Everyone Is Wrong About Tokens - CourtneyR.dev's 2nd Brain

Everyone Is Wrong About Tokens ![rw-book-cover](https://img.youtube.com/vi/0zw-Uk9KJiA/maxresdefault.jpg) 🔗 **Source:** [Everyone Is Wrong About Tokens](https://www.youtube.com/watch?v=0zw-Uk9KJiA) > [!note]+ Document Notes > tags: [ai-economics, token-efficiency, enterprise-ai] > author: [MISSING] > source: YouTube > reference: https://www.youtube.com/watch?v=0zw-Uk9KJiA > date created: {{date:YYYY-MM-DD}} {{time:HH:mm}} > > # 🏷️ Title: Everyone is Wrong about Tokens >  ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ## 📝 Summary - Developer discovers $1.3 million monthly OpenAI token spend (603 billion tokens) for a research project - Argues current "token maxing" culture mirrors past tech excesses like microservices over-adoption - Predicts companies will shift from unlimited token spending to token efficiency optimization - Warns against AI influencer advice promoting maximum token usage without considering costs - Bottom line: Token efficiency, not token maximization, will become the competitive advantage ## 📖 Main Content The video centers on a case study where Pete, an OpenAI cloud researcher, spent $1.3 million in tokens over 30 days (603 billion tokens) on a research project. The speaker argues this represents a fundamental misunderstanding about the future of AI adoption in enterprise settings. **Historical Parallel**: The speaker draws parallels to the 2016-2020 era when startups over-adopted microservices and Kubernetes, creating unnecessary complexity. One example cited: a startup managing 10 microservices for just 3 customers, prioritizing architectural sophistication over business value. **Current AI Hype Cycle**: AI influencers (described as "crypto NFT bros moved over to AI") promote unlimited token spending as necessary for success, despite most users paying full price while researchers like Pete receive free tokens for experimental purposes. **Corporate Contradiction**: Companies previously required VP approval for $400 RAM upgrades but now encourage unlimited AI token spending. This represents a temporary anomaly driven by AI adoption pressure and FOMO. **The Coming Correction**: The speaker predicts a shift toward "token efficiency" as companies recognize the unsustainable economics of unlimited AI usage. This will favor engineers who optimize AI usage rather than those who simply maximize it. ### 🗒️ Notes #### ⏱️ [00:00:00](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=0) Token Spend Revelation - $1.3 million in 30 days, 603 billion tokens - Speaker questions physical possibility of such usage - Triggers broader reflection on AI economics #### ⏱️ [00:02:15](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=135) AI Influencer Criticism - Compares AI influencers to crypto promoters - Warns against "buy my course" mentality around token usage - Emphasizes different contexts: research vs. commercial use #### ⏱️ [00:04:30](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=270) Microservices Parallel - 2016-2020: Friend managed 10 microservices for 3 customers - Pattern: Adopting Google-scale solutions for non-Google problems - Same dynamic happening with AI token usage #### ⏱️ [00:07:45](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=465) Corporate Spending Paradox - Historical: VP approval needed for $400 RAM upgrade - Current: Unlimited AI spending encouraged - $1.3M monthly ≈ 30 engineers at $50K/month each #### ⏱️ [00:10:20](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=620) Future Prediction - Token efficiency will replace token maximization - Consultant class will emerge for AI optimization - Buy vs. Build vs. "Vibe" decision framework #### ⏱️ [00:12:00](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=720) Long-term Perspective - OpenAI's vision: Token utopia with infinite usage - Current reality: Limited energy and compute resources - 10x cheaper promise from 2 years ago hasn't materialized ## 📚 References - [OpenAI Cloud Platform](https://openai.com) - [Cursor Cloud Agents](https://cursor.com/agents) ## 🗣️ Paraphrase A developer discovers an extreme case of AI token spending ($1.3 million monthly) and uses it to critique the current "more tokens = better results" mentality in AI adoption. Drawing parallels to past tech bubbles like microservices over-adoption, they argue that companies will inevitably shift from unlimited AI spending to optimizing for token efficiency. The prediction is that sustainable AI practices, not maximum usage, will determine