@jakevanclief

There's a massive AI problem for companies that sell advice. Firms like McKinsey and Deloitte bill for hours and for what they hand over at the end: the research, the strategy, the slide decks, the training. AI cuts the hours it takes to make all of it, and the templates a new MBA follows can go straight into an AI. People will still want the deck, and plenty will still pay someone to make it. The squeeze is headcount, because the product no longer pays for that many people. More on AI and how work is changing on my YouTube. Link in my bio. youtube.com/@JEVanClief #shorts #ai #consulting #mckinsey

Advice companies are like Mckinsey, these companies that come in and are basically giving strategy or they're saying, hey, we just went through and collected all this data. We have data on the past 30 years for your type of company. Here's what you want to do. Here's your business model. Here's slide decks. We'll do some enablement training. We'll collect some data. We'll we'll work with your people. Here's our advice to make your company successful. Now a lot of this is billable hours and training hours. So they are making money for the artifacts that they're giving at the end as well as the research and hours that they are putting in to create those artifacts. AI means that you can take less hours to get the same amount of research, which means less billable time. The biggest part of the advice sector is also IP Mckinsey, Deloitte all these people have a huge database and history of knowledge of how to run a company. The problem is some of this is hard to read hard to understand. They're passing it off to an MBA who just got hired after getting out of college and they're asking seniors for advice. They're playing with a template. Those templates can be uploaded to an AI and I can follow them tomorrow. So that brings billable hours and search down a lot. So the problem here with advice consultancies is they can make a lot of money by delivering artifacts advice Slide decks, all of these things. But if that is faster to delivery and if that cost is going down, the problem is not the artifact. People still need that. Just because the AI could do it doesn't mean they won't want to pay someone else to do it. The problem is the massive amount of employees that they have to pay for, but the product does not support that price. And you'll see this is a common theme among a lot of them.

Upgrading your Claude plan to stop hitting the limit mostly buys it more room to reread your PDF. Every message carries the whole chat back in, so the PDF you pasted at nine gets read again with every message after it. Give each task its own fresh chat, and have the AI boil big documents down to one page you keep in a project, with the PDF left out. Then send each job to the cheapest model that can do it, because a big model answering yes or no is a classifier in a very expensive coat. Clief Notes is free, link in my bio. https://www.skool.com/cliefnotes/about #shorts #claude #ai #contextengineering #productivity

Upgrading your Claude plan to stop hitting the limit mostly buys it more room to reread your PDF. I'll show you the fix right here. And honestly, it's boring, which is exactly why nobody posts about it. Every message you send carries the whole chat back in with it. So that PDF you pasted at 9 gets read again with every message after it. And by lunch, you're out. And anthropics own docs say the longer that pile gets, the worse it remembers what's in it. So on a bigger plan, you're paying more for worse answers. So give every task its own fresh chat and keep what matters in files. Have the AI boil any big document down to one page once put that page in your Claude project and keep the PDF out. And in a project, that page gets cached so it counts for less every time. Need an exact number from the PDF, open it in a chat of its own. Then for whatever you hand your AI, write down what it needs and what to leave alone. And that's what people mean by context engineering. And send each job to the cheapest thing that can do it. Because a big model answering yes or no is a classifier in a very expensive coat. Now you might be thinking, I'm paying for the big plan, so why not use it? Payne from my community said it probably took a week before the routing improvements and context Engineering paid for themselves via saved compute, that community's cliff notes. It's free link in my bio.

Your AI is quietly building a folder about you. Every time you chat with Claude about a client or an investor, it saves what it learns about you, and the next new chat pulls those notes back in before it answers your first message. The hard part is keeping that pile organized, so the right notes come in at the right time. Open your Claude settings and find the section called memory to see what it's kept on you. The full video, Do You Actually Need AI Agents?, is on my YouTube. Link in my bio. youtube.com/@JEVanClief #shorts #ai #claude #aimemory

A lot of these tools are making a folder on you in your profile. So when you're sitting there and chatting with Claude, let's say you uploaded something you're working on, you chatted with it on some sort of investor or maybe some sort of client, as you're having those chats, it's saying, oh, that's information about you. Oh, that's information about your investor, and it's adding it to its own folder structure. The problem, and this is something anthropics working on too, it's why they're in the space as well, they've made skills, all of these things, is this needs to be organized because the next time you talk to Claude, when it wants to bring this information in, it needs to bring it in every time. So every time you open a new chat, that model knows nothing about you. As soon as you send that first message, anthropic is saying, okay, make sure that information is already in there. Depending on your question, it's gonna go, oh, I think the best place to look for that would be where you said investors inside of the folder. It's literally doing that. So if you open up your Claude account, there is literally a section that says memory.

