{
  "text": "In the past week or so, so many people have been complaining about hitting their cloud code limit\ninsanely fast. Claims like one prompt that is about 1% of the limit is now around 10%.\nYou could go through X and find tons and tons of threads about this topic.\nEven on a $200 per month plan, people are reaching the session limit way too fast.\nAnd then we got this post from an Anthropic employee that basically said that they\nare working on a little change with peak hours and off-peak hours. But even after that,\nsome people were saying they were still hitting it really quick, even during off-peak hours.\nSo anyways, I've been playing around a ton, trying different things, doing research, and I have 18 token management hacks for you guys that I've organized from tier 1 all the way up to tier 3 so they get more advanced as we go.\nI'm very confident that by the end of this video, you will feel like your Claude code usage has doubled, tripled, maybe even 5x'd.\nSo let's not waste any time and just get straight into the video.\nSo as I've been optimizing my own token management, I think that what's really important to realize first is how tokens actually work.\nBecause once you realize how Claude uses tokens, it makes it very clear how you should actually\nreverse engineer the way that you work in order to use less tokens.\nSo a token is the smallest unit of text that an AI model reads and charges you for.\nIt's roughly one token is one word, but that's not explicitly true.\nKind of just a good baseline.\nSo every time that you send a message, Claude rereads the entire conversation from the beginning,\nand all of those are tokens that it's charging you for.\nSo message one, it will read it, then it will read its reply, and then message two,\nand then the reply all the way up to your latest prompt.\nAnd it does that every single time.\nAnd I think that alone is a huge light bulb moment\nfor a lot of people.\nThis means as you're having a conversation with Claude,\nyour cost is compounding, not just adding,\nit's exponentially growing.\nMeaning message one might cost 500 tokens,\nmessage 30 costs 15,000\nbecause it's rereading everything before it.\nOne developer actually tracked a 100 plus message chat\nand found that 98.5% of all the tokens\nwere just spent rereading the old chat history in the session.\nLike that's a huge waste.\nNow, yes, the argument has to be made that,\nwell, it needs the context and it needs to understand what we're doing,\nbut still 98.5% is crazy.\nSo take a quick look at this graphic here.\nAlong the x-axis, we have message number.\nAnd as it increases, you can see that we have our per message cost\nand our cumulative tokens increasing, but it's not linear.\nIt's basically each message is rereading all of the past ones\nand it has to count that in.\nSo message one could be 500,\nmessage 30 could be 15,500, which is 31 times more. And then after 30 messages, you might already be at\nalmost a quarter million cumulative tokens. Now, on top of all of your own messages, Claude will\nalso reload your Claude.md, your MCP servers, your system prompts, your skills, your files,\non every single turn. And this is invisible overhead, but it is constantly dripping into\nyour context and your tokens. And a really important thing to realize is that bloated\ncontext doesn't just cost you more money, but it also produces worse output. So you're paying more\nand you're getting less. There's this phenomenon called lost in the middle, which basically says\nthat models are paying the most intention in the beginning of your session and kind of at the end.\nSo all that stuff in the middle of your session, kind of in this dip, is getting ignored. All right,\nso now that we kind of understand a little bit more about how Cloud Code works and how tokens\nwork, let's move into the hacks. We're going to start here with tier one hacks. These are the\nones that are going to be super easy to implement and everyone should be able to understand.\nSo we've got nine of these.\nNumber one is to start fresh conversations.\nUse slash clear between unrelated tasks.\nDon't carry context about topic A into a conversation about topic B.\nSo every single message in a long chat is exponentially more expensive than the same message in a fresh chat.\nSo this one habit is the number one thing that extends your session life.\nAnd it's pretty obvious based on what we just talked about.\nSo that's why this was number one.\nOkay, number two is to disconnect MCP servers.\nEvery single connected MCP server loads all of its tool definitions into your context on every message.\nThis is another source of completely invisible tokens that might just be eating up and eating away.\nSo one server alone might be something like 18,000 tokens per message.\nSo run MCP at the start of each session and disconnect the ones that you don't need.\nAnd better yet, if you're able to find CLIs for something,\nso for example, rather than having the Google Workspace or Google Calendar MCP server,\nwhich eats a lot of tokens, just use the Google Workspace CLI. It's faster, it's cheaper,\nand I think the future is moving towards having our agents use CLIs rather than MCPs.\nAll right, number three, batch prompts into one message. Three separate messages cost three times\nwhat one combined message costs because of the way the tokens work, right? Instead of summarize\nthis as one message and then now extract the issues, now suggest a fix, send it all in one\nprompt. If clogged something slightly wrong, edit your original message and regenerate instead of\nsending a full follow-up correction. Follow-ups stack onto history permanently while edits replace\nthe bad exchange entirely. Now I will say there is an argument to be made here that potentially\ndoing it this way where you're doing task one, task two, then task three might actually be better\noutput quality. I think it depends on the actual use case. Basically the idea would be if you can\ngive AI one specific task at a time, it's going to do better because it's more specialized and it's\nmore focused. But this is definitely something that you should be aware of. Okay, number four\nis to use plan mode before any real task. This lets Claude map out the approach, ask you the\nright questions, and it prevents the single biggest source of token waste, which is just\nhaving Claude go down the wrong path, writing code, and then basically everything that it just did,\nyou have to basically like scrap and redo. It's just a huge waste of time and tokens. So you can\nadd something like this to your Claude.md. Do not make any changes until you have 95% confidence in\nwhat you need to build, ask me follow-up questions until you reach that confidence level. This is\nsomething that I'm putting into all of my Cloud.nmds when I am having it help me build things.\nNumber five, we have run slash context and slash cost. Slash context shows you exactly what's\neating your tokens right now. So your conversation history, your MCP overhead, loaded files, stuff\nlike that. And slash cost shows your actual token usage and estimated spend for that current session.\nMost people have no idea where their tokens are going, and these two commands make the invisible\nvisible. Because if you don't actually know that you're bleeding because of MCPs, then how would\nyou be able to fix that? So when you run slash context, this is what it will look like. It'll\nbasically give you a screenshot of how many tokens you're at, what is the cap, and it will estimate\nbased on the different categories. And what I did here is this was ran in a completely fresh\nsession, no chats. So what that tells me is, okay, before I even talk to Claude, I'm already down 51,000\ntokens because of things like the system prompt, the system tools, my custom agents, my skills,\nmemory files. And here I've actually cleared out all the MCPs. So there wasn't anything in there,\nbut those can, like I said, completely blow up your tokens right from the get go. Okay. Number\nsix is to set up a status line. This kind of goes hand in hand with having more visibility.\nYou only actually see this in your terminal though. So you will have to do it there.\nAnd it basically lets you see what's going on. So right here, you can see in my terminal,\nI've got this set up so that I can see the model I'm using. I can see a visual kind of progress\nbar of my usage. And then I can see 5% of my whole 1 million context window. And I can see 52,000\ntokens out of 1000,000, which is a million. And just to clarify, this isn't my session, like my\nfive hour session. This is basically just indicating that I'm 5% of the way or 52k out of 1000k. So\nall you have to do is in cloud code in the terminal, do slash status line and explain that you want to\nreplicate this setup. Number seven is just super simple, but keep your dashboard open. Same thing\nwith visibility. You might run into issues with your limit and just get hit out of nowhere. But\nif you have it pulled up next to you or you have it ready so that you can switch into that tab and\ncheck every 20, 40 minutes, then you're going to be able to pace yourself a little bit better.\nYou could even set up automation to basically check in on it every 30 minutes and send you\na text or a Slack message and say, hey, by the way, you're getting near your usage.\nAll right. So number eight, we have be smart with pasting. Before you drop a document or a file or\nsomething large, just ask yourself, does Claude need to read this whole thing?\nSometimes it does, sometimes it needs that full context, but sometimes it just needs\none section or one page.\nSo if the bug...\nSo if the bug is in one function,\nthen paste just that function.\nOr if it just needs the context of one little paragraph,\njust paste that.\nClaude needs to be precise about what it reads,\nbut you also need to be precise about what you feed it.\nAnd number nine, our last tier one hack\nis to actually watch Claude code work.\nDon't just fire off a prompt and walk away or switch tabs.\nWatch what Claude is doing, especially on longer tasks.\nAnd this is because if you actually sit and watch it,\nsometimes you'll realize it's going down the wrong path.\nSometimes it gets stuck in its own loops,\nrereads the same files, things like that.\nSo if it's doing that,\nyou might as well just stop it right there.\nKind of the same idea as plan mode.\nWhy would you let it go down the wrong path,\nwaste all your tokens, and then just have to scrap it all?\nIn a bad loop, 80% of the tokens are being used,\nproducing zero value.\nSo if you're able to just watch your session run until you know it's going down the right path, it could save you thousands of tokens.\nAll right, let's kick it up a little bit. Let's move into our tier two hacks.\nAnd for these ones, we have five of them.\nSo number one is to keep your Claude.md file lean.\nPlace it in your project root, whether that is globally or in local project.\nClaude auto reads it at the start of every single chat as system context.\nSo keep it under 200 lines.\ninclude things like your tech stack, your coding conventions, your build commands,\nthe 95% confidence rule, only the most important things. And you need to treat this like an index\nroute to where more data lives. And it's a complete mindset shift. This file basically\njust tells cloud code, where is everything that it needs and what to do every single time.\nSo it can point to files that are huge, but that way it just says, okay, I don't need this right\nnow, but if I do need this, I know exactly where to look. And because it knows exactly where to\nlook, it's not going to waste time or tokens searching through and reading other files.\nIt's just able to grab it right there by the file name.\nAnd the reason I say this is a mindset shift\nbecause you should be doing this with other things,\nnot just your Cloud.md, with your skills\nor with your master reference guide sheets.\nI saw someone talking about how they created an index\nthat's super, super lean,\nand it shows Cloud Code exactly where to go\nin the Cloud Code documentation.\nSo if it needs help with something related to Cloud Code,\nit doesn't have to search through the whole database.\nIt can just say, okay, here's my index file.\nI know exactly which URL to look up at.\nSuper simple.\nYou wanna keep this lean and trim it all the time.\nIt's always a work in progress because every single chat, not just like your session, every\nsingle message, cloud.md gets read. So if your cloud.md file is a thousand lines, every single\ntime you shoot off a message, even if you just say hi, the whole thing's going to get read.\nOkay, number two here is to be surgical with file references. Don't just say something like,\nhere's my whole repo, go find the bug. Say something more like, check the verify user\nfunction inside the auth.js file. Or you can also use at file name to point at specific files\ninstead of once again, letting Claude explore freely.\nThe whole idea of being specific and routing.\nAll right, so number three,\nI'm saying to compact at around 60% capacity.\nAuto-compact triggers at like 95%,\nby which point your context is already pretty degraded.\nSo run slash context to check your capacity percentage,\nor you should have the status line set up.