long-term success. ## 📰 Newsletter 🤖 **The Token Economics Reality Check** This week's deep dive explores why the "spend all the tokens" mentality might be setting teams up for failure. A fascinating case study shows $1.3M monthly in AI costs - but here's the kicker: it's research-funded, not real-world sustainable. Remember the microservices craze of 2018? Companies were running 10 services for 3 customers. Sound familiar? We're seeing the same pattern with AI adoption. **What This Means for WordPress Teams:** - Start tracking your AI tool costs *now* - Focus on efficiency over usage volume - Document which AI tasks actually save time vs. money The future belongs to teams who optimize AI usage, not maximize it. Get ahead of the curve before the "token efficiency consultants" arrive! 😅 ## ✅ Action Items #### DevRel - Document AI tool ROI metrics: cost per feature delivered vs. traditional development - Create developer guidelines for cost-effective AI usage patterns - Track and publish token efficiency benchmarks across different development tasks #### Community - WordPress Performance: Establish AI usage guidelines that don't compromise site performance budgets - Training: Develop curriculum on sustainable AI practices, not just maximum adoption #### WordPress - **Core Program Team:** Evaluate AI integration costs vs. benefits before implementation - CLI: Consider token costs in any AI-powered scaffolding or generation features - Documentation: Create cost-awareness sections in AI tool integration guides ## 🧪 Hypotheses - Token efficiency will become a competitive hiring skill within 18 months - Companies currently encouraging unlimited AI usage will implement strict budgets by end of 2025 - A consulting industry focused on AI cost optimization will emerge similar to cloud cost optimization - Developers who learn prompt efficiency now will have career advantages over those who maximize usage ### 📊 Methodology - Survey current AI spending across different company sizes and track changes over 12 months - A/B test teams using unlimited AI access vs. efficiency-focused training - Analyze correlation between AI spend and actual feature delivery velocity - Interview CTOs about AI budget allocation decision-making processes ### ❓ Questions - What specific metrics should companies track for AI ROI beyond raw token usage? - How will token efficiency consulting avoid becoming another "agile coaching" consultant category? - What role will open-source AI models play in reducing enterprise token dependency? - How might companies balance innovation exploration with cost optimization? ### 🔮 Future Research - Long-term sustainability of current AI pricing models under mass adoption - Development of standardized AI efficiency metrics across different use cases - Impact of specialized AI hardware on token economics - Evolution of AI tooling toward more cost-transparent interfaces ### 💡 Implications - **Practice**: Development teams need AI cost monitoring tools integrated into workflows - **Policy**: Companies should establish AI spending governance before costs spiral - **Security**: Token usage monitoring becomes part of security audit requirements - **Education**: Computer science curricula should include AI resource optimization - **Funding**: Startups need realistic AI cost projections in their financial models ### 🧠 Learning Styles Assessment - Reading level estimate: Flesch-Kincaid Grade 8.2 - accessible to general developer audience #### 👀 Visual learners - Present: Cost comparison charts, token usage graphs, timeline of microservices adoption - Improve: Architecture diagrams showing efficient vs. inefficient AI integration patterns #### 🔊 Aural learners - Present: Video format with clear narration and examples - Improve: Podcast-style summary highlighting key economic principles #### 🗣️ Verbal learners - Present: Clear analogies (microservices bubble, corporate spending paradox) - Improve: Glossary defining token efficiency, AI ROI metrics #### 🧩 Kinesthetic learners - Present: Specific examples (10 microservices for 3 customers, $1.3M spend breakdown) - Improve: Interactive cost calculators, hands-on token optimization exercises #### 🧠 Logical learners - Present: Clear prediction framework, cost-benefit analysis structure - Improve: Decision matrices for build vs. buy vs. AI implementation #### 🧑‍🤝‍🧑 Social learners - Present: References to community experiences, shared industry challenges - Improve: Discussion prompts about team AI adoption experiences #### 🧘‍♀️ Solitary learners - Present: Individual reflection on AI usage patterns and costs - Improve: Self-assessment tools for personal AI efficiency #### 🌿 Naturalistic learners - Present: Pattern recognition between different tech adoption cycles - Improve: Ecosystem thinking about sustainable AI practices ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure, descriptive content - Improve: Alt text for any referenced screenshots of token usage #### Keyboard navigation - Present: Video can be navigated with standard controls - Improve: Transcript with proper heading navigation #### Contrast and readability - Present: High contrast video presentation - Improve: Ensure subtitle contrast meets WCAG standards #### Media and captions - Present: Video has auto-generated captions - Improve: Human-reviewed captions for technical terminology accuracy ### 🌍 DEIB Assessment #### Cultural breadth - Present: Examples focus on Silicon Valley tech culture - Improve: Include global perspectives on AI adoption costs and constraints #### Inclusive language - Present: Generally accessible technical language - Improve: Define technical terms like "microservices" and "token efficiency" more clearly #### Representation - Present: Single perspective from one developer - Improve: Include diverse viewpoints on AI economics from different roles and backgrounds ### ✅ Logical Validity #### Strong reasoning - **Historical analogy**: The microservices comparison provides strong pattern recognition for tech over-adoption cycles - **Economic logic**: The spending paradox ($400 RAM approval vs. unlimited AI tokens) effectively highlights unsustainable current practices #### Fallacies - **Hasty generalization**: Assumes all companies will follow same pattern as microservices adoption without considering different contexts - **Appeal to probability**: Prediction about token efficiency consulting treats likely outcome as certain #### Tone of voice - Conversational and engaging with humor ("crypto NFT bros"), making complex economic arguments accessible while maintaining credibility through specific examples ### 🛡️ DARVO Analysis - **Deny**: No significant denial patterns present - **Attack**: Mild criticism of AI influencers as "pull-overs from crypto days" but focused on practices rather than personal attacks - **Reverse victim/offender**: No victim/offender reversal present - Overall: Content focuses on systemic critique rather than personal blame ## 🎯 Conclusion The video presents a compelling argument that current AI token maximization culture is unsustainable and mirrors past tech bubbles. The prediction that companies will shift toward token efficiency optimization appears well-reasoned based on historical patterns and economic realities. For developers and organizations, the practical takeaway is to start optimizing AI usage for value rather than volume before market correction forces the issue. ## 🔄 Refactor Appendix — Everyone is Wrong about Tokens ### 📚 Literature Atoms #### Token efficiency will replace token maximization as competitive advantage > [!quote] ID:rw_video_00_10_20 > Token efficiency as an entire argument as opposed to simply token maxing **Context & use**: - Prediction made around 10:20 timestamp - Core thesis of entire economic argument about AI adoption sustainability **Tags**: #ai-economics #prediction **Highlight**: [view highlight](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=620)  #### AI adoption mirrors microservices over-adoption pattern from 2016-2020 > [!quote] ID:rw_video_00_04_30 > I actually had a friend lament to me that he was managing 10 different microservices and he had three customers **Context & use**: - Historical parallel used around 4:30 to explain current AI hype cycle - Demonstrates pattern of adopting enterprise solutions for non-enterprise problems **Tags**: #tech-cycles #microservices #ai-adoption **Highlight**: [view highlight](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=270)  #### Corporate spending paradox reveals AI adoption as temporary anomaly > [!quote] ID:rw_video_00_07_45 > You wanted a new computer. Oh, you want 32 GB of RAM? Well, we're going to have to get a vice president to sign off on those $400 **Context & use**: - Contrasts historical cost controls with current unlimited AI spending - Evidence for unsustainable nature of current practices **Tags**: #corporate-spending #ai-economics **Highlight**: [view highlight](https://www.youtube.com/watch?v=0zw-Uk9KJiA?t=465)  ### 🧠 Permanent Candidates #### Tech adoption cycles follow predictable patterns