994e-13,
      "compression_ratio": 1.99288256227758
    },
    {
      "start": 629.28,
      "end": 633.32,
      "text": "around all these different nodes all these different lines that you see are connections",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.046571267169454825,
      "no_speech_prob": 9.657287990127994e-13,
      "compression_ratio": 1.99288256227758
    },
    {
      "start": 633.32,
      "end": 638.34,
      "text": "between different documents so it can piece together different bits of information work",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.046571267169454825,
      "no_speech_prob": 9.657287990127994e-13,
      "compression_ratio": 1.99288256227758
    },
    {
      "start": 638.34,
      "end": 642.48,
      "text": "between different documents to find the right information to give you a high level overview of",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.046571267169454825,
      "no_speech_prob": 9.657287990127994e-13,
      "compression_ratio": 1.99288256227758
    },
    {
      "start": 642.48,
      "end": 648.06,
      "text": "how this works we create a folder somewhere on our computer and we then point obsidian at this",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.046571267169454825,
      "no_speech_prob": 9.657287990127994e-13,
      "compression_ratio": 1.99288256227758
    },
    {
      "start": 648.06,
      "end": 654.22,
      "text": "folder and then that then becomes our vault now once we have our vault we can then point cloud code",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.046571267169454825,
      "no_speech_prob": 9.657287990127994e-13,
      "compression_ratio": 1.99288256227758
    },
    {
      "start": 654.22,
      "end": 660.42,
      "text": "at this vault and inside of our vault we will have a bunch of different folders in there. Now these",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.060979227224985756,
      "no_speech_prob": 1.2350504733499301e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 660.42,
      "end": 666.14,
      "text": "folders can be whatever you like. For me I like to split this into three separate layers. So the",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.060979227224985756,
      "no_speech_prob": 1.2350504733499301e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 666.14,
      "end": 672.88,
      "text": "first layer is my working folders. So for me that's things like agency, school, content and inside of",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.060979227224985756,
      "no_speech_prob": 1.2350504733499301e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 672.88,
      "end": 678.26,
      "text": "these folders I will have subfolders. So for agency that will be things like clients, it could be",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.060979227224985756,
      "no_speech_prob": 1.2350504733499301e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 678.26,
      "end": 684.98,
      "text": "deliverables, outreach, pipeline, for school, it might be frameworks, strategy, courses. So that is",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08916924285888672,
      "no_speech_prob": 1.3996227137338946e-12,
      "compression_ratio": 1.6305084745762712
    },
    {
      "start": 684.98,
      "end": 690.1,
      "text": "our first layer, all of the working files that you'll be in there messing around with on a day",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08916924285888672,
      "no_speech_prob": 1.3996227137338946e-12,
      "compression_ratio": 1.6305084745762712
    },
    {
      "start": 690.1,
      "end": 695.64,
      "text": "to day basis. The second layer here originates from Andre Carpathie's Obsidian RAG system. I'm",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08916924285888672,
      "no_speech_prob": 1.3996227137338946e-12,
      "compression_ratio": 1.6305084745762712
    },
    {
      "start": 695.64,
      "end": 699.84,
      "text": "not going to go into that here so make sure you check out that video linked on screen if you're",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08916924285888672,
      "no_speech_prob": 1.3996227137338946e-12,
      "compression_ratio": 1.6305084745762712
    },
    {
      "start": 699.84,
      "end": 705.58,
      "text": "interested in that. And the third layer is our clod.md. So in there as we covered in tip number",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08916924285888672,
      "no_speech_prob": 1.3996227137338946e-12,
      "compression_ratio": 1.6305084745762712