\nAnd at about 60%, just run the slash compact\nwith specific instructions\non what it should actually be preserving.\nAfter three to four compacts in a row,\nthe quality does start to degrade.\nSo at that point, once you've done three or four,\njust get a session summary slash clear, give the session summary back, and then keep going.\nAll right, so number four, short breaks are actually costing you. Cloud Code uses prompt\ntraining to avoid reprocessing unchanged context, but the cache has a five-minute timeout. So if you\nstep away and you come back and it's been longer than five minutes, your next message reprocesses\neverything from scratch at full cost. And that is why some people feel like their usage just\nrandomly spikes if they might have, you know, stepped away and came back. So if you're going\nto do that, just consider doing a slash compact or a slash clear before you step away. All right,\nnumber five, command output bloat. When Claude runs shell commands, the full output enters your\ncontext window. So if you have a command that it comes back with 200 commits or, you know, just tons\nand tons of data, then all of that is tokens that get sent to your model. So really the takeaway\nhere is to be intentional about what you let Claude run. If you know in a certain project that\nthat doesn't need to use certain commands,\nthen you can go ahead and in that project,\ndeny those permissions.\nAnd this is another one that seems like it's invisible\nbecause when it runs like a bash or certain commands,\nit basically just has like one line\nand you don't actually see all the tokens\nthat it has sent there.\nAll right, so sitting here editing this video\nand there's just one more thing\nthat I wanted to get off my chest\nand it's basically about hitting your limit.\nAnd the goal of this video and your goal should be\nto optimize so that you don't hit your limit,\nBut I don't think that you should associate hitting your limit with like, it shouldn't be a negative connotation.\nBecause ultimately, if you're doing a lot of these hacks and you are not just like being wasteful with tokens,\nthen hitting your limit is actually a good thing if you think about it.\nBecause it means that you are using this tool so much.\nAnd I think that's what you want to be.\nI think you want to be a power user of this tool to the point where it's like, got to wait again.\nAnd, you know, waiting sucks.\nbut people that are using it so much\nare going to be so much more productive\nand so much farther ahead\nthan people who are never hitting their limits,\nnot getting their money's worth\nand not truly getting the leverage\nthat you are now getting.\nSo anyways, quick little raw rant there,\nbut I think it's an important mindset shift to have,\njust something to think about.\nAll right, so we're moving on to tier three now.\nI hope you guys feel like you already have a lot of things\nthat you wanna implement\nand these ones are getting a little crazier as well.\nSo we've got four of these here\nand I've got a few bonus ones also,\nbut number one is to pick the right model.\nSo Sonnet for your default, most coding work,\nHaiku for sub-agents, formatting, simple tasks,\nOpus for deep architectural planning\nand only when Sonnet wasn't enough,\ntry to keep this under 20% of usage\nor unless you just really, really need it for that project.\nNow, a little tip here is when you have a huge code base\nand you wanna do certain things like maybe a review,\nthen try bringing in Codex.\nThere is an official plugin now\nand I made a video about this.\nI'll tag that right up here. But you could basically have, you know, Opus and Sonnet working\ntogether to build you, you know, a project or a code base. And then you could just bring in codecs\nto actually review everything. And that way you're saving yourself on the Claude tokens.\nThe next one, number two here is the cost of sub agents. Agent workflows use roughly seven to 10\ntimes more tokens than a standard single agent session. Now, why is that? Because they wake up\nwith their own full context and it's a separate instance. So they basically have to reload\neverything when you start up the new session. All of those files, all of the system tools,\nlike everything like that. Now what you can do though, which is helpful, is to delegate to sub\nagents for one-off tasks, especially if you want that one-off task to use haiku. So maybe you need\nto process a lot of information, or maybe you need to do a ton of research and get just like a summary\nback. Now yes, tokens are still tokens no matter what at the end of the day, but if you can make\n80% of your tokens a cheaper model rather than 80% of your tokens an expensive model, then you're\ngoing to be saving money. And then of course, agent teams are cool. Sometimes I really do actually\nlike them and it helps me get more higher quality outputs, but they're very, very expensive. So\ntry to use them very sparingly. All right. So number three is to understand peak hours. So we\njust talked about at the beginning, how they've adjusted how fast your five hour session window\ndrains based on demand during the peak hours, which are 8am to 2pm Eastern time on weekdays,\nbut off-peak this is when your usage is kind of either normal or it lasts a little longer\nand these are afternoons evenings weekends so if you actually think about this strategically\nmaybe you want to make sure that you're running big refactors or multi-agent sessions or big\nprojects during off-peak hours only otherwise you're going to you know drain right through that\npeak session and on top of this we'll call this a little hack 3.5 which is the one i kind of\nalluded to earlier when i said hey just keep open your clot account so you can see your usage at all\ntimes. If you're near a reset and you have room left in your allocation, then go heavy. Try to\nhit that usage limit before it resets. Get your money's worth. Let your agents go loose at that\npoint. And on the other side, if you're getting near your limit, but you still have lots of time,\nthen step away.\nThis is your time to take a break, take a walk, make some lunch, come back with a full\nbudget instead of burning the last 5% on something small and getting stuck mid-task and having\nto just kind of, you know, lose that flow state that you might've been in.\nOkay, number four, your system's constitution, which is claw.md. This should contain stable\ndecisions, architecture rules, and progress summaries. Think of it like the source of truth\nthat makes every prompt shorter and shorter. Save decisions, not conversations. Every architectural\ncall that you store there is a paragraph that you never have to type again. So this builds on top of\nthe way that you were thinking about it back in tier one. You can add rules in there that basically\ntell it, hey, I want you to help me make sure I'm being smart about tokens. Use subagents for any\nexploration or research. If a task needs three plus files or multi-file analysis, spawn a sub\nagent and only return summarized insights. Spawn that sub agent in Haiku. And here's a little prompt\nthat I have at the bottom of mycloud.md. And I will say before I read this out, you have to be\ncareful because when you make a file like this kind of self-learning or self-evolving, you have\nto check on it frequently because you don't want it to accidentally get too bloated. But here I said\napplied learning. When something fails repeatedly, when Nate has to re-explain, or when a workaround\nis found for a platform tool or limitation, add a one-line bullet here. Keep each bullet under 15\nwords, no explanations, only add things that will save time in future sessions. And then it's got\nsome bullets. Now, I'm not saying this is the most optimal prompt, but I think this sort of system of\nhaving your Cloud.MD actually learn and continuously think about how it can save you time and tokens\nis a good idea to play with. All right. So I know that we just went through a ton of stuff.\nThis whole slide deck will be available for download for free in my FreeSchool community.\nthe link for that will be down in the description. But right now, what you should go do are these\nthings. Go run slash context, see what it looks like. Go to some of your active sessions, run slash\ncost. Status line, make sure it's showing your model, your context percentage, and your token\ncount. Make sure you pull up your cloud usage dashboard so you can see your remaining allocation\nand what time it resets. Disconnect unused MCP servers. Start complex tasks in plan mode. Use\nslash clear when you're switching to an unrelated task. Manually compact at 60% context. Batch your\nmulti-step instructions into single messages and schedule heavy sessions for off-peak hours and\nreally just be mindful about the actual timing. So I wanted to kind of leave you guys with one,\nmaybe two messages. The first thing is just the idea that there is a balance between quality\nand cost. And so that's kind of a game that you have to play a little bit. And sometimes you do\nhave to go for the higher quality, which ultimately is going to cost you more money. And that's just\nthe way it works. But the other thing is just to keep it simple and think about what we talked\nabout at the beginning of this video, how tokens actually work, how Claude Code actually charges\nyou. Most people don't need a bigger plan. They need to stop resending their entire conversation\nhistory 30 times when you could just send it, you know, five times. It's not a limits problem. It's\na context hygiene problem. But anyways, that is going to do it for this one. If you guys enjoyed\nor learned something new, please give it a like. It helps me out a ton. And as always, I appreciate\nyou guys making it to the end of the video. I'll see you on the next one. Thanks everyone.",
  "language": "en",
  "duration": 1136.570375,
  "segments": [
    {
      "start": 0.0,
      "end": 3.84,
      "text": "In the past week or so, so many people have been complaining about hitting their cloud code limit",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07983482284033859,
      "no_speech_prob": 2.1090045942279145e-12,
      "compression_ratio": 1.7719298245614035
    },
    {
      "start": 3.84,
      "end": 8.54,
      "text": "insanely fast. Claims like one prompt that is about 1% of the limit is now around 10%.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07983482284033859,
      "no_speech_prob": 2.1090045942279145e-12,
      "compression_ratio": 1.7719298245614035
    },
    {
      "start": 8.54,
      "end": 11.96,
      "text": "You could go through X and find tons and tons of threads about this topic.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07983482284033859,
      "no_speech_prob": 2.1090045942279145e-12,
      "compression_ratio": 1.7719298245614035
    },
    {
      "start": 12.3,
      "end": 16.72,
      "text": "Even on a $200 per month plan, people are reaching the session limit way too fast.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07983482284033859,
      "no_speech_prob": 2.1090045942279145e-12,
      "compression_ratio": 1.7719298245614035
    },
    {
      "start": 17.12,
      "end": 20.36,
      "text": "And then we got this post from an Anthropic employee that basically said that they",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07983482284033859,
      "no_speech_prob": 2.1090045942279145e-12,
      "compression_ratio": 1.7719298245614035
    },
    {
      "start": 20.36,
      "end": 25.32,
      "text": "are working on a little change with peak hours and off-peak hours. But even after that,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07983482284033859,
      "no_speech_prob": 2.1090045942279145e-12,
      "compression_ratio": 1.7719298245614035
    },
    {
      "start": 25.4,
      "end": 28.68,
      "text": "some people were saying they were still hitting it really quick, even during off-peak hours.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07983482284033859,
      "no_speech_prob": 2.1090045942279145e-12,
      "compression_ratio": 1.7719298245614035
    },
    {
      "start": 28.68,
      "end": 39.02,
      "text": "So anyways, I've been playing around a ton, trying different things, doing research, and I have 18 token management hacks for you guys that I've organized from tier 1 all the way up to tier 3 so they get more advanced as we go.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061447073431576,
      "no_speech_prob": 2.380799550560919e-12,
      "compression_ratio": 1.6636904761904763
    },
    {
      "start": 39.3,
      "end": 45.26,
      "text": "I'm very confident that by the end of this video, you will feel like your Claude code usage has doubled, tripled, maybe even 5x'd.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061447073431576,
      "no_speech_prob": 2.380799550560919e-12,
      "compression_ratio": 1.6636904761904763
    },
    {
      "start": 45.5,
      "end": 47.82,
      "text": "So let's not waste any time and just get straight into the video.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061447073431576,
      "no_speech_prob": 2.380799550560919e-12,
      "compression_ratio": 1.6636904761904763
    },