of excess followed by optimization **Claim**: Enterprise technology adoption consistently follows a pattern where early excess adoption gets corrected toward efficiency optimization. **Evidence pointer**: [[#Token efficiency will replace token maximization as competitive advantage]] **Scope & conditions**: Applies when new technologies enable previously impossible capabilities but have clear cost structures that eventually hit budget constraints. **Links**: - [[Historical tech cycles]] — because extends pattern recognition - [[Economic sustainability]] — because evidence-for in enterprise contexts  ### ✏️ Fleeting Queue - Research actual token cost trajectories - has the promised "10x cheaper every year" materialized? #fleeting #research - Investigate emergence of AI cost optimization tools and consulting services #fleeting #market-research ### → Things 3 - [ ] Track personal AI tool costs for 30 days to establish baseline efficiency metrics #things - [ ] Research existing token efficiency measurement tools and frameworks #things --- ## Full Transcript [00:00] You see this right here? Yeah. That's [00:02] $1.3 million [00:05] spent in OpenAI tokens in the last 30 [00:08] days. 603 [00:11] billion tokens spent. Now, even if I [00:13] were to try my hardest, I am not [00:15] actually sure it's possible for me to [00:18] spend this amount of money or that [00:20] amount of tokens. I have no idea how we [00:23] accomplished such things. And when I saw [00:25] this, I thought this is just the most [00:27] ridiculous thing and I [00:29] This is so stupid. But then, I started [00:31] thinking about it more and more and I [00:33] realized that there's a future that is [00:36] developing in which I think a lot of [00:37] people are wrong and I think this post [00:40] right here really helps it kind of [00:42] crystallize in my mind [00:45] where things are going. So, I got a lot [00:48] of yapping to do, so I hope you're going [00:49] to you know, strap down cuz I I I think [00:51] that yeah yeah you you you you you [00:53] probably aren't going to see this one [00:55] coming. I don't think you you understand [00:56] what's going to happen here in the next [00:58] year. And I think I might be right on [00:59] this one. I'm going to In fact, I'm [01:00] going to do something I normally don't [01:01] do. I'm going to make a tech prediction. [01:03] I know. [01:05] Kind of dangerous. I I do just got to [01:07] yap about this for a second, okay? The [01:09] reason why I have to yap about this is [01:11] that whenever a post like this happens, [01:13] there's always the exact same thing that [01:15] happens. There's this entire [01:17] fluencer market when it comes to AI and [01:20] I largely think they're just simply [01:21] pull-overs from the crypto days. The [01:23] crypto NFT bros moved over to AI. When [01:26] they see someone make a post like this, [01:27] Papa Pete, of course, they go, "Oh, hey [01:30] bros. Hey bros. Everybody. Uh I don't [01:31] know if you know this. If you aren't [01:33] spending like $100,000 if you A- if [01:36] you're not even hitting 10 billion, if [01:38] you're not even in the B's when it comes [01:40] to token usage a month, you're not going [01:42] to make it. You know how I know that? [01:43] Look at Papa Pete, okay? Open cloud guy, [01:46] he knows what he's doing. Do you know [01:47] what you're doing? Not going to make it. [01:49] Permanent underclass. Hey, buy my course [01:51] and I'm going to teach you how to do AI [01:53] properly." And it's such a bad takeaway. [01:56] And let me explain it in more simple [01:57] terms. Like, you know, the funny thing [01:58] about history and about tech, they don't [02:01] repeat but they do rhyme. I've heard [02:02] that once and it makes me feel like I'm [02:04] really smart saying that. [02:07] You know what I mean? So, what I mean by [02:09] they they rhyme is that in 2016, 2018, [02:13] 2020, if you would see any startup, you [02:16] if you went and talked to any of your [02:17] friends in the Silicon Valley, there was [02:19] an entire culture [02:21] that had more microservices and [02:24] Kubernetes usage than they did literal [02:27] customers. I actually had a friend [02:29] lament to me that he was managing 10 [02:32] different microservices and he had three [02:34] customers. Unironic, that's not me [02:36] making up or exaggerating things. He had [02:38] triple the services for three customers [02:41] and he was just like, "What the hell [02:43] have I done with my life?" And it's just [02:45] like, "Brother, you have to quit [02:46] listening to Google for how to run a [02:48] company. Just because it works for them [02:51] does not mean it works for you." And [02:53] this is kind of that same vibe. Just [02:55] because this works for Pete, which by [02:57] the way, guess how many dollars he paid [02:59] for those tokens? Yeah, zero. You know [03:02] how much money you're going to pay for [03:03] those tokens? Yeah, full price, okay, [03:05] buddy. It's not going to be cheap, okay? [03:06] You're not spending six six hundred [03:08] three billion dollars in tokens per [03:10] month. And if you if you are, I mean, [03:12] well, hey, nice to meet you, sir. I did [03:14] not realize you [03:15] I was not aware of your game. And so, I [03:17] just wanted to kind of get that out of [03:18] the way, okay? Now to the future, the [03:20] thing that I think all of you have [03:22] wrong, okay? But first, the bag. You see [03:25] these people walking around with their [03:26] laptops cracked just so their agents [03:29] don't stop running? [03:30] Mine never stop running. When making [03:32] changes with Cloud Agents, you can see [03:33] the diffs inline just like with any [03:35] other agent. It will create a PR and you [03:38] can actually see your CI running live [03:40] within the Cloud Agent. You can see the [03:41] status of the CI when it completes and [03:43] you can even go back and fix the failing [03:45] CI. Not only that, but you can also just [03:47] run live commands in the terminal. That [03:49] is my project right there. This is not [03:52] on my computer. This is in the cloud [03:54] running where I can ask it to do things. [03:57] I can ask for changes. I can ask for [03:59] changes on my phone and see the game [04:01] played via MP4. What's even crazier is I [04:05] can just take over the desktop and I can [04:06] place towers and I can just play the [04:08] game. Start round. And I can watch the [04:11] bats happen. This is my game. Try Cloud [04:14] Agents today. cursor.com/agents [04:17] and never have to worry about your [04:19] laptop being open again. [04:21] Okay, welcome back. Let's talk about the [04:22] future here for a second. So, something [04:24] that you need to kind of keep in mind [04:26] when you see these things is that what [04:28] Pete's entire goal is, it's a research [04:31] project. How far can OpenAI take token [04:34] usage? Cuz remember, they believe the [04:36] future is going to be this token Utopia [04:39] where everybody just sits back and [04:42] relaxes like we're in Wall-E and we just [04:44] are able to out anything and you [04:46] have billions upon billions of tokens [04:48] for free cuz everything gets 10x cheaper [04:50] every single year, which by the way, [04:52] that promise is 2 years old and I feel [04:54] like things have never been more [04:55] expensive. I don't know. It feels that [04:58] way to me. Maybe I'm wrong, but things [04:59] kind of feel a little costly. [05:01] Nonetheless, 10x cheaper. Remember that. [05:03] 10x cheaper every single year. And so, [05:04] at some distant point in the future, you [05:07] spending 603 billion tokens and every [05:10] last person on Earth doing that, which [05:12] by the way, we don't even have enough. [05:13] Like I don't even think there's enough [05:14] power on Earth to do that currently. We [05:16] might have to 10x all power on Earth and [05:18] only use it to power GP used to make [05:20] this happen. But again, I digress. If [05:22] this were to come across, this is how [05:24] projects could look. So, I think a lot [05:27] of people look at this and they're like, [05:28] "Oh, well, you know, OpenAI is being [05:29] evil." No, I think they actually just [05:30] believe this, right? Like I think they [05:32] actually believe that every last person [05:34] will be using Infinity tokens at all [05:36] times. And yeah, sure. They are the [05:38] benefactors of it. And I mean, it's a [05:39] good future for them, but I actually [05:41] also think they they think this is just [05:43] like how the world should work. This is [05:44] how projects should be ran. And so, this [05:47] is a research project which got me to [05:49] think about something for a second. And [05:51] it's kind of this funny conundrum that [05:53] you see. Uh right now, if you go to any [05:56] of the big companies, what are they all [05:58] about? Hey, what's your token spend? I [06:00] mean, there is literal people getting [06:02] fired because they're not using AI [06:05] enough. You've