    },
    {
      "start": 705.58,
      "end": 711.5,
      "text": "one. This is where we store our project map, our rules, our context, so that Claude knows everything",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05896076965332031,
      "no_speech_prob": 1.5732351402522604e-12,
      "compression_ratio": 1.8125
    },
    {
      "start": 711.5,
      "end": 716.2,
      "text": "which it needs to know about us at session start without us having to explain every time. So how",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05896076965332031,
      "no_speech_prob": 1.5732351402522604e-12,
      "compression_ratio": 1.8125
    },
    {
      "start": 716.2,
      "end": 722.0,
      "text": "that relates to our memory system here is, first of all, we give Claude a map of the project folders,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05896076965332031,
      "no_speech_prob": 1.5732351402522604e-12,
      "compression_ratio": 1.8125
    },
    {
      "start": 722.44,
      "end": 728.08,
      "text": "what's contained in each folder, so that it knows where things are. Then we need to give it a map",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05896076965332031,
      "no_speech_prob": 1.5732351402522604e-12,
      "compression_ratio": 1.8125
    },
    {
      "start": 728.08,
      "end": 733.0,
      "text": "of where it needs to put things when we ask for them to be saved. So you can see here, anything",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.05896076965332031,
      "no_speech_prob": 1.5732351402522604e-12,
      "compression_ratio": 1.8125
    },
    {
      "start": 733.0,
      "end": 738.94,
      "text": "for a client goes in agency clients and then the client's name. If it's an idea or a script or a",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.050039232822886685,
      "no_speech_prob": 1.7691013584619775e-12,
      "compression_ratio": 1.7718631178707225
    },
    {
      "start": 738.94,
      "end": 743.12,
      "text": "hook it goes in content. So we give it the information which it needs so that things don't",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.050039232822886685,
      "no_speech_prob": 1.7691013584619775e-12,
      "compression_ratio": 1.7718631178707225
    },
    {
      "start": 743.12,
      "end": 749.28,
      "text": "get lost when it saves them. And finally we give it the knowledge base rules. So this is how it",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.050039232822886685,
      "no_speech_prob": 1.7691013584619775e-12,
      "compression_ratio": 1.7718631178707225
    },
    {
      "start": 749.28,
      "end": 754.18,
      "text": "knows how to operate our knowledge base system and this is arguably the most important part.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.050039232822886685,
      "no_speech_prob": 1.7691013584619775e-12,
      "compression_ratio": 1.7718631178707225
    },
    {
      "start": 754.54,
      "end": 758.42,
      "text": "You can see here we have information on wiki links to make sure that it's always creating",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.050039232822886685,
      "no_speech_prob": 1.7691013584619775e-12,
      "compression_ratio": 1.7718631178707225
    },
    {
      "start": 758.42,
      "end": 763.94,
      "text": "those links between different documents and that is how we get our nice graph here inside of",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07911000468514183,
      "no_speech_prob": 1.3670687845634966e-12,
      "compression_ratio": 1.578512396694215
    },
    {
      "start": 763.94,
      "end": 770.14,
      "text": "obsidian.md. Number seven is creating custom skills. This one is hands down the most powerful",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07911000468514183,
      "no_speech_prob": 1.3670687845634966e-12,
      "compression_ratio": 1.578512396694215
    },
    {
      "start": 770.14,
      "end": 775.84,
      "text": "tip in this video, not only from a token saving perspective, but also in maximizing the value",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07911000468514183,
      "no_speech_prob": 1.3670687845634966e-12,
      "compression_ratio": 1.578512396694215
    },
    {
      "start": 775.84,
      "end": 782.3,
      "text": "which you get out of cloud code for you specifically. So creating a custom skill is just packaging up",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07911000468514183,
      "no_speech_prob": 1.3670687845634966e-12,
      "compression_ratio": 1.578512396694215
    },
    {