    {
      "start": 48.08,
      "end": 55.2,
      "text": "So as I've been optimizing my own token management, I think that what's really important to realize first is how tokens actually work.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.061447073431576,
      "no_speech_prob": 2.380799550560919e-12,
      "compression_ratio": 1.6636904761904763
    },
    {
      "start": 55.2,
      "end": 59.98,
      "text": "Because once you realize how Claude uses tokens, it makes it very clear how you should actually",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10610677109283655,
      "no_speech_prob": 1.818095804115294e-12,
      "compression_ratio": 1.821875
    },
    {
      "start": 59.98,
      "end": 63.3,
      "text": "reverse engineer the way that you work in order to use less tokens.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10610677109283655,
      "no_speech_prob": 1.818095804115294e-12,
      "compression_ratio": 1.821875
    },
    {
      "start": 63.72,
      "end": 67.98,
      "text": "So a token is the smallest unit of text that an AI model reads and charges you for.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10610677109283655,
      "no_speech_prob": 1.818095804115294e-12,
      "compression_ratio": 1.821875
    },
    {
      "start": 68.26,
      "end": 71.46,
      "text": "It's roughly one token is one word, but that's not explicitly true.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10610677109283655,
      "no_speech_prob": 1.818095804115294e-12,
      "compression_ratio": 1.821875
    },
    {
      "start": 71.64,
      "end": 72.6,
      "text": "Kind of just a good baseline.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10610677109283655,
      "no_speech_prob": 1.818095804115294e-12,
      "compression_ratio": 1.821875
    },
    {
      "start": 72.6,
      "end": 76.98,
      "text": "So every time that you send a message, Claude rereads the entire conversation from the beginning,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10610677109283655,
      "no_speech_prob": 1.818095804115294e-12,
      "compression_ratio": 1.821875
    },
    {
      "start": 77.38,
      "end": 79.54,
      "text": "and all of those are tokens that it's charging you for.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10610677109283655,
      "no_speech_prob": 1.818095804115294e-12,
      "compression_ratio": 1.821875
    },
    {
      "start": 79.96,
      "end": 83.48,
      "text": "So message one, it will read it, then it will read its reply, and then message two,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.10610677109283655,
      "no_speech_prob": 1.818095804115294e-12,
      "compression_ratio": 1.821875
    },
    {
      "start": 83.48,
      "end": 86.36,
      "text": "and then the reply all the way up to your latest prompt.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 86.64,
      "end": 87.94,
      "text": "And it does that every single time.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 88.22,
      "end": 90.3,
      "text": "And I think that alone is a huge light bulb moment",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 90.3,
      "end": 91.24,
      "text": "for a lot of people.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 91.58,
      "end": 93.74,
      "text": "This means as you're having a conversation with Claude,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 94.1,
      "end": 96.3,
      "text": "your cost is compounding, not just adding,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 96.66,
      "end": 97.84,
      "text": "it's exponentially growing.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 98.3,
      "end": 100.46,
      "text": "Meaning message one might cost 500 tokens,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 100.84,
      "end": 102.56,
      "text": "message 30 costs 15,000",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 102.56,
      "end": 104.54,
      "text": "because it's rereading everything before it.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 105.04,
      "end": 107.8,
      "text": "One developer actually tracked a 100 plus message chat",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 107.8,
      "end": 110.82,
      "text": "and found that 98.5% of all the tokens",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.12621915162499273,
      "no_speech_prob": 1.12024430530544e-12,
      "compression_ratio": 1.6084142394822007
    },
    {
      "start": 110.82,
      "end": 114.16,
      "text": "were just spent rereading the old chat history in the session.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 114.3,
      "end": 115.28,
      "text": "Like that's a huge waste.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 115.7,
      "end": 117.42,
      "text": "Now, yes, the argument has to be made that,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 117.68,
      "end": 120.06,
      "text": "well, it needs the context and it needs to understand what we're doing,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 120.2,
      "end": 122.48,
      "text": "but still 98.5% is crazy.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 122.82,
      "end": 124.4,
      "text": "So take a quick look at this graphic here.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 124.68,
      "end": 126.48,
      "text": "Along the x-axis, we have message number.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 126.88,
      "end": 130.04,
      "text": "And as it increases, you can see that we have our per message cost",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 130.04,
      "end": 133.06,
      "text": "and our cumulative tokens increasing, but it's not linear.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 133.3,
      "end": 136.56,
      "text": "It's basically each message is rereading all of the past ones",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 136.56,
      "end": 137.64,
      "text": "and it has to count that in.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 137.64,
      "end": 139.12,
      "text": "So message one could be 500,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.14601814222035917,
      "no_speech_prob": 2.1759824844758135e-12,
      "compression_ratio": 1.6897590361445782
    },
    {
      "start": 139.12,
      "end": 144.92,
      "text": "message 30 could be 15,500, which is 31 times more. And then after 30 messages, you might already be at",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0861426154486567,
      "no_speech_prob": 2.2982303993918096e-12,
      "compression_ratio": 1.68955223880597
    },
    {
      "start": 144.92,
      "end": 149.72,
      "text": "almost a quarter million cumulative tokens. Now, on top of all of your own messages, Claude will",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0861426154486567,
      "no_speech_prob": 2.2982303993918096e-12,
      "compression_ratio": 1.68955223880597
    },
    {
      "start": 149.72,
      "end": 154.32,
      "text": "also reload your Claude.md, your MCP servers, your system prompts, your skills, your files,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0861426154486567,
      "no_speech_prob": 2.2982303993918096e-12,
      "compression_ratio": 1.68955223880597
    },
    {
      "start": 154.44,
      "end": 159.14,
      "text": "on every single turn. And this is invisible overhead, but it is constantly dripping into",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0861426154486567,
      "no_speech_prob": 2.2982303993918096e-12,
      "compression_ratio": 1.68955223880597
    },
    {
      "start": 159.14,
      "end": 162.56,
      "text": "your context and your tokens. And a really important thing to realize is that bloated",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0861426154486567,
      "no_speech_prob": 2.2982303993918096e-12,
      "compression_ratio": 1.68955223880597
    },
    {
      "start": 162.56,
      "end": 166.98,
      "text": "context doesn't just cost you more money, but it also produces worse output. So you're paying more",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.0861426154486567,
      "no_speech_prob": 2.2982303993918096e-12,
      "compression_ratio": 1.68955223880597
    },
    {
      "start": 166.98,
      "end": 171.04,
      "text": "and you're getting less. There's this phenomenon called lost in the middle, which basically says",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07000185973452826,
      "no_speech_prob": 2.2104473199752173e-12,
      "compression_ratio": 1.8144654088050314
    },
    {
      "start": 171.04,
      "end": 175.36,
      "text": "that models are paying the most intention in the beginning of your session and kind of at the end.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07000185973452826,
      "no_speech_prob": 2.2104473199752173e-12,
      "compression_ratio": 1.8144654088050314
    },
    {
      "start": 175.46,
      "end": 180.08,
      "text": "So all that stuff in the middle of your session, kind of in this dip, is getting ignored. All right,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07000185973452826,
      "no_speech_prob": 2.2104473199752173e-12,
      "compression_ratio": 1.8144654088050314
    },
    {
      "start": 180.12,
      "end": 183.32,
      "text": "so now that we kind of understand a little bit more about how Cloud Code works and how tokens",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07000185973452826,
      "no_speech_prob": 2.2104473199752173e-12,
      "compression_ratio": 1.8144654088050314
    },
    {
      "start": 183.32,
      "end": 187.82,
      "text": "work, let's move into the hacks. We're going to start here with tier one hacks. These are the",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07000185973452826,
      "no_speech_prob": 2.2104473199752173e-12,
      "compression_ratio": 1.8144654088050314
    },
    {
      "start": 187.82,
      "end": 190.82,
      "text": "ones that are going to be super easy to implement and everyone should be able to understand.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07000185973452826,
      "no_speech_prob": 2.2104473199752173e-12,
      "compression_ratio": 1.8144654088050314
    },
    {
      "start": 191.28,
      "end": 192.64,
      "text": "So we've got nine of these.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 193.0,
      "end": 194.94,
      "text": "Number one is to start fresh conversations.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 195.58,
      "end": 198.12,
      "text": "Use slash clear between unrelated tasks.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 198.54,
      "end": 201.8,
      "text": "Don't carry context about topic A into a conversation about topic B.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 202.44,
      "end": 207.34,
      "text": "So every single message in a long chat is exponentially more expensive than the same message in a fresh chat.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 207.34,
      "end": 211.64,
      "text": "So this one habit is the number one thing that extends your session life.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 211.78,
      "end": 214.14,
      "text": "And it's pretty obvious based on what we just talked about.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 214.42,
      "end": 215.58,
      "text": "So that's why this was number one.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 216.22,
      "end": 219.44,
      "text": "Okay, number two is to disconnect MCP servers.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.18763860066731772,
      "no_speech_prob": 1.4272582681690293e-12,
      "compression_ratio": 1.7244897959183674
    },
    {
      "start": 219.44,
      "end": 224.58,
      "text": "Every single connected MCP server loads all of its tool definitions into your context on every message.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07435927671544693,
      "no_speech_prob": 1.7830883254285612e-12,
      "compression_ratio": 1.6611842105263157
    },
    {
      "start": 225.0,
      "end": 229.68,
      "text": "This is another source of completely invisible tokens that might just be eating up and eating away.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07435927671544693,
      "no_speech_prob": 1.7830883254285612e-12,
      "compression_ratio": 1.6611842105263157
    },
    {
      "start": 230.02,
      "end": 233.82,
      "text": "So one server alone might be something like 18,000 tokens per message.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07435927671544693,
      "no_speech_prob": 1.7830883254285612e-12,
      "compression_ratio": 1.6611842105263157
    },
    {
      "start": 234.14,
      "end": 238.5,
      "text": "So run MCP at the start of each session and disconnect the ones that you don't need.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07435927671544693,
      "no_speech_prob": 1.7830883254285612e-12,
      "compression_ratio": 1.6611842105263157
    },
    {
      "start": 238.9,
      "end": 241.48,
      "text": "And better yet, if you're able to find CLIs for something,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07435927671544693,