seen this, you've seen [06:07] the articles, you've seen [06:09] potentially these rage posts on Reddit. [06:11] I can never tell if what I'm reading on [06:12] Reddit is real or not, if it's just [06:14] there to rage bait me into a frothy [06:16] mouth just to go off and tweet a story [06:18] that doesn't even exist. But let's [06:19] pretend they do exist. People are [06:21] getting fired for not using enough AI. [06:23] The I've read stories about people who [06:24] are interviewing, if they use too much [06:26] AI, people don't like it. If they don't [06:28] use enough AI, people are not liking it. [06:29] Like interviewing sounds like hell. [06:31] Working at companies right now sounds [06:33] pretty awful cuz you're constantly being [06:34] shoved down the throat, you must use [06:35] this. People at Amazon, you better use [06:37] Kiro. Hey, if you're over at Google, [06:38] better use that Gemini, buddy. And just [06:40] keeps on going and going and going, [06:42] right? Well, there's kind of a problem [06:43] there. I don't think people realize what [06:46] the problem is. Because right now it's [06:48] like spend all the money you want, [06:49] right? Okay. Well, let's just rewind [06:52] like 18 months, okay? Not even that long [06:54] ago. Let's just go back a little bit. [06:58] You wanted a new computer. [07:01] Oh, [07:02] you want 32 GB of RAM? Well, we're going [07:05] to have to get a vice president to sign [07:08] off on those $400. Oh, and a chair? [07:11] Yeah, that chair, it's going to be a [07:13] used Herman Miller. Okay. You're I'm [07:15] sorry, but those buns of yours do not [07:17] get the luxury of sitting on brand new [07:19] Herman Miller, okay? You know what? [07:21] We're getting you a lifetime chair. [07:22] That's what you get. Yeah, you. You get [07:24] a lifetime chair and I'm going to go [07:25] grab some patio furniture padding and [07:27] duct tape that right onto your chair. [07:28] That's what you get. That's what you [07:29] deserve, okay? Because let's just face [07:31] it, we can't be bothering our VPs for [07:33] these $50 upcharges. We can't do that, [07:36] okay? Us as a multi-billion dollar [07:38] company, we are very concerned if you [07:41] spend $25. And now, all of a sudden, [07:45] you can spend infinity on tokens. In [07:47] fact, you're even encouraged to do so. [07:49] Going back to this for a second, if you [07:51] really think about that, that means it [07:54] takes $1.3 million a month to run [07:57] OpenClaw. So, how many engineers is [08:00] that? Well, like if you think about [08:02] that, let's just pretend we're a big [08:03] tech Google company. It costs $50,000 a [08:07] month, and you're spending $1.3 million [08:10] a month on just AI agents. To replace [08:12] those with just engineers with Well, [08:14] that kind of math you I mean, it's a [08:16] number. That kind of math you can't just [08:17] do off the top of your head. So, let's [08:18] just say 30 engineers. That's like 30 [08:20] engineers worth of people working on [08:22] something. You can't just do this for [08:24] every single project. Your company at [08:27] some point's going to go, "Okay, [08:28] timeout. We've made a mistake. We have [08:31] decided that we let you use all the [08:33] tokens you want. That's bad. We're going [08:35] to go back to the old days. Who's the [08:37] most token efficient? Oh, you're not [08:39] token efficient. You're spending 603 [08:42] billion tokens on maintaining a simple [08:44] project? No, we're not going to do [08:46] that." [08:47] You're gone. There's going to come a [08:49] world where there's an entire consultant [08:51] class going through these companies [08:54] teaching people how to be efficient with [08:56] tokens. No longer will we see this world [08:59] of infinity token usage. Instead, it's [09:01] going to be, "Okay, who's the top [09:03] performers by features and things [09:04] delivered, not just by how much you [09:06] spend." Because in the old world, we [09:08] used to do buy versus build. Do you [09:10] build the thing or do you buy the thing? [09:12] Depending on the cost and the trade-off, [09:14] sometimes it's better to, you know, [09:15] trade the time for the money or the [09:17] money for the time. But now, we kind of [09:19] have a new world. It's like buy versus [09:21] build versus vibe. Do you vibe it? Well, [09:23] vibing takes both time and money. So, [09:26] which is the proper trade-off? And I [09:27] think companies are going to quickly [09:29] snap back to the old