Somebody at a family dinner asks what you want to do with your life, you give them a list, and everyone else has one word. Here's what your list looks like from above, and trust me, I've got a long list myself. Knowledge works like the map in a strategy game. The middle is what people already know, and the edges are fog, which is where the new ideas are. A specialist clears one line all the way to the edge and gets to see the walls of the box. If you hyperfixate on everything, you clear patches all over, which looks scattered from the middle and like more of the map than almost anyone from above. So take one patch all the way to the edge and finish something. Join Clief Notes, free, link in my bio. The long version is on my YouTube: The Hidden Advantage of Hyperfixating on Everything. #shorts #learning #polymath #hyperfocus #creativity

Somebody at a family dinner asks what you want to do with your life. You give them a list and you watch their face do the thing because everyone else has one word, nurse, lawyer, accountant. I'm going to show you what that list looks like from above. And trust me, I've got a long list myself. So picture Civ 6, a strategy game I've lost way more nights to than I'll admit in public, where the whole map starts hidden and everything you haven't explored is fog. That map is the best picture I've found of how knowledge works. The middle is everything people already know written down and lit up. And out at the edges, it's fog, which is where every new idea comes from. A specialist picks one direction and clears it for years all the way to the edge. And after that long inside one box, they and finally see its walls. Then there's you. Probably you hyperfixate for three months, clear a big patch, then something else grabs you and you go clear a patch somewhere completely different from the middle of the map that looks scattered. And from above, it looks like someone who's seen more of the map than almost anyone. So take one of those patches all the way to the edge and finish something because once you've seen the walls, you'll recognize the edge everywhere. Nobody in Cleef notes has the same map, which is my favorite thing about it. Links In my bio. And the next time.

Every video I make, this one included, comes out of one folder on my computer, and you can build the same kind of AI second brain for your business tonight. Put a short CLAUDE.md at the top that says what the place is and where to go for what. Give every job you repeat its own folder, with a note on what goes in and who checks what comes out. Anything every job shares lives in one system folder, written like the note you'd leave whoever covers your vacation. Skip MCP, skills and hooks for the first week, they bolt onto a folder like this later anyway. Clief Notes is free and has 40+ lessons on exactly this. Join Clief Notes, link in my bio. https://www.skool.com/cliefnotes/about #shorts #AI #ClaudeCode #SecondBrain #SmallBusiness

Every video I make, including this one, comes out of one folder on my computer. And you can build the same kind of AI second brain for your business. Tonight I'll walk you through the whole thing right here, file names and all. And honestly, it's way less work than the apps want you to think. At the top you put one short file called Claude dot MD saying what this place is and where to go for what because Claude code reads that file first when you open it in that folder. Mine literally says new video, start in the voice folder. Then make a folder for every job you repeat, like client emails or invoices, each with a short note saying what goes in and who checks what comes out. So the AI only opens what that one job needs. Anything every job shares goes in one system folder, like how I talk and how I post. So you only write it once and you write all of it like the note you'd leave whoever covers your vacation. The odd little rules included. Mine says no big laughs on my cloned voice because they sound off. That's the stuff. It would never work out on its own. Now somebody's already typing. That's way too simple to work. This one folder has made dozens of my videos and a lot of the setup advice online comes from people whose favorite part is the setup itself, which, hey, great hobby Skip MCP skills and hooks for the first week. They all bolt onto a folder like this later. Anyway, Clief notes is free and it's got 40+ lessons on exactly this link in my bio. And that one folder is the whole reason.

Mr jobs

Mr jobs, you're a bright and influential man. Here it comes. It's sad and clear that on several accounts you've discussed, you don't know what you're talking about. I would like, for example, for you to express in clear terms how, say, Java in any of its incarnations addresses the ideas embodied in Open Dock. And when you're finished with that, perhaps you could tell us what you personally have been.