      "no_speech_prob": 1.7830883254285612e-12,
      "compression_ratio": 1.6611842105263157
    },
    {
      "start": 241.58,
      "end": 246.2,
      "text": "so for example, rather than having the Google Workspace or Google Calendar MCP server,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07435927671544693,
      "no_speech_prob": 1.7830883254285612e-12,
      "compression_ratio": 1.6611842105263157
    },
    {
      "start": 246.2,
      "end": 250.78,
      "text": "which eats a lot of tokens, just use the Google Workspace CLI. It's faster, it's cheaper,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07873801674161639,
      "no_speech_prob": 2.0205803176465142e-12,
      "compression_ratio": 1.5972222222222223
    },
    {
      "start": 251.3,
      "end": 256.48,
      "text": "and I think the future is moving towards having our agents use CLIs rather than MCPs.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07873801674161639,
      "no_speech_prob": 2.0205803176465142e-12,
      "compression_ratio": 1.5972222222222223
    },
    {
      "start": 257.12,
      "end": 262.16,
      "text": "All right, number three, batch prompts into one message. Three separate messages cost three times",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07873801674161639,
      "no_speech_prob": 2.0205803176465142e-12,
      "compression_ratio": 1.5972222222222223
    },
    {
      "start": 262.16,
      "end": 267.12,
      "text": "what one combined message costs because of the way the tokens work, right? Instead of summarize",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07873801674161639,
      "no_speech_prob": 2.0205803176465142e-12,
      "compression_ratio": 1.5972222222222223
    },
    {
      "start": 267.12,
      "end": 272.18,
      "text": "this as one message and then now extract the issues, now suggest a fix, send it all in one",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.07873801674161639,
      "no_speech_prob": 2.0205803176465142e-12,
      "compression_ratio": 1.5972222222222223
    },
    {
      "start": 272.18,
      "end": 277.1,
      "text": "prompt. If clogged something slightly wrong, edit your original message and regenerate instead of",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05953714384961484,
      "no_speech_prob": 2.436565055621842e-12,
      "compression_ratio": 1.68
    },
    {
      "start": 277.1,
      "end": 282.36,
      "text": "sending a full follow-up correction. Follow-ups stack onto history permanently while edits replace",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05953714384961484,
      "no_speech_prob": 2.436565055621842e-12,
      "compression_ratio": 1.68
    },
    {
      "start": 282.36,
      "end": 287.1,
      "text": "the bad exchange entirely. Now I will say there is an argument to be made here that potentially",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05953714384961484,
      "no_speech_prob": 2.436565055621842e-12,
      "compression_ratio": 1.68
    },
    {
      "start": 287.1,
      "end": 292.94,
      "text": "doing it this way where you're doing task one, task two, then task three might actually be better",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05953714384961484,
      "no_speech_prob": 2.436565055621842e-12,
      "compression_ratio": 1.68
    },
    {
      "start": 292.94,
      "end": 297.24,
      "text": "output quality. I think it depends on the actual use case. Basically the idea would be if you can",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05953714384961484,
      "no_speech_prob": 2.436565055621842e-12,
      "compression_ratio": 1.68
    },
    {
      "start": 297.24,
      "end": 301.64,
      "text": "give AI one specific task at a time, it's going to do better because it's more specialized and it's",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05953714384961484,
      "no_speech_prob": 2.436565055621842e-12,
      "compression_ratio": 1.68
    },
    {
      "start": 301.64,
      "end": 305.12,
      "text": "more focused. But this is definitely something that you should be aware of. Okay, number four",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05249885490962437,
      "no_speech_prob": 2.53434057594093e-12,
      "compression_ratio": 1.6873156342182891
    },
    {
      "start": 305.12,
      "end": 310.36,
      "text": "is to use plan mode before any real task. This lets Claude map out the approach, ask you the",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05249885490962437,
      "no_speech_prob": 2.53434057594093e-12,
      "compression_ratio": 1.6873156342182891
    },
    {
      "start": 310.36,
      "end": 314.3,
      "text": "right questions, and it prevents the single biggest source of token waste, which is just",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05249885490962437,
      "no_speech_prob": 2.53434057594093e-12,
      "compression_ratio": 1.6873156342182891
    },
    {
      "start": 314.3,
      "end": 318.46,
      "text": "having Claude go down the wrong path, writing code, and then basically everything that it just did,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05249885490962437,
      "no_speech_prob": 2.53434057594093e-12,
      "compression_ratio": 1.6873156342182891
    },
    {
      "start": 318.72,
      "end": 323.06,
      "text": "you have to basically like scrap and redo. It's just a huge waste of time and tokens. So you can",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05249885490962437,
      "no_speech_prob": 2.53434057594093e-12,
      "compression_ratio": 1.6873156342182891
    },
    {
      "start": 323.06,
      "end": 327.5,
      "text": "add something like this to your Claude.md. Do not make any changes until you have 95% confidence in",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05249885490962437,
      "no_speech_prob": 2.53434057594093e-12,
      "compression_ratio": 1.6873156342182891
    },
    {
      "start": 327.5,
      "end": 332.08,
      "text": "what you need to build, ask me follow-up questions until you reach that confidence level. This is",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09901515175314511,
      "no_speech_prob": 1.6749370987390044e-12,
      "compression_ratio": 1.7026239067055393
    },
    {
      "start": 332.08,
      "end": 336.44,
      "text": "something that I'm putting into all of my Cloud.nmds when I am having it help me build things.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09901515175314511,
      "no_speech_prob": 1.6749370987390044e-12,
      "compression_ratio": 1.7026239067055393
    },
    {
      "start": 336.74,
      "end": 341.76,
      "text": "Number five, we have run slash context and slash cost. Slash context shows you exactly what's",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09901515175314511,
      "no_speech_prob": 1.6749370987390044e-12,
      "compression_ratio": 1.7026239067055393
    },
    {
      "start": 341.76,
      "end": 346.26,
      "text": "eating your tokens right now. So your conversation history, your MCP overhead, loaded files, stuff",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09901515175314511,
      "no_speech_prob": 1.6749370987390044e-12,
      "compression_ratio": 1.7026239067055393
    },
    {
      "start": 346.26,
      "end": 351.94,
      "text": "like that. And slash cost shows your actual token usage and estimated spend for that current session.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09901515175314511,
      "no_speech_prob": 1.6749370987390044e-12,
      "compression_ratio": 1.7026239067055393
    },
    {
      "start": 352.6,
      "end": 356.34,
      "text": "Most people have no idea where their tokens are going, and these two commands make the invisible",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.09901515175314511,
      "no_speech_prob": 1.6749370987390044e-12,
      "compression_ratio": 1.7026239067055393
    },
    {
      "start": 356.34,
      "end": 361.36,
      "text": "visible. Because if you don't actually know that you're bleeding because of MCPs, then how would",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05501332154145112,
      "no_speech_prob": 1.9737254365603807e-12,
      "compression_ratio": 1.6590909090909092
    },
    {
      "start": 361.36,
      "end": 365.4,
      "text": "you be able to fix that? So when you run slash context, this is what it will look like. It'll",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05501332154145112,
      "no_speech_prob": 1.9737254365603807e-12,
      "compression_ratio": 1.6590909090909092
    },
    {
      "start": 365.4,
      "end": 370.74,
      "text": "basically give you a screenshot of how many tokens you're at, what is the cap, and it will estimate",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05501332154145112,
      "no_speech_prob": 1.9737254365603807e-12,
      "compression_ratio": 1.6590909090909092
    },
    {
      "start": 370.74,
      "end": 375.06,
      "text": "based on the different categories. And what I did here is this was ran in a completely fresh",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05501332154145112,
      "no_speech_prob": 1.9737254365603807e-12,
      "compression_ratio": 1.6590909090909092
    },
    {
      "start": 375.06,
      "end": 381.4,
      "text": "session, no chats. So what that tells me is, okay, before I even talk to Claude, I'm already down 51,000",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05501332154145112,
      "no_speech_prob": 1.9737254365603807e-12,
      "compression_ratio": 1.6590909090909092
    },
    {
      "start": 381.4,
      "end": 386.28,
      "text": "tokens because of things like the system prompt, the system tools, my custom agents, my skills,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05501332154145112,
      "no_speech_prob": 1.9737254365603807e-12,
      "compression_ratio": 1.6590909090909092
    },
    {
      "start": 386.34,
      "end": 391.02,
      "text": "memory files. And here I've actually cleared out all the MCPs. So there wasn't anything in there,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08674131027639728,
      "no_speech_prob": 2.534266416512332e-12,
      "compression_ratio": 1.7076923076923076
    },
    {
      "start": 391.02,
      "end": 396.16,
      "text": "but those can, like I said, completely blow up your tokens right from the get go. Okay. Number",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08674131027639728,
      "no_speech_prob": 2.534266416512332e-12,
      "compression_ratio": 1.7076923076923076
    },
    {
      "start": 396.16,
      "end": 400.64,
      "text": "six is to set up a status line. This kind of goes hand in hand with having more visibility.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08674131027639728,
      "no_speech_prob": 2.534266416512332e-12,
      "compression_ratio": 1.7076923076923076
    },
    {
      "start": 400.86,
      "end": 403.64,
      "text": "You only actually see this in your terminal though. So you will have to do it there.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08674131027639728,
      "no_speech_prob": 2.534266416512332e-12,
      "compression_ratio": 1.7076923076923076
    },
    {
      "start": 404.24,
      "end": 407.96,
      "text": "And it basically lets you see what's going on. So right here, you can see in my terminal,",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08674131027639728,
      "no_speech_prob": 2.534266416512332e-12,
      "compression_ratio": 1.7076923076923076
    },
    {
      "start": 407.96,
      "end": 412.6,
      "text": "I've got this set up so that I can see the model I'm using. I can see a visual kind of progress",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.08674131027639728,
      "no_speech_prob": 2.534266416512332e-12,
      "compression_ratio": 1.7076923076923076
    },
    {
      "start": 412.6,
      "end": 420.86,
      "text": "bar of my usage. And then I can see 5% of my whole 1 million context window. And I can see 52,000",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.057333700584642815,
      "no_speech_prob": 1.3671886973237735e-12,
      "compression_ratio": 1.6466666666666667
    },
    {
      "start": 420.86,
      "end": 425.68,
      "text": "tokens out of 1000,000, which is a million. And just to clarify, this isn't my session, like my",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.057333700584642815,
      "no_speech_prob": 1.3671886973237735e-12,
      "compression_ratio": 1.6466666666666667
    },
    {
      "start": 425.68,
      "end": 432.02,
      "text": "five hour session. This is basically just indicating that I'm 5% of the way or 52k out of 1000k. So",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.057333700584642815,
      "no_speech_prob": 1.3671886973237735e-12,
      "compression_ratio": 1.6466666666666667
    },
    {
      "start": 432.02,
      "end": 436.54,
      "text": "all you have to do is in cloud code in the terminal, do slash status line and explain that you want to",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.057333700584642815,
      "no_speech_prob": 1.3671886973237735e-12,