      "start": 782.3,
      "end": 788.86,
      "text": "some amount of rich information on a specific topic that's specific to you so that it's repeatable",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.058224875351478314,
      "no_speech_prob": 1.3618749140562048e-12,
      "compression_ratio": 1.6576271186440679
    },
    {
      "start": 788.86,
      "end": 794.06,
      "text": "and Claude can do it over and over again without us having to go back 20 different times, change",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.058224875351478314,
      "no_speech_prob": 1.3618749140562048e-12,
      "compression_ratio": 1.6576271186440679
    },
    {
      "start": 794.06,
      "end": 799.9,
      "text": "this, change that, because Claude already knows what a good output looks like. Now I would highly",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.058224875351478314,
      "no_speech_prob": 1.3618749140562048e-12,
      "compression_ratio": 1.6576271186440679
    },
    {
      "start": 799.9,
      "end": 806.08,
      "text": "recommend that you install the skill creator skill by Anthropics. This is an official Anthropics skill",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.058224875351478314,
      "no_speech_prob": 1.3618749140562048e-12,
      "compression_ratio": 1.6576271186440679
    },
    {
      "start": 806.08,
      "end": 811.24,
      "text": "and it tells Claude Code how to build custom skills. So first of all you're going to explain",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.058224875351478314,
      "no_speech_prob": 1.3618749140562048e-12,
      "compression_ratio": 1.6576271186440679
    },
    {
      "start": 811.24,
      "end": 815.32,
      "text": "what it is that you're trying to do. Then Cloud Code is going to ask you a bunch of questions,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07767491658528645,
      "no_speech_prob": 1.804004103218848e-12,
      "compression_ratio": 1.9757785467128028
    },
    {
      "start": 815.72,
      "end": 819.16,
      "text": "it's going to test it, it's going to run evaluations, it's going to make sure that it",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07767491658528645,
      "no_speech_prob": 1.804004103218848e-12,
      "compression_ratio": 1.9757785467128028
    },
    {
      "start": 819.16,
      "end": 823.3,
      "text": "triggers when it should, and it's going to make sure that you're happy with the output. So then",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07767491658528645,
      "no_speech_prob": 1.804004103218848e-12,
      "compression_ratio": 1.9757785467128028
    },
    {
      "start": 823.3,
      "end": 828.42,
      "text": "all you need to do when you want to use it going forward is run a slash command and then that task",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07767491658528645,
      "no_speech_prob": 1.804004103218848e-12,
      "compression_ratio": 1.9757785467128028
    },
    {
      "start": 828.42,
      "end": 834.48,
      "text": "is now repeatable on the mat. So the reason that this is so powerful is without a skill, you ask",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07767491658528645,
      "no_speech_prob": 1.804004103218848e-12,
      "compression_ratio": 1.9757785467128028
    },
    {
      "start": 834.48,
      "end": 839.02,
      "text": "it to do something, you then have to go back saying no that's not right, you might need to go back",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07767491658528645,
      "no_speech_prob": 1.804004103218848e-12,
      "compression_ratio": 1.9757785467128028
    },
    {
      "start": 839.02,
      "end": 844.28,
      "text": "again and again and again and eventually you will have a product that you're happy with but by that",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.04283719270125679,
      "no_speech_prob": 1.8041583851879928e-12,
      "compression_ratio": 1.712280701754386
    },
    {
      "start": 844.28,
      "end": 849.1,
      "text": "point you've burned a bunch of tokens and if you're doing this multiple times a week then I can",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.04283719270125679,
      "no_speech_prob": 1.8041583851879928e-12,
      "compression_ratio": 1.712280701754386
    },
    {
      "start": 849.1,
      "end": 854.84,
      "text": "guarantee you're wasting a huge amount of your plan limits on tasks that are repeatable. So what I",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.04283719270125679,
      "no_speech_prob": 1.8041583851879928e-12,