Let AI automate everything it wants. Seriously, let it. We're still never running out of work. I watched it happen in a room in Dubai. Every hour the AI handed back, the team already had something waiting for it. One person got seven hours back on every customer map and spent them on a project that was six months away. Douglas Engelbart, the guy who invented the mouse, was making this case in the sixties, and every layer of computing since has handed us a tool and sent us off to find a bigger problem. So let AI automate the task, hand that automation to a person, and watch what they do with the time. Join Clief Notes, link in my bio. https://www.skool.com/cliefnotes/about #shorts #ai #futureofwork #automation #jobs

Let AI automate everything it wants. Seriously, let it. We're still never running out of work. I watched it happen in a room in Dubai. A company paid me 120 grand to train their people. And every hour the AI handed back, they already had something waiting for it. And trust me, they had a list. I'm going to show you why that keeps happening all the way back to the guy who invented the mouse right here. One person got seven hours back on every customer map and spent them on a project that was six months away. Another stopped paying a vendor for training decks and started making them in house, which is money that used to walk out the door every single year. Now you've seen the headline, most desk work automated. In 18 months, Klarnas said its AI was doing the work of 700 people and a year later, it was hiring people again. And this argument is old. Douglas Engelbart, the guy who invented the mouse, was making the same case in the sixties. The mouse and the keyboard are automations you can hold. Every click runs thousands of instructions you never think about. And all that ever did was move us up a floor so we could take on bigger problems. Excelle was supposed to take the numbers out of a bookkeeper's hands. And 30 years later, bookkeepers are building models and getting paid more than before. AI is just the next layer. And Every layer of computing does the same thing. It hands you a tool and you go find a bigger problem, because people never stop wanting things. And wanting things is where all the work comes from. So let AI automate the task, hand that automation to a person and watch what they do with the time. Which is exactly why I keep saying.

A company paid me $120K to teach their people that agents are just a naming convention. The value comes from everything else, the instructions you write, what the AI can reach, and the outcome you want. An agent is three parts. The instructions, the model (the part you rent), and what it's allowed to touch. A good harness like Claude Code already handles the looping and the tool calls, so for most people the agent problem is solved. Write the instructions down, put them in a folder, and one agent turns into every agent you need. No keyword to comment, the whole setup is in the video. Join Clief Notes, link in my bio. https://www.skool.com/cliefnotes/about #shorts #ai #aiagents #claudecode #automation

Company paid me 120 grand to teach their people that agents are just a naming convention and the value comes from everything else. The instructions you write, what the AI can reach and the outcome you actually want. I'm going to show you the whole setup right here and you don't have to comment the word agent to get it. Okay. This drives me nuts in computing, everything good got built on top of something that already worked. And right now everybody's out here building their own AI agent from scratch. Like everybody, that's a brand new wheel and the wheel is done. An agent is really three parts. The instructions, so what you want done and how you do it. The model, which is the part you rent, Claud or chat GBT or co pilot. I genuinely don't care which and what it's allowed to reach. Your files and your tools, the looping and the tool calls. A good harness like Claude Code already handles that. So for most people, the agent problem is solved. One good model, a good harness and a good set of folders. I watched this click in a room in Dubai. One person wrote a brief with Claude. Claude couldn't reach the government website it needed. So somebody on a completely different team ran the exact same brief in chat GPT and it worked. Same instructions in a different AI and it still worked. The part worth owning was the instructions the whole time Now, if you've built an agent, I'd honestly love to see it and an automation people can talk to is a great thing to have. The name just gets in the way of the part that matters. And Gardner's saying up to 70% of companies will walk away from the agents vendors built for them by 2028. So write down the outcome you want, how you do it, what it's allowed to touch and where a person checks it. Put that in a folder, point one good agent at it, and that one agent turns into every agent you need, which is exactly why.

I went from $5,000 a month to $62,270 a month on ONE product, in seven months. These are the three things I think made it possible. 1. Amplify what you're already good at. Not sure what that is? Think about the last person who paid you, and why they paid you. 2. Post. Social media is the most important one, because nobody knows you exist until you do. 3. Have a counter-narrative, a take that pushes against what everybody keeps repeating, with the how right behind it. No keyword to comment, it's all in the video. Join Clief Notes, link in my bio. https://www.skool.com/cliefnotes/about #shorts #ai #entrepreneur #creator #business