      "compression_ratio": 1.6466666666666667
    },
    {
      "start": 436.54,
      "end": 441.46,
      "text": "replicate this setup. Number seven is just super simple, but keep your dashboard open. Same thing",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.057333700584642815,
      "no_speech_prob": 1.3671886973237735e-12,
      "compression_ratio": 1.6466666666666667
    },
    {
      "start": 441.46,
      "end": 446.66,
      "text": "with visibility. You might run into issues with your limit and just get hit out of nowhere. But",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05642498789967357,
      "no_speech_prob": 2.645542855964078e-12,
      "compression_ratio": 1.700906344410876
    },
    {
      "start": 446.66,
      "end": 449.92,
      "text": "if you have it pulled up next to you or you have it ready so that you can switch into that tab and",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05642498789967357,
      "no_speech_prob": 2.645542855964078e-12,
      "compression_ratio": 1.700906344410876
    },
    {
      "start": 449.92,
      "end": 454.44,
      "text": "check every 20, 40 minutes, then you're going to be able to pace yourself a little bit better.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05642498789967357,
      "no_speech_prob": 2.645542855964078e-12,
      "compression_ratio": 1.700906344410876
    },
    {
      "start": 454.84,
      "end": 459.42,
      "text": "You could even set up automation to basically check in on it every 30 minutes and send you",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05642498789967357,
      "no_speech_prob": 2.645542855964078e-12,
      "compression_ratio": 1.700906344410876
    },
    {
      "start": 459.42,
      "end": 464.94,
      "text": "a text or a Slack message and say, hey, by the way, you're getting near your usage.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05642498789967357,
      "no_speech_prob": 2.645542855964078e-12,
      "compression_ratio": 1.700906344410876
    },
    {
      "start": 465.48,
      "end": 470.06,
      "text": "All right. So number eight, we have be smart with pasting. Before you drop a document or a file or",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.05642498789967357,
      "no_speech_prob": 2.645542855964078e-12,
      "compression_ratio": 1.700906344410876
    },
    {
      "start": 470.06,
      "end": 474.16,
      "text": "something large, just ask yourself, does Claude need to read this whole thing?",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.1985227476875737,
      "no_speech_prob": 8.131946179802674e-13,
      "compression_ratio": 1.4855072463768115
    },
    {
      "start": 474.42,
      "end": 477.72,
      "text": "Sometimes it does, sometimes it needs that full context, but sometimes it just needs",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.1985227476875737,
      "no_speech_prob": 8.131946179802674e-13,
      "compression_ratio": 1.4855072463768115
    },
    {
      "start": 477.72,
      "end": 478.98,
      "text": "one section or one page.",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.1985227476875737,
      "no_speech_prob": 8.131946179802674e-13,
      "compression_ratio": 1.4855072463768115
    },
    {
      "start": 479.3,
      "end": 480.0,
      "text": "So if the bug...",
      "chunk": 0,
      "language": "en",
      "avg_logprob": -0.1985227476875737,
      "no_speech_prob": 8.131946179802674e-13,
      "compression_ratio": 1.4855072463768115
    },
    {
      "start": 480.0,
      "end": 480.98,
      "text": "So if the bug is in one function,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.16097480494801591,
      "no_speech_prob": 2.8048394769969454e-12,
      "compression_ratio": 1.873846153846154
    },
    {
      "start": 481.1,
      "end": 482.28,
      "text": "then paste just that function.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.16097480494801591,
      "no_speech_prob": 2.8048394769969454e-12,
      "compression_ratio": 1.873846153846154
    },
    {
      "start": 482.42,
      "end": 484.38,
      "text": "Or if it just needs the context of one little paragraph,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.16097480494801591,
      "no_speech_prob": 2.8048394769969454e-12,
      "compression_ratio": 1.873846153846154
    },
    {
      "start": 484.58,
      "end": 485.24,
      "text": "just paste that.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.16097480494801591,
      "no_speech_prob": 2.8048394769969454e-12,
      "compression_ratio": 1.873846153846154
    },
    {
      "start": 485.6,
      "end": 487.18,
      "text": "Claude needs to be precise about what it reads,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.16097480494801591,
      "no_speech_prob": 2.8048394769969454e-12,
      "compression_ratio": 1.873846153846154
    },
    {
      "start": 487.18,
      "end": 489.58,
      "text": "but you also need to be precise about what you feed it.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.16097480494801591,
      "no_speech_prob": 2.8048394769969454e-12,
      "compression_ratio": 1.873846153846154
    },
    {
      "start": 489.94,
      "end": 491.42,
      "text": "And number nine, our last tier one hack",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.16097480494801591,
      "no_speech_prob": 2.8048394769969454e-12,
      "compression_ratio": 1.873846153846154
    },
    {
      "start": 491.42,
      "end": 494.14,
      "text": "is to actually watch Claude code work.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.16097480494801591,
      "no_speech_prob": 2.8048394769969454e-12,
      "compression_ratio": 1.873846153846154
    },
    {
      "start": 494.14,
      "end": 496.46,
      "text": "Don't just fire off a prompt and walk away or switch tabs.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 496.88,
      "end": 499.38,
      "text": "Watch what Claude is doing, especially on longer tasks.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 499.78,
      "end": 501.72,
      "text": "And this is because if you actually sit and watch it,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 501.92,
      "end": 504.0,
      "text": "sometimes you'll realize it's going down the wrong path.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 504.26,
      "end": 505.76,
      "text": "Sometimes it gets stuck in its own loops,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 506.26,
      "end": 507.9,
      "text": "rereads the same files, things like that.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 508.2,
      "end": 509.68,
      "text": "So if it's doing that,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 510.08,
      "end": 511.8,
      "text": "you might as well just stop it right there.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 512.06,
      "end": 513.3,
      "text": "Kind of the same idea as plan mode.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 513.38,
      "end": 515.02,
      "text": "Why would you let it go down the wrong path,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 515.12,
      "end": 517.14,
      "text": "waste all your tokens, and then just have to scrap it all?",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 517.44,
      "end": 520.18,
      "text": "In a bad loop, 80% of the tokens are being used,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 520.58,
      "end": 522.48,
      "text": "producing zero value.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10045786938035345,
      "no_speech_prob": 1.4898707267896327e-12,
      "compression_ratio": 1.701449275362319
    },
    {
      "start": 522.48,
      "end": 528.04,
      "text": "So if you're able to just watch your session run until you know it's going down the right path, it could save you thousands of tokens.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14192650534889914,
      "no_speech_prob": 1.661905964407684e-12,
      "compression_ratio": 1.6026490066225165
    },
    {
      "start": 528.52,
      "end": 531.12,
      "text": "All right, let's kick it up a little bit. Let's move into our tier two hacks.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14192650534889914,
      "no_speech_prob": 1.661905964407684e-12,
      "compression_ratio": 1.6026490066225165
    },
    {
      "start": 531.46,
      "end": 533.42,
      "text": "And for these ones, we have five of them.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14192650534889914,
      "no_speech_prob": 1.661905964407684e-12,
      "compression_ratio": 1.6026490066225165
    },
    {
      "start": 533.78,
      "end": 536.74,
      "text": "So number one is to keep your Claude.md file lean.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14192650534889914,
      "no_speech_prob": 1.661905964407684e-12,
      "compression_ratio": 1.6026490066225165
    },
    {
      "start": 537.26,
      "end": 541.18,
      "text": "Place it in your project root, whether that is globally or in local project.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14192650534889914,
      "no_speech_prob": 1.661905964407684e-12,
      "compression_ratio": 1.6026490066225165
    },
    {
      "start": 541.74,
      "end": 545.42,
      "text": "Claude auto reads it at the start of every single chat as system context.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14192650534889914,
      "no_speech_prob": 1.661905964407684e-12,
      "compression_ratio": 1.6026490066225165
    },
    {
      "start": 545.62,
      "end": 547.18,
      "text": "So keep it under 200 lines.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14192650534889914,
      "no_speech_prob": 1.661905964407684e-12,
      "compression_ratio": 1.6026490066225165
    },
    {
      "start": 547.18,
      "end": 550.6,
      "text": "include things like your tech stack, your coding conventions, your build commands,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07417691138482863,
      "no_speech_prob": 1.890767490492218e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 550.88,
      "end": 555.78,
      "text": "the 95% confidence rule, only the most important things. And you need to treat this like an index",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07417691138482863,
      "no_speech_prob": 1.890767490492218e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 555.78,
      "end": 560.46,
      "text": "route to where more data lives. And it's a complete mindset shift. This file basically",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07417691138482863,
      "no_speech_prob": 1.890767490492218e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 560.46,
      "end": 564.44,
      "text": "just tells cloud code, where is everything that it needs and what to do every single time.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07417691138482863,
      "no_speech_prob": 1.890767490492218e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 564.44,
      "end": 568.24,
      "text": "So it can point to files that are huge, but that way it just says, okay, I don't need this right",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07417691138482863,
      "no_speech_prob": 1.890767490492218e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 568.24,
      "end": 572.34,
      "text": "now, but if I do need this, I know exactly where to look. And because it knows exactly where to",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07417691138482863,
      "no_speech_prob": 1.890767490492218e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 572.34,
      "end": 575.66,
      "text": "look, it's not going to waste time or tokens searching through and reading other files.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07417691138482863,
      "no_speech_prob": 1.890767490492218e-12,
      "compression_ratio": 1.8
    },
    {
      "start": 575.66,
      "end": 577.78,
      "text": "It's just able to grab it right there by the file name.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 578.12,
      "end": 579.44,
      "text": "And the reason I say this is a mindset shift",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 579.44,
      "end": 581.24,
      "text": "because you should be doing this with other things,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 581.3,
      "end": 583.22,
      "text": "not just your Cloud.md, with your skills",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 583.22,
      "end": 586.26,
      "text": "or with your master reference guide sheets.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 586.58,
      "end": 588.72,
      "text": "I saw someone talking about how they created an index",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 588.72,
      "end": 589.72,
      "text": "that's super, super lean,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 590.02,
      "end": 592.08,
      "text": "and it shows Cloud Code exactly where to go",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 592.08,
      "end": 593.56,
      "text": "in the Cloud Code documentation.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 593.88,
      "end": 596.5,