      "compression_ratio": 1.712280701754386
    },
    {
      "start": 854.84,
      "end": 859.9,
      "text": "would suggest is that you copy the prompt which you see on screen and you have Claude go through",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.04283719270125679,
      "no_speech_prob": 1.8041583851879928e-12,
      "compression_ratio": 1.712280701754386
    },
    {
      "start": 859.9,
      "end": 865.66,
      "text": "your last 10, 15, 20 sessions and pick out things which you've asked it to do multiple different",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.04283719270125679,
      "no_speech_prob": 1.8041583851879928e-12,
      "compression_ratio": 1.712280701754386
    },
    {
      "start": 865.66,
      "end": 871.06,
      "text": "times and tell it to build those into skills. To install the skill creator skill just run",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08089208602905273,
      "no_speech_prob": 9.582062790794432e-13,
      "compression_ratio": 1.7078651685393258
    },
    {
      "start": 871.06,
      "end": 877.3,
      "text": "forward slash plugins and then on the discover tab here you will see it listed fourth from the top.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08089208602905273,
      "no_speech_prob": 9.582062790794432e-13,
      "compression_ratio": 1.7078651685393258
    },
    {
      "start": 877.44,
      "end": 883.2,
      "text": "You can see skill creator create new skills, improve existing skills and measure skills as",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08089208602905273,
      "no_speech_prob": 9.582062790794432e-13,
      "compression_ratio": 1.7078651685393258
    },
    {
      "start": 883.2,
      "end": 888.54,
      "text": "well. So all you need to do is enter this and install it. I would recommend that you go for",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08089208602905273,
      "no_speech_prob": 9.582062790794432e-13,
      "compression_ratio": 1.7078651685393258
    },
    {
      "start": 888.54,
      "end": 892.54,
      "text": "user scope which means you'll have access to across all of your different projects.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.08089208602905273,
      "no_speech_prob": 9.582062790794432e-13,
      "compression_ratio": 1.7078651685393258
    },
    {
      "start": 892.54,
      "end": 894.64,
      "text": "Number eight is using Caveman.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 894.98,
      "end": 898.32,
      "text": "Caveman is a plugin that makes Cloud Code less verbose.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 898.78,
      "end": 901.7,
      "text": "So we know that these large language models love to talk.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 902.04,
      "end": 904.88,
      "text": "You ask it a simple question and it gives you a huge reply,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 905.2,
      "end": 906.82,
      "text": "which will take you five minutes to read.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 907.44,
      "end": 910.6,
      "text": "And Caveman was designed to try and stop that from happening.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 910.84,
      "end": 912.6,
      "text": "And it actually does a very, very good job.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 912.98,
      "end": 916.02,
      "text": "I've been using it on and off now for around about three or four months.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 916.28,
      "end": 917.7,
      "text": "And it is very effective,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 917.88,
      "end": 921.74,
      "text": "especially if you're not wanting a big walls of text back from the model.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.13866254261561803,
      "no_speech_prob": 1.920496530902205e-12,
      "compression_ratio": 1.64576802507837
    },
    {
      "start": 922.32,
      "end": 926.24,
      "text": "So it says here, why use many token when few do trick?",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.1385005596464714,
      "no_speech_prob": 1.5311772032580717e-12,
      "compression_ratio": 1.5387596899224807
    },
    {
      "start": 926.58,
      "end": 927.7,
      "text": "And that is kind of how it works.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.1385005596464714,
      "no_speech_prob": 1.5311772032580717e-12,
      "compression_ratio": 1.5387596899224807
    },
    {
      "start": 928.12,
      "end": 930.52,