way in which [09:31] they've always done things. It's going [09:33] to be, "Okay, who's the most efficient? [09:35] Who knows how to use these things the [09:37] best? It's not going to be the people [09:39] spending Infinity. It's not going to be [09:40] the fluencers that's telling you [09:42] you need to run 500 agents in the cloud [09:44] at all times or you're not going to make [09:46] it. It's going to be the people that are [09:47] just being engineers. They're the people [09:50] like learning. People that actually want [09:52] to just do good work and use things to [09:54] help speed them up in certain areas. And [09:56] that's my prediction. Yes, I I'm doing a [09:59] prediction. I'm doing an actual [10:00] prediction. I know you're not supposed [10:02] to do predictions. Tech predictions [10:03] almost are largely you're always wrong, [10:06] but I do think in the near future we are [10:09] going to see token efficiency as an [10:11] entire argument as opposed to simply [10:13] token maxing. Token maxing is because [10:16] we're just trying to figure out is this [10:17] even viable? And by we, I don't mean me. [10:20] I'm out here still hand coding stuff for [10:23] my video game, okay? This is it's a [10:25] different world. But nonetheless, this [10:26] was very interesting to see. I was very [10:28] happy I got to read about this and kind [10:30] of see the live reaction from everybody [10:33] because people were just, you know, [10:35] instantaneously suspicious. Like, "Oh, [10:37] this is just open code trying to make [10:38] money." Yeah. [10:40] They are they are trying to make money. [10:41] I'll tell you that much. But they also [10:42] this is just like what they think the [10:44] future looks like. You and 100 agents [10:46] non-stop doing stuff. And maybe at some [10:49] point in the future, maybe hey, you know [10:50] what? Maybe in 10 years, some large [10:53] amount of time when we have, you know, [10:55] 100x more energy and 1,000x more GPUs. [10:58] Yeah, maybe that future does exist in in [11:00] some far away place. But right now, to [11:02] me at least, the big takeaway here is [11:05] I think you got to start thinking about [11:06] token efficiency. You got to start [11:08] thinking about how you're actually using [11:09] it. Maybe having a kajillion agents does [11:12] work for one person. [11:13] But I'm not sure if this is really a [11:15] sustainable approach for anybody, even [11:18] if the promised 10x is going to happen. [11:20] Okay, sorry. I made a future prediction [11:22] and I'm probably going to be wrong, but [11:23] I you know, honestly, I think I'm right. [11:25] Also, the consulting class, can we all [11:27] just agree that's going to be the most [11:28] annoying people in the universe? [11:29] Honestly, [11:31] I'd almost rather take the crypto bros [11:34] who are going to be like, "Oh, you got a [11:35] token max." than [11:37] the new class of agile coaches that are [11:39] going to be coming out. These agile [11:40] coaches for token efficiency is just [11:43] going to be the worst. Oh my gosh. There [11:45] is actually going to be prompt trainers. [11:47] Like [11:47] it's going to be like Pokémon trainers, [11:49] but they're going to be prompt trainers [11:50] and you're going to have to go in there [11:52] and they're going to like one-v-one you [11:53] on prompts. It's going to be so [11:55] ridiculous. It's all horoscopes, baby. [11:57] The name [11:59] is the Brian Magen.  📄 **Full text:** ![[Full Document Contents/Articles/Everyone Is Wrong About Tokens.md|Everyone Is Wrong About Tokens by YouTube]] ## Highlights - customers. I actually had a friend lament to me that he was managing 10 ([View Highlight](https://read.readwise.io/read/01ksfsxbdkngks68pepjv86tg8)) - Note: AI adoption mirrors microservices over-adoption (10 services, 3 customers). - Tags: ai-apis, atom/literature/h2 - you want 32 GB of RAM? Well, we're going to have to get a vice president to sign off on those $400. Oh, and a chair? ([View Highlight](https://read.readwise.io/read/01ksfsxdz2bx4vc07kex5s3pa5)) - Note: Corporate paradox: VP sign-off for $400 RAM, but unlimited AI spend. - Tags: atom/literature/h2, ai-apis - going to see token efficiency as an entire argument as opposed to simply token maxing. Token maxing is because we're just trying to figure out is this ([View Highlight](https://read.readwise.io/read/01ksfsx9k2wevz9gzba08v04px)) - Note: Token efficiency replaces token-maxing as the competitive argument. - Tags: atom/literature/h1, ai-apis