      "text": "So if it needs help with something related to Cloud Code,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 596.56,
      "end": 598.44,
      "text": "it doesn't have to search through the whole database.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 598.44,
      "end": 600.28,
      "text": "It can just say, okay, here's my index file.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 600.38,
      "end": 602.12,
      "text": "I know exactly which URL to look up at.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 602.3,
      "end": 602.86,
      "text": "Super simple.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 603.24,
      "end": 605.04,
      "text": "You wanna keep this lean and trim it all the time.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12904069664772977,
      "no_speech_prob": 1.7283221295100182e-12,
      "compression_ratio": 1.7263157894736842
    },
    {
      "start": 605.04,
      "end": 609.12,
      "text": "It's always a work in progress because every single chat, not just like your session, every",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.081459018546091,
      "no_speech_prob": 1.7555412418907013e-12,
      "compression_ratio": 1.7831715210355987
    },
    {
      "start": 609.12,
      "end": 614.9,
      "text": "single message, cloud.md gets read. So if your cloud.md file is a thousand lines, every single",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.081459018546091,
      "no_speech_prob": 1.7555412418907013e-12,
      "compression_ratio": 1.7831715210355987
    },
    {
      "start": 614.9,
      "end": 618.58,
      "text": "time you shoot off a message, even if you just say hi, the whole thing's going to get read.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.081459018546091,
      "no_speech_prob": 1.7555412418907013e-12,
      "compression_ratio": 1.7831715210355987
    },
    {
      "start": 618.92,
      "end": 623.22,
      "text": "Okay, number two here is to be surgical with file references. Don't just say something like,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.081459018546091,
      "no_speech_prob": 1.7555412418907013e-12,
      "compression_ratio": 1.7831715210355987
    },
    {
      "start": 623.34,
      "end": 627.54,
      "text": "here's my whole repo, go find the bug. Say something more like, check the verify user",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.081459018546091,
      "no_speech_prob": 1.7555412418907013e-12,
      "compression_ratio": 1.7831715210355987
    },
    {
      "start": 627.54,
      "end": 634.12,
      "text": "function inside the auth.js file. Or you can also use at file name to point at specific files",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.081459018546091,
      "no_speech_prob": 1.7555412418907013e-12,
      "compression_ratio": 1.7831715210355987
    },
    {
      "start": 634.12,
      "end": 636.86,
      "text": "instead of once again, letting Claude explore freely.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 637.16,
      "end": 639.2,
      "text": "The whole idea of being specific and routing.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 639.58,
      "end": 640.26,
      "text": "All right, so number three,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 640.36,
      "end": 643.42,
      "text": "I'm saying to compact at around 60% capacity.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 643.92,
      "end": 645.76,
      "text": "Auto-compact triggers at like 95%,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 645.76,
      "end": 648.28,
      "text": "by which point your context is already pretty degraded.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 648.54,
      "end": 650.74,
      "text": "So run slash context to check your capacity percentage,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 650.86,
      "end": 652.46,
      "text": "or you should have the status line set up.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 652.74,
      "end": 655.38,
      "text": "And at about 60%, just run the slash compact",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 655.38,
      "end": 656.86,
      "text": "with specific instructions",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 656.86,
      "end": 658.36,
      "text": "on what it should actually be preserving.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 658.74,
      "end": 660.28,
      "text": "After three to four compacts in a row,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 660.46,
      "end": 661.98,
      "text": "the quality does start to degrade.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 662.08,
      "end": 664.1,
      "text": "So at that point, once you've done three or four,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.14903469205652392,
      "no_speech_prob": 1.2208085020323778e-12,
      "compression_ratio": 1.6929577464788732
    },
    {
      "start": 664.12,
      "end": 669.64,
      "text": "just get a session summary slash clear, give the session summary back, and then keep going.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06799956222078693,
      "no_speech_prob": 1.721375429350469e-12,
      "compression_ratio": 1.6916167664670658
    },
    {
      "start": 670.1,
      "end": 674.46,
      "text": "All right, so number four, short breaks are actually costing you. Cloud Code uses prompt",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06799956222078693,
      "no_speech_prob": 1.721375429350469e-12,
      "compression_ratio": 1.6916167664670658
    },
    {
      "start": 674.46,
      "end": 680.26,
      "text": "training to avoid reprocessing unchanged context, but the cache has a five-minute timeout. So if you",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06799956222078693,
      "no_speech_prob": 1.721375429350469e-12,
      "compression_ratio": 1.6916167664670658
    },
    {
      "start": 680.26,
      "end": 684.4,
      "text": "step away and you come back and it's been longer than five minutes, your next message reprocesses",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06799956222078693,
      "no_speech_prob": 1.721375429350469e-12,
      "compression_ratio": 1.6916167664670658
    },
    {
      "start": 684.4,
      "end": 688.74,
      "text": "everything from scratch at full cost. And that is why some people feel like their usage just",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06799956222078693,
      "no_speech_prob": 1.721375429350469e-12,
      "compression_ratio": 1.6916167664670658
    },
    {
      "start": 688.74,
      "end": 692.7,
      "text": "randomly spikes if they might have, you know, stepped away and came back. So if you're going",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06799956222078693,
      "no_speech_prob": 1.721375429350469e-12,
      "compression_ratio": 1.6916167664670658
    },
    {
      "start": 692.7,
      "end": 697.12,
      "text": "to do that, just consider doing a slash compact or a slash clear before you step away. All right,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.049463844299316405,
      "no_speech_prob": 2.0046115375288442e-12,
      "compression_ratio": 1.7052631578947368
    },
    {
      "start": 697.14,
      "end": 702.82,
      "text": "number five, command output bloat. When Claude runs shell commands, the full output enters your",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.049463844299316405,
      "no_speech_prob": 2.0046115375288442e-12,
      "compression_ratio": 1.7052631578947368
    },
    {
      "start": 702.82,
      "end": 708.48,
      "text": "context window. So if you have a command that it comes back with 200 commits or, you know, just tons",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.049463844299316405,
      "no_speech_prob": 2.0046115375288442e-12,
      "compression_ratio": 1.7052631578947368
    },
    {
      "start": 708.48,
      "end": 714.64,
      "text": "and tons of data, then all of that is tokens that get sent to your model. So really the takeaway",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.049463844299316405,
      "no_speech_prob": 2.0046115375288442e-12,
      "compression_ratio": 1.7052631578947368
    },
    {
      "start": 714.64,
      "end": 718.86,
      "text": "here is to be intentional about what you let Claude run. If you know in a certain project that",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.049463844299316405,
      "no_speech_prob": 2.0046115375288442e-12,
      "compression_ratio": 1.7052631578947368
    },
    {
      "start": 718.86,
      "end": 720.66,
      "text": "that doesn't need to use certain commands,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 720.66,
      "end": 722.74,
      "text": "then you can go ahead and in that project,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 722.74,
      "end": 724.8,
      "text": "deny those permissions.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 724.8,
      "end": 726.88,
      "text": "And this is another one that seems like it's invisible",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 726.88,
      "end": 730.04,
      "text": "because when it runs like a bash or certain commands,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 730.04,
      "end": 731.94,
      "text": "it basically just has like one line",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 731.94,
      "end": 733.6,
      "text": "and you don't actually see all the tokens",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 733.6,
      "end": 736.04,
      "text": "that it has sent there.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 736.04,
      "end": 738.08,
      "text": "All right, so sitting here editing this video",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 738.08,
      "end": 739.06,
      "text": "and there's just one more thing",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 739.06,
      "end": 741.02,
      "text": "that I wanted to get off my chest",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 741.02,
      "end": 743.72,
      "text": "and it's basically about hitting your limit.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 743.72,
      "end": 746.1,
      "text": "And the goal of this video and your goal should be",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 746.1,
      "end": 748.64,
      "text": "to optimize so that you don't hit your limit,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.10320627689361572,
      "no_speech_prob": 2.882966651865382e-12,
      "compression_ratio": 1.8697068403908794
    },
    {
      "start": 748.64,
      "end": 753.14,
      "text": "But I don't think that you should associate hitting your limit with like, it shouldn't be a negative connotation.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11773789723714193,
      "no_speech_prob": 2.2191276594085707e-12,
      "compression_ratio": 1.8404255319148937
    },
    {
      "start": 753.52,
      "end": 759.92,
      "text": "Because ultimately, if you're doing a lot of these hacks and you are not just like being wasteful with tokens,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11773789723714193,
      "no_speech_prob": 2.2191276594085707e-12,
      "compression_ratio": 1.8404255319148937
    },
    {
      "start": 760.34,
      "end": 762.88,
      "text": "then hitting your limit is actually a good thing if you think about it.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11773789723714193,
      "no_speech_prob": 2.2191276594085707e-12,
      "compression_ratio": 1.8404255319148937
    },
    {
      "start": 762.88,
      "end": 765.86,
      "text": "Because it means that you are using this tool so much.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11773789723714193,
      "no_speech_prob": 2.2191276594085707e-12,
      "compression_ratio": 1.8404255319148937
    },
    {
      "start": 765.9,
      "end": 767.08,
      "text": "And I think that's what you want to be.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11773789723714193,
      "no_speech_prob": 2.2191276594085707e-12,
      "compression_ratio": 1.8404255319148937
    },
    {
      "start": 767.14,
      "end": 772.54,
      "text": "I think you want to be a power user of this tool to the point where it's like, got to wait again.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11773789723714193,
      "no_speech_prob": 2.2191276594085707e-12,
      "compression_ratio": 1.8404255319148937
    },
    {
      "start": 772.54,
      "end": 773.98,
      "text": "And, you know, waiting sucks.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11773789723714193,
      "no_speech_prob": 2.2191276594085707e-12,
      "compression_ratio": 1.8404255319148937
    },
    {
      "start": 773.98,
      "end": 776.94,
      "text": "but people that are using it so much",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 776.94,
      "end": 779.0,
      "text": "are going to be so much more productive",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 779.0,
      "end": 780.3,
      "text": "and so much farther ahead",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 780.3,
      "end": 782.66,