      "text": "Make your AI coding agent talk like a caveman.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.1385005596464714,
      "no_speech_prob": 1.5311772032580717e-12,
      "compression_ratio": 1.5387596899224807
    },
    {
      "start": 930.52,
      "end": 934.42,
      "text": "Same answers, up to 65% fewer output tokens.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.1385005596464714,
      "no_speech_prob": 1.5311772032580717e-12,
      "compression_ratio": 1.5387596899224807
    },
    {
      "start": 934.52,
      "end": 939.54,
      "text": "Now for me personally, I probably haven't seen as much as 65% fewer output tokens,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.1385005596464714,
      "no_speech_prob": 1.5311772032580717e-12,
      "compression_ratio": 1.5387596899224807
    },
    {
      "start": 939.7,
      "end": 942.8,
      "text": "but it does definitely reduce the token output.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.1385005596464714,
      "no_speech_prob": 1.5311772032580717e-12,
      "compression_ratio": 1.5387596899224807
    },
    {
      "start": 943.02,
      "end": 944.1,
      "text": "You can see an example here.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.1385005596464714,
      "no_speech_prob": 1.5311772032580717e-12,
      "compression_ratio": 1.5387596899224807
    },
    {
      "start": 944.22,
      "end": 948.2,
      "text": "I said, explain how prompt caching reduces cost in LLMs.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.1385005596464714,
      "no_speech_prob": 1.5311772032580717e-12,
      "compression_ratio": 1.5387596899224807
    },
    {
      "start": 948.2,
      "end": 951.9,
      "text": "and this was the response that I got using Caveman.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.2301800467751243,
      "no_speech_prob": 7.403765100999493e-13,
      "compression_ratio": 1.375
    },
    {
      "start": 952.36,
      "end": 954.26,
      "text": "Now if we compare that to this other terminal,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.2301800467751243,
      "no_speech_prob": 7.403765100999493e-13,
      "compression_ratio": 1.375
    },
    {
      "start": 954.74,
      "end": 956.14,
      "text": "I gave it the exact same prompt,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.2301800467751243,
      "no_speech_prob": 7.403765100999493e-13,
      "compression_ratio": 1.375
    },
    {
      "start": 956.26,
      "end": 959.32,
      "text": "explain how prompt caching reduces cost in LLMs,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.2301800467751243,
      "no_speech_prob": 7.403765100999493e-13,
      "compression_ratio": 1.375
    },
    {
      "start": 959.32,
      "end": 960.0,
      "text": "and...",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.2301800467751243,
      "no_speech_prob": 7.403765100999493e-13,
      "compression_ratio": 1.375
    },
    {
      "start": 960.0,
      "end": 964.58,
      "text": "reduces cost in LLMs and the response which I got from this one was much much longer. You can see",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.059414579848612634,
      "no_speech_prob": 2.645283731644854e-12,
      "compression_ratio": 1.8401486988847584
    },
    {
      "start": 964.58,
      "end": 969.34,
      "text": "the difference here in the size of the response. So this is definitely one that you should be trying",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.059414579848612634,
      "no_speech_prob": 2.645283731644854e-12,
      "compression_ratio": 1.8401486988847584
    },
    {
      "start": 969.34,
      "end": 974.7,
      "text": "out. Number nine is using ponytail. This one has been blown up over the past few weeks. You can see",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.059414579848612634,
      "no_speech_prob": 2.645283731644854e-12,
      "compression_ratio": 1.8401486988847584
    },
    {
      "start": 974.7,
      "end": 980.06,
      "text": "that it's number one repository of the day and of the week and it was designed because these AI",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.10072683446547565,
      "no_speech_prob": 1.1114378731594265e-12,
      "compression_ratio": 1.6422413793103448
    },
    {
      "start": 980.06,
      "end": 986.2,
      "text": "coding agents love writing unnecessary amounts of code and ponytail is essentially designed to make",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.10072683446547565,
      "no_speech_prob": 1.1114378731594265e-12,
      "compression_ratio": 1.6422413793103448
    },
    {
      "start": 986.2,