      "text": "than people who are never hitting their limits,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 782.86,
      "end": 785.28,
      "text": "not getting their money's worth",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 785.28,
      "end": 789.0,
      "text": "and not truly getting the leverage",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 789.0,
      "end": 789.8,
      "text": "that you are now getting.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 789.92,
      "end": 792.52,
      "text": "So anyways, quick little raw rant there,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 792.64,
      "end": 794.9,
      "text": "but I think it's an important mindset shift to have,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 794.96,
      "end": 796.48,
      "text": "just something to think about.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 797.0,
      "end": 798.94,
      "text": "All right, so we're moving on to tier three now.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 799.06,
      "end": 800.96,
      "text": "I hope you guys feel like you already have a lot of things",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 800.96,
      "end": 801.62,
      "text": "that you wanna implement",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.11434101879148555,
      "no_speech_prob": 1.5371785875434307e-12,
      "compression_ratio": 1.7640845070422535
    },
    {
      "start": 801.62,
      "end": 804.52,
      "text": "and these ones are getting a little crazier as well.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 804.66,
      "end": 805.72,
      "text": "So we've got four of these here",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 805.72,
      "end": 807.06,
      "text": "and I've got a few bonus ones also,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 807.36,
      "end": 808.94,
      "text": "but number one is to pick the right model.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 809.32,
      "end": 811.5,
      "text": "So Sonnet for your default, most coding work,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 811.96,
      "end": 814.42,
      "text": "Haiku for sub-agents, formatting, simple tasks,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 814.76,
      "end": 816.26,
      "text": "Opus for deep architectural planning",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 816.26,
      "end": 817.94,
      "text": "and only when Sonnet wasn't enough,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 818.24,
      "end": 819.86,
      "text": "try to keep this under 20% of usage",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 819.86,
      "end": 822.46,
      "text": "or unless you just really, really need it for that project.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 822.72,
      "end": 824.92,
      "text": "Now, a little tip here is when you have a huge code base",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 824.92,
      "end": 826.78,
      "text": "and you wanna do certain things like maybe a review,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 827.28,
      "end": 828.92,
      "text": "then try bringing in Codex.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 829.24,
      "end": 830.42,
      "text": "There is an official plugin now",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 830.42,
      "end": 831.48,
      "text": "and I made a video about this.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.12638135728889338,
      "no_speech_prob": 2.8051515103821867e-12,
      "compression_ratio": 1.672
    },
    {
      "start": 831.48,
      "end": 834.76,
      "text": "I'll tag that right up here. But you could basically have, you know, Opus and Sonnet working",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0831239810888318,
      "no_speech_prob": 1.9204950130191634e-12,
      "compression_ratio": 1.680597014925373
    },
    {
      "start": 834.76,
      "end": 839.18,
      "text": "together to build you, you know, a project or a code base. And then you could just bring in codecs",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0831239810888318,
      "no_speech_prob": 1.9204950130191634e-12,
      "compression_ratio": 1.680597014925373
    },
    {
      "start": 839.18,
      "end": 844.06,
      "text": "to actually review everything. And that way you're saving yourself on the Claude tokens.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0831239810888318,
      "no_speech_prob": 1.9204950130191634e-12,
      "compression_ratio": 1.680597014925373
    },
    {
      "start": 844.48,
      "end": 849.32,
      "text": "The next one, number two here is the cost of sub agents. Agent workflows use roughly seven to 10",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0831239810888318,
      "no_speech_prob": 1.9204950130191634e-12,
      "compression_ratio": 1.680597014925373
    },
    {
      "start": 849.32,
      "end": 854.08,
      "text": "times more tokens than a standard single agent session. Now, why is that? Because they wake up",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0831239810888318,
      "no_speech_prob": 1.9204950130191634e-12,
      "compression_ratio": 1.680597014925373
    },
    {
      "start": 854.08,
      "end": 858.96,
      "text": "with their own full context and it's a separate instance. So they basically have to reload",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0831239810888318,
      "no_speech_prob": 1.9204950130191634e-12,
      "compression_ratio": 1.680597014925373
    },
    {
      "start": 858.96,
      "end": 863.2,
      "text": "everything when you start up the new session. All of those files, all of the system tools,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0571234146220572,
      "no_speech_prob": 2.772172465539957e-12,
      "compression_ratio": 1.7852760736196318
    },
    {
      "start": 863.32,
      "end": 867.22,
      "text": "like everything like that. Now what you can do though, which is helpful, is to delegate to sub",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0571234146220572,
      "no_speech_prob": 2.772172465539957e-12,
      "compression_ratio": 1.7852760736196318
    },
    {
      "start": 867.22,
      "end": 871.92,
      "text": "agents for one-off tasks, especially if you want that one-off task to use haiku. So maybe you need",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0571234146220572,
      "no_speech_prob": 2.772172465539957e-12,
      "compression_ratio": 1.7852760736196318
    },
    {
      "start": 871.92,
      "end": 875.88,
      "text": "to process a lot of information, or maybe you need to do a ton of research and get just like a summary",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0571234146220572,
      "no_speech_prob": 2.772172465539957e-12,
      "compression_ratio": 1.7852760736196318
    },
    {
      "start": 875.88,
      "end": 881.04,
      "text": "back. Now yes, tokens are still tokens no matter what at the end of the day, but if you can make",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0571234146220572,
      "no_speech_prob": 2.772172465539957e-12,
      "compression_ratio": 1.7852760736196318
    },
    {
      "start": 881.04,
      "end": 886.26,
      "text": "80% of your tokens a cheaper model rather than 80% of your tokens an expensive model, then you're",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.0571234146220572,
      "no_speech_prob": 2.772172465539957e-12,
      "compression_ratio": 1.7852760736196318
    },
    {
      "start": 886.26,
      "end": 891.42,
      "text": "going to be saving money. And then of course, agent teams are cool. Sometimes I really do actually",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06299538082546657,
      "no_speech_prob": 1.064784545257158e-12,
      "compression_ratio": 1.627986348122867
    },
    {
      "start": 891.42,
      "end": 896.42,
      "text": "like them and it helps me get more higher quality outputs, but they're very, very expensive. So",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06299538082546657,
      "no_speech_prob": 1.064784545257158e-12,
      "compression_ratio": 1.627986348122867
    },
    {
      "start": 896.42,
      "end": 902.44,
      "text": "try to use them very sparingly. All right. So number three is to understand peak hours. So we",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06299538082546657,
      "no_speech_prob": 1.064784545257158e-12,
      "compression_ratio": 1.627986348122867
    },
    {
      "start": 902.44,
      "end": 906.16,
      "text": "just talked about at the beginning, how they've adjusted how fast your five hour session window",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06299538082546657,
      "no_speech_prob": 1.064784545257158e-12,
      "compression_ratio": 1.627986348122867
    },
    {
      "start": 906.16,
      "end": 912.62,
      "text": "drains based on demand during the peak hours, which are 8am to 2pm Eastern time on weekdays,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.06299538082546657,
      "no_speech_prob": 1.064784545257158e-12,
      "compression_ratio": 1.627986348122867
    },
    {
      "start": 912.62,
      "end": 917.9,
      "text": "but off-peak this is when your usage is kind of either normal or it lasts a little longer",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07204151153564453,
      "no_speech_prob": 1.2944372136669546e-12,
      "compression_ratio": 1.7546583850931676
    },
    {
      "start": 917.9,
      "end": 922.34,
      "text": "and these are afternoons evenings weekends so if you actually think about this strategically",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07204151153564453,
      "no_speech_prob": 1.2944372136669546e-12,
      "compression_ratio": 1.7546583850931676
    },
    {
      "start": 922.34,
      "end": 926.98,
      "text": "maybe you want to make sure that you're running big refactors or multi-agent sessions or big",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07204151153564453,
      "no_speech_prob": 1.2944372136669546e-12,
      "compression_ratio": 1.7546583850931676
    },
    {
      "start": 926.98,
      "end": 932.04,
      "text": "projects during off-peak hours only otherwise you're going to you know drain right through that",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07204151153564453,
      "no_speech_prob": 1.2944372136669546e-12,
      "compression_ratio": 1.7546583850931676
    },
    {
      "start": 932.04,
      "end": 937.18,
      "text": "peak session and on top of this we'll call this a little hack 3.5 which is the one i kind of",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07204151153564453,
      "no_speech_prob": 1.2944372136669546e-12,
      "compression_ratio": 1.7546583850931676
    },
    {
      "start": 937.18,
      "end": 941.32,
      "text": "alluded to earlier when i said hey just keep open your clot account so you can see your usage at all",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.07204151153564453,
      "no_speech_prob": 1.2944372136669546e-12,
      "compression_ratio": 1.7546583850931676
    },
    {
      "start": 941.32,
      "end": 946.02,
      "text": "times. If you're near a reset and you have room left in your allocation, then go heavy. Try to",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.054354132675543065,
      "no_speech_prob": 1.3251162320204735e-12,
      "compression_ratio": 1.5510204081632653
    },
    {
      "start": 946.02,
      "end": 950.5,
      "text": "hit that usage limit before it resets. Get your money's worth. Let your agents go loose at that",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.054354132675543065,
      "no_speech_prob": 1.3251162320204735e-12,
      "compression_ratio": 1.5510204081632653
    },
    {
      "start": 950.5,
      "end": 955.1,
      "text": "point. And on the other side, if you're getting near your limit, but you still have lots of time,",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.054354132675543065,
      "no_speech_prob": 1.3251162320204735e-12,
      "compression_ratio": 1.5510204081632653
    },
    {
      "start": 955.4,
      "end": 956.22,
      "text": "then step away.",
      "chunk": 1,
      "language": "en",
      "avg_logprob": -0.054354132675543065,
      "no_speech_prob": 1.3251162320204735e-12,
      "compression_ratio": 1.5510204081632653
    },
    {
      "start": 956.57,
      "end": 960.35,
      "text": "This is your time to take a break, take a walk, make some lunch, come back with a full",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.08771834247990658,
      "no_speech_prob": 2.1506342713639715e-12,
      "compression_ratio": 1.6993865030674846
    },
    {
      "start": 960.35,
      "end": 965.27,
      "text": "budget instead of burning the last 5% on something small and getting stuck mid-task and having",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.08771834247990658,
      "no_speech_prob": 2.1506342713639715e-12,
      "compression_ratio": 1.6993865030674846
    },
    {
      "start": 965.27,
      "end": 968.61,