      "end": 991.56,
      "text": "your AI coding agent as lazy as possible without making it complacent. So if we look at the",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.10072683446547565,
      "no_speech_prob": 1.1114378731594265e-12,
      "compression_ratio": 1.6422413793103448
    },
    {
      "start": 991.56,
      "end": 997.06,
      "text": "benchmarks here you can see the impact which this has. So this is versus a no skill baseline,",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.10072683446547565,
      "no_speech_prob": 1.1114378731594265e-12,
      "compression_ratio": 1.6422413793103448
    },
    {
      "start": 997.06,
      "end": 1008.08,
      "text": "minus 54% on lines of code, minus 22% on tokens, minus 20% on cost, and minus 27% on time. And if",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.03391525480482313,
      "no_speech_prob": 1.102785831402775e-12,
      "compression_ratio": 1.6041666666666667
    },
    {
      "start": 1008.08,
      "end": 1013.6,
      "text": "we keep scrolling down here, you can see this hierarchy of how it works. So before writing any",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.03391525480482313,
      "no_speech_prob": 1.102785831402775e-12,
      "compression_ratio": 1.6041666666666667
    },
    {
      "start": 1013.6,
      "end": 1018.34,
      "text": "code, the agent will stop at the first rung that holds. The first thing which it's going to do is",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.03391525480482313,
      "no_speech_prob": 1.102785831402775e-12,
      "compression_ratio": 1.6041666666666667
    },
    {
      "start": 1018.34,
      "end": 1023.48,
      "text": "check, does this even need to exist? Only if that is a yes, is it then going to see if there's",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.03391525480482313,
      "no_speech_prob": 1.102785831402775e-12,
      "compression_ratio": 1.6041666666666667
    },
    {
      "start": 1023.48,
      "end": 1028.46,
      "text": "something already in the code base that can handle it. And if that is a no, it's going to see,",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07001619954263011,
      "no_speech_prob": 2.2272320706479e-12,
      "compression_ratio": 1.6992753623188406
    },
    {
      "start": 1028.9,
      "end": 1033.96,
      "text": "can I handle this with one single line of code? And only if that is a no, is it then going to",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07001619954263011,
      "no_speech_prob": 2.2272320706479e-12,
      "compression_ratio": 1.6992753623188406
    },
    {
      "start": 1033.96,
      "end": 1039.0,
      "text": "write the minimum amount of code to get the job done. Number 10 is using the Codex plugin for",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07001619954263011,
      "no_speech_prob": 2.2272320706479e-12,
      "compression_ratio": 1.6992753623188406
    },
    {
      "start": 1039.0,
      "end": 1043.98,
      "text": "CloudCode. This is an amazing plugin if you're someone who's already paying for an OpenAI",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07001619954263011,
      "no_speech_prob": 2.2272320706479e-12,
      "compression_ratio": 1.6992753623188406
    },
    {
      "start": 1043.98,
      "end": 1051.02,
      "text": "subscription. So it allows us to use Codex inside of the CloudCode ecosystem, which means we can",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07001619954263011,
      "no_speech_prob": 2.2272320706479e-12,
      "compression_ratio": 1.6992753623188406
    },
    {
      "start": 1051.02,
      "end": 1058.0,
      "text": "offload tasks from the Anthropic models onto the GPT models. Now we know that these GPT models are",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06184981487415455,
      "no_speech_prob": 1.0857881434234162e-12,
      "compression_ratio": 1.6750902527075813
    },
    {
      "start": 1058.0,
      "end": 1063.16,
      "text": "substantially cheaper than the Anthropic models and they're now very very competitive as well.",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06184981487415455,
      "no_speech_prob": 1.0857881434234162e-12,
      "compression_ratio": 1.6750902527075813
    },
    {
      "start": 1063.7,
      "end": 1068.38,
      "text": "GPT 5.6 has just been released so that is definitely something that you can take advantage of",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06184981487415455,
      "no_speech_prob": 1.0857881434234162e-12,
      "compression_ratio": 1.6750902527075813
    },