      "text": "to just kind of, you know, lose that flow state that you might've been in.",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.08771834247990658,
      "no_speech_prob": 2.1506342713639715e-12,
      "compression_ratio": 1.6993865030674846
    },
    {
      "start": 968.61,
      "end": 975.21,
      "text": "Okay, number four, your system's constitution, which is claw.md. This should contain stable",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.050861569418423416,
      "no_speech_prob": 1.7486399698021793e-12,
      "compression_ratio": 1.7218934911242603
    },
    {
      "start": 975.21,
      "end": 979.29,
      "text": "decisions, architecture rules, and progress summaries. Think of it like the source of truth",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.050861569418423416,
      "no_speech_prob": 1.7486399698021793e-12,
      "compression_ratio": 1.7218934911242603
    },
    {
      "start": 979.29,
      "end": 984.49,
      "text": "that makes every prompt shorter and shorter. Save decisions, not conversations. Every architectural",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.050861569418423416,
      "no_speech_prob": 1.7486399698021793e-12,
      "compression_ratio": 1.7218934911242603
    },
    {
      "start": 984.49,
      "end": 988.67,
      "text": "call that you store there is a paragraph that you never have to type again. So this builds on top of",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.050861569418423416,
      "no_speech_prob": 1.7486399698021793e-12,
      "compression_ratio": 1.7218934911242603
    },
    {
      "start": 988.67,
      "end": 993.15,
      "text": "the way that you were thinking about it back in tier one. You can add rules in there that basically",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.050861569418423416,
      "no_speech_prob": 1.7486399698021793e-12,
      "compression_ratio": 1.7218934911242603
    },
    {
      "start": 993.15,
      "end": 998.05,
      "text": "tell it, hey, I want you to help me make sure I'm being smart about tokens. Use subagents for any",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.050861569418423416,
      "no_speech_prob": 1.7486399698021793e-12,
      "compression_ratio": 1.7218934911242603
    },
    {
      "start": 998.05,
      "end": 1002.83,
      "text": "exploration or research. If a task needs three plus files or multi-file analysis, spawn a sub",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.042668392783717105,
      "no_speech_prob": 1.8759906822429784e-12,
      "compression_ratio": 1.6581920903954803
    },
    {
      "start": 1002.83,
      "end": 1007.73,
      "text": "agent and only return summarized insights. Spawn that sub agent in Haiku. And here's a little prompt",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.042668392783717105,
      "no_speech_prob": 1.8759906822429784e-12,
      "compression_ratio": 1.6581920903954803
    },
    {
      "start": 1007.73,
      "end": 1012.17,
      "text": "that I have at the bottom of mycloud.md. And I will say before I read this out, you have to be",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.042668392783717105,
      "no_speech_prob": 1.8759906822429784e-12,
      "compression_ratio": 1.6581920903954803
    },
    {
      "start": 1012.17,
      "end": 1017.87,
      "text": "careful because when you make a file like this kind of self-learning or self-evolving, you have",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.042668392783717105,
      "no_speech_prob": 1.8759906822429784e-12,
      "compression_ratio": 1.6581920903954803
    },
    {
      "start": 1017.87,
      "end": 1022.07,
      "text": "to check on it frequently because you don't want it to accidentally get too bloated. But here I said",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.042668392783717105,
      "no_speech_prob": 1.8759906822429784e-12,
      "compression_ratio": 1.6581920903954803
    },
    {
      "start": 1022.07,
      "end": 1026.59,
      "text": "applied learning. When something fails repeatedly, when Nate has to re-explain, or when a workaround",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.042668392783717105,
      "no_speech_prob": 1.8759906822429784e-12,
      "compression_ratio": 1.6581920903954803
    },
    {
      "start": 1026.59,
      "end": 1031.85,
      "text": "is found for a platform tool or limitation, add a one-line bullet here. Keep each bullet under 15",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07502996004544772,
      "no_speech_prob": 2.1932559934878526e-12,
      "compression_ratio": 1.615598885793872
    },
    {
      "start": 1031.85,
      "end": 1035.87,
      "text": "words, no explanations, only add things that will save time in future sessions. And then it's got",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07502996004544772,
      "no_speech_prob": 2.1932559934878526e-12,
      "compression_ratio": 1.615598885793872
    },
    {
      "start": 1035.87,
      "end": 1041.13,
      "text": "some bullets. Now, I'm not saying this is the most optimal prompt, but I think this sort of system of",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07502996004544772,
      "no_speech_prob": 2.1932559934878526e-12,
      "compression_ratio": 1.615598885793872
    },
    {
      "start": 1041.13,
      "end": 1046.51,
      "text": "having your Cloud.MD actually learn and continuously think about how it can save you time and tokens",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07502996004544772,
      "no_speech_prob": 2.1932559934878526e-12,
      "compression_ratio": 1.615598885793872
    },
    {
      "start": 1046.51,
      "end": 1050.81,
      "text": "is a good idea to play with. All right. So I know that we just went through a ton of stuff.",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07502996004544772,
      "no_speech_prob": 2.1932559934878526e-12,
      "compression_ratio": 1.615598885793872
    },
    {
      "start": 1051.01,
      "end": 1055.47,
      "text": "This whole slide deck will be available for download for free in my FreeSchool community.",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.07502996004544772,
      "no_speech_prob": 2.1932559934878526e-12,
      "compression_ratio": 1.615598885793872
    },
    {
      "start": 1055.47,
      "end": 1059.05,
      "text": "the link for that will be down in the description. But right now, what you should go do are these",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.0670105278069246,
      "no_speech_prob": 2.307484715455277e-12,
      "compression_ratio": 1.6772334293948126
    },
    {
      "start": 1059.05,
      "end": 1064.15,
      "text": "things. Go run slash context, see what it looks like. Go to some of your active sessions, run slash",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.0670105278069246,
      "no_speech_prob": 2.307484715455277e-12,
      "compression_ratio": 1.6772334293948126
    },
    {
      "start": 1064.15,
      "end": 1068.13,
      "text": "cost. Status line, make sure it's showing your model, your context percentage, and your token",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.0670105278069246,
      "no_speech_prob": 2.307484715455277e-12,
      "compression_ratio": 1.6772334293948126
    },
    {
      "start": 1068.13,
      "end": 1072.57,
      "text": "count. Make sure you pull up your cloud usage dashboard so you can see your remaining allocation",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.0670105278069246,
      "no_speech_prob": 2.307484715455277e-12,
      "compression_ratio": 1.6772334293948126
    },
    {
      "start": 1072.57,
      "end": 1079.17,
      "text": "and what time it resets. Disconnect unused MCP servers. Start complex tasks in plan mode. Use",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.0670105278069246,
      "no_speech_prob": 2.307484715455277e-12,
      "compression_ratio": 1.6772334293948126
    },
    {
      "start": 1079.17,
      "end": 1085.39,
      "text": "slash clear when you're switching to an unrelated task. Manually compact at 60% context. Batch your",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.0670105278069246,
      "no_speech_prob": 2.307484715455277e-12,
      "compression_ratio": 1.6772334293948126
    },
    {
      "start": 1085.39,
      "end": 1090.13,
      "text": "multi-step instructions into single messages and schedule heavy sessions for off-peak hours and",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04199752600296684,
      "no_speech_prob": 2.8272551400726487e-12,
      "compression_ratio": 1.7668711656441718
    },
    {
      "start": 1090.13,
      "end": 1094.55,
      "text": "really just be mindful about the actual timing. So I wanted to kind of leave you guys with one,",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04199752600296684,
      "no_speech_prob": 2.8272551400726487e-12,
      "compression_ratio": 1.7668711656441718
    },
    {
      "start": 1094.71,
      "end": 1099.97,
      "text": "maybe two messages. The first thing is just the idea that there is a balance between quality",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04199752600296684,
      "no_speech_prob": 2.8272551400726487e-12,
      "compression_ratio": 1.7668711656441718
    },
    {
      "start": 1099.97,
      "end": 1103.75,
      "text": "and cost. And so that's kind of a game that you have to play a little bit. And sometimes you do",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04199752600296684,
      "no_speech_prob": 2.8272551400726487e-12,
      "compression_ratio": 1.7668711656441718
    },
    {
      "start": 1103.75,
      "end": 1107.11,
      "text": "have to go for the higher quality, which ultimately is going to cost you more money. And that's just",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04199752600296684,
      "no_speech_prob": 2.8272551400726487e-12,
      "compression_ratio": 1.7668711656441718
    },
    {
      "start": 1107.11,
      "end": 1111.37,
      "text": "the way it works. But the other thing is just to keep it simple and think about what we talked",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.04199752600296684,
      "no_speech_prob": 2.8272551400726487e-12,
      "compression_ratio": 1.7668711656441718
    },
    {
      "start": 1111.37,
      "end": 1115.71,
      "text": "about at the beginning of this video, how tokens actually work, how Claude Code actually charges",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06513780437103689,
      "no_speech_prob": 2.729545539245648e-12,
      "compression_ratio": 1.6695402298850575
    },
    {
      "start": 1115.71,
      "end": 1119.71,
      "text": "you. Most people don't need a bigger plan. They need to stop resending their entire conversation",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06513780437103689,
      "no_speech_prob": 2.729545539245648e-12,
      "compression_ratio": 1.6695402298850575
    },
    {
      "start": 1119.71,
      "end": 1124.87,
      "text": "history 30 times when you could just send it, you know, five times. It's not a limits problem. It's",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06513780437103689,
      "no_speech_prob": 2.729545539245648e-12,
      "compression_ratio": 1.6695402298850575
    },
    {
      "start": 1124.87,
      "end": 1128.91,
      "text": "a context hygiene problem. But anyways, that is going to do it for this one. If you guys enjoyed",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06513780437103689,
      "no_speech_prob": 2.729545539245648e-12,
      "compression_ratio": 1.6695402298850575
    },
    {
      "start": 1128.91,
      "end": 1132.65,
      "text": "or learned something new, please give it a like. It helps me out a ton. And as always, I appreciate",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06513780437103689,
      "no_speech_prob": 2.729545539245648e-12,
      "compression_ratio": 1.6695402298850575
    },
    {
      "start": 1132.65,
      "end": 1136.17,
      "text": "you guys making it to the end of the video. I'll see you on the next one. Thanks everyone.",
      "chunk": 2,
      "language": "en",
      "avg_logprob": -0.06513780437103689,
      "no_speech_prob": 2.729545539245648e-12,
      "compression_ratio": 1.6695402298850575
    }
  ],
  "segmented_transcription": {
    "status": "SEGMENTED_TRANSCRIPT_OK",
    "created_at": "2026-08-01T21:42:44",
    "manifest": "/Users/sagawa/AI_WORK/video_notes/url/20260801_214052__18_Claude_Code_Token_Hacks_in_18_Minutes/segmented_work/chunk_manifest.json",
    "audio": "/Users/sagawa/AI_WORK/video_notes/url/20260801_214052__18_Claude_Code_Token_Hacks_in_18_Minutes/audio_16k_mono.wav",
    "audio_sha256": "4b897502fa1922a2c01b9d8faba78c10c5c24e840230b12ad8d7720919cda6be",
    "audio_duration": 1136.570375,
    "model": "mlx-community/whisper-large-v3-turbo",
    "language_hint": "en",
    "condition_on_previous_text": false,
    "chunk_count": 3,
    "resumed_chunks": 0,
    "merged_segment_count": 316,
    "final_segment_end": 1136.17,
    "tail_repeat_run": 1,
    "tail_repeat_text": "exploration or research. If a task needs three plus files or multi-file analysis, spawn a sub",
    "qc_ok": true,
    "qc_errors": []
  }
}