    {
      "start": 1068.38,
      "end": 1073.08,
      "text": "if you're already paying for an OpenAI plan. Once you have the plugin installed just do",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06184981487415455,
      "no_speech_prob": 1.0857881434234162e-12,
      "compression_ratio": 1.6750902527075813
    },
    {
      "start": 1073.08,
      "end": 1077.12,
      "text": "forward slash codex and you will see all the different slash commands which you now have",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06184981487415455,
      "no_speech_prob": 1.0857881434234162e-12,
      "compression_ratio": 1.6750902527075813
    },
    {
      "start": 1077.12,
      "end": 1082.44,
      "text": "available. One which you should definitely check out is Codex Adversarial Review, which is going to",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05638682255979444,
      "no_speech_prob": 1.4897402972682827e-12,
      "compression_ratio": 1.6666666666666667
    },
    {
      "start": 1082.44,
      "end": 1087.18,
      "text": "have the Codex agent do a deep dive into your code base to see if it can find any security",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05638682255979444,
      "no_speech_prob": 1.4897402972682827e-12,
      "compression_ratio": 1.6666666666666667
    },
    {
      "start": 1087.18,
      "end": 1091.82,
      "text": "vulnerabilities. This is definitely one that you should check out. Go in there, have a play around",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05638682255979444,
      "no_speech_prob": 1.4897402972682827e-12,
      "compression_ratio": 1.6666666666666667
    },
    {
      "start": 1091.82,
      "end": 1096.86,
      "text": "with it, and this will save you a bunch of tokens. And finally, tip number 11 is offloading research",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05638682255979444,
      "no_speech_prob": 1.4897402972682827e-12,
      "compression_ratio": 1.6666666666666667
    },
    {
      "start": 1096.86,
      "end": 1104.22,
      "text": "heavy tasks to Google servers using Notebook LMPi. Notebook LMPi allows us to use Google servers to",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05638682255979444,
      "no_speech_prob": 1.4897402972682827e-12,
      "compression_ratio": 1.6666666666666667
    },
    {
      "start": 1104.22,
      "end": 1109.74,
      "text": "carry out any kind of deep web research. We can also use it to generate infographics and slide",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05597217459427683,
      "no_speech_prob": 1.0120705272106956e-12,
      "compression_ratio": 1.7050359712230216
    },
    {
      "start": 1109.74,
      "end": 1114.12,
      "text": "decks and a bunch of other cool stuff as well and we can do this all from the command line",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05597217459427683,
      "no_speech_prob": 1.0120705272106956e-12,
      "compression_ratio": 1.7050359712230216
    },
    {
      "start": 1114.12,
      "end": 1120.5,
      "text": "inside of Cloud Code. Now one of the main benefits of this is that we know that research flows,",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05597217459427683,
      "no_speech_prob": 1.0120705272106956e-12,
      "compression_ratio": 1.7050359712230216
    },
    {
      "start": 1120.92,
      "end": 1125.84,
      "text": "sub-agents, agent teams, all these things are super token heavy and we don't really want to",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05597217459427683,
      "no_speech_prob": 1.0120705272106956e-12,
      "compression_ratio": 1.7050359712230216
    },
    {
      "start": 1125.84,
      "end": 1131.68,
      "text": "be chewing through our Cloud subscription on those tasks and that is when we can offload those tasks",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.05597217459427683,
      "no_speech_prob": 1.0120705272106956e-12,
      "compression_ratio": 1.7050359712230216
    },
    {
      "start": 1131.68,
      "end": 1137.56,
      "text": "to Google servers essentially for free using Notebook LMPi and one of the massive benefits",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.08408194630085922,
      "no_speech_prob": 9.001602631689143e-13,
      "compression_ratio": 1.6202090592334495
    },
    {
      "start": 1137.56,
      "end": 1142.46,
      "text": "to this is because it's Gemini under the hood it has access to all of the YouTube content",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.08408194630085922,
      "no_speech_prob": 9.001602631689143e-13,
      "compression_ratio": 1.6202090592334495
    },
    {
      "start": 1142.46,
      "end": 1147.1,
      "text": "transcripts and all of that rich information which you normally would not be able to surface",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.08408194630085922,
      "no_speech_prob": 9.001602631689143e-13,
      "compression_ratio": 1.6202090592334495
    },
    {
      "start": 1147.1,
      "end": 1153.34,
      "text": "just using Claude Code alone. So I asked Claude Code to use the Notebook LMPi CLI to carry out",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.08408194630085922,
      "no_speech_prob": 9.001602631689143e-13,
      "compression_ratio": 1.6202090592334495
    },
    {
      "start": 1153.34,
      "end": 1160.56,
      "text": "research on GPT 5.6 SOL versus Claude Fable 5 and then turn that into an infographic. So it went",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.08408194630085922,
      "no_speech_prob": 9.001602631689143e-13,
      "compression_ratio": 1.6202090592334495
    },
    {
      "start": 1160.56,
      "end": 1165.24,
      "text": "through the process here and this is what we got at the end. Now bear in mind I did not need to do",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04898752388379252,
      "no_speech_prob": 1.6363169499933328e-12,
      "compression_ratio": 1.8164556962025316
    },
    {
      "start": 1165.24,
      "end": 1171.04,
      "text": "anything for this. I did not use any of my clod tokens or my plan limits and this is what we got",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04898752388379252,
      "no_speech_prob": 1.6363169499933328e-12,
      "compression_ratio": 1.8164556962025316
    },
    {
      "start": 1171.04,
      "end": 1176.0,
      "text": "at the end. So this is an awesome tool and you should definitely be leveraging this to save",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04898752388379252,
      "no_speech_prob": 1.6363169499933328e-12,
      "compression_ratio": 1.8164556962025316
    },
    {
      "start": 1176.0,
      "end": 1180.58,
      "text": "yourself a bunch of tokens. That is everything for this video today. If you got some kind of",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04898752388379252,
      "no_speech_prob": 1.6363169499933328e-12,
      "compression_ratio": 1.8164556962025316
    },
    {
      "start": 1180.58,
      "end": 1185.42,
      "text": "value out of this then please hit subscribe. I would massively appreciate it and make sure that",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04898752388379252,
      "no_speech_prob": 1.6363169499933328e-12,
      "compression_ratio": 1.8164556962025316
    },
    {
      "start": 1185.42,
      "end": 1190.4,
      "text": "you check out the pinned comment for all of the free resources. You'll also find a full clod code",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04898752388379252,
      "no_speech_prob": 1.6363169499933328e-12,
      "compression_ratio": 1.8164556962025316
    },
    {
      "start": 1190.4,
      "end": 1194.96,
      "text": "masterclass in the school which will take you from complete beginner to proficient",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.13806490216936385,
      "no_speech_prob": 9.76981082839723e-13,
      "compression_ratio": 1.3392857142857142
    },
    {
      "start": 1194.96,
      "end": 1198.96,
      "text": "cloud code user so thanks for watching i'll see you in the next one",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.13806490216936385,
      "no_speech_prob": 9.76981082839723e-13,
      "compression_ratio": 1.3392857142857142
    }
  ],
  "segmented_transcription": {
    "status": "SEGMENTED_TRANSCRIPT_OK",
    "created_at": "2026-07-28T15:41:44",
    "manifest": "/Users/sagawa/AI_WORK/video_notes/url/20260728_154100__Never_Hit_Claude_Code_Usage_Limits_Again/segmented_work/chunk_manifest.json",
    "audio": "/Users/sagawa/AI_WORK/video_notes/url/20260728_154100__Never_Hit_Claude_Code_Usage_Limits_Again/audio_16k_mono.wav",
    "audio_sha256": "cb53c488581677d649c3af00afd01a740426b50e896776ca6c9f7bd8f63c9e5c",
    "audio_duration": 1199.194563,
    "model": "mlx-community/whisper-large-v3-turbo",
    "language_hint": "en",
    "condition_on_previous_text": false,
    "chunk_count": 3,
    "resumed_chunks": 0,
    "merged_segment_count": 228,
    "final_segment_end": 1198.96,
    "tail_repeat_run": 1,
    "tail_repeat_text": "CloudCode. This is an amazing plugin if you're someone who's already paying for an OpenAI",
    "qc_ok": true,
    "qc_errors": []
  }
}