You wanna make five grand a month or maybe 60 grand a month? Well, I've done both. I had one subscription service go from 5,000 a month to 60,000 a month in seven months. And I'm telling you, if you don't understand these three things, you will never be able to grow no matter what you're selling. And you don't need to comment anything or pay me. I'm explaining it right here. Number one, you amplify what you're already good at. A lot of people go learn AI so they can make money and they've got it backwards. AI amplifies you, same as every tool anybody's ever built. So if you don't know where to start, think about the last person who paid you a job, a contract, whatever, and ask why they paid you. For me, it was a college paying me to teach their professors, which was wild because those were the same professors who taught me. Then you've basically got two options. Teach other people how to do that thing or teach the version of you from a year ago how to do it better and skip all the dumb mistakes you made. And trust me, I made plenty. Get somebody there in a few months instead of a few years and people will pay for that. Now, No. 2 is the one almost everybody skips, and it's the most important, social media. Nobody knows you exist. You can be the best in the world at something. And if you never post, nobody is ever going to find out. And I know nobody wants to hear. Just post more. You don't have to dance. I promise. People found me because I posted videos, which case in point, you're watching one. I get millions of views a month. I've got over 50,000 students in my community, and 37,000 of them joined in under 90 days without a single ad. And some of my biggest contracts came from CTO's scrolling through YouTube and finding my stuff. So post videos that answer the questions people are already asking and show them what's possible. No. 3, you need a counter narrative. And it has to tie back to the first two. That's your take that pushes against the story everybody keeps repeating in the thing you're actually good at. Everybody was saying, go build AI agents. And I said, agents are just a naming convention. The value comes from everything else. And yeah, some people get pretty mad at me in the comments for that one. Everybody's saying, AI is taking all the jobs. And I'm saying, let it automate everything it wants. We're never running out of work. That's what makes people remember you out of the 10,000 other people posting the same thing. It only works if they walk away understanding what actually works, though. So every take needs the how right behind it. And somebody's already typing. So that's the only way he makes money. Nope. And honestly, Fair question. This year, a company in Dubai paid me 120 grand for three weeks of training their people. The community is one product out of a few, and it grew on these exact three things. So amplify what you're good at post, so people can actually find you and give them a take that teaches them what works. Get those right, and it plays out the same way. Weather.

It still says running, and the report's been wrong since Tuesday. Split the job where you'd naturally pause, one folder or one sheet tab per step, and have each step save what it made to a file a person reads before the next one runs. Give every step one check you can do in about a minute. When a check fails, fix that one file and rerun from there, so you catch it the day it broke. Buying an AI setup? Ask where a person checks it. And where do the instructions live? Can you open them? Clief Notes is free, with 40+ free lessons. Link in my bio. #shorts #ai #aiautomation #automation #workflow

It still says running and the report's been wrong since Tuesday because that morning the step that grabs the numbers started pulling last week's sheet and every step after it built a perfectly nice report on the wrong numbers. And the light stayed green the whole time since all green actually means is nothing crashed. So open yours today, find the spots where you'd naturally pause and split it there. One folder per step or a sheet tab per step works too. And each step saves what it made to a file a person reads before the next one runs, which is actually rule 1 in the folder that makes these videos. Then give every step one check you can do in about a minute. Like does the numbers file have this week's dates on it? Things like that. And if checking a step takes longer than doing it yourself, that steps too big. So split it again. When a check fails, you fix that one file and rerun from there so you catch it on Tuesday, the day it broke. And if someone's selling you an AI setup, whatever they call it, ask where a person checks it. And if the answer is the very end, that's a Tuesday waiting to happen. One Cliff Notes member built a proposal system with a person signing off at two gates before anything goes out and his client closed a major deal off one of those proposals. There are a lot more Wins like that in Cliff Notes. It's free link in my bio. And trust me, I've built some truly terrible automations myself.

Em dash, em dash, little blue diamonds, hashtag future of work... and everybody reading it can tell you didn't write it. Most AI writing sounds like that because nobody gave the model their voice. Put a plain voice file, with a few rules and three or four things you actually wrote, in your project instructions so Claude or ChatGPT reads it every time. Any rule with one right answer goes in a tiny search script that checks every draft, and you still read the draft yourself, because a search only catches the obvious ones. Setting this up is the kind of thing we get into in Clief Notes. It's free, link in my bio. My own videos run on the same idea, and I walk through it on my YouTube in "How to Make AI Videos That Don't Feel Like AI Slop". #shorts #ai #chatgpt #claudeai #writing

M dash, m dash, little blue diamonds, hashtag future of work, and everybody reading it can tell you didn't write it. That's Claude's real answer when I asked for a LinkedIn post about AI at work. And most AI writing sounds like this because nobody gave the model their voice. So you get its default voice, pretty much the same one everybody else is posting. So first, all you need is a voice file, paste it into your project instructions in Claude or chat. GPT so it gets read every single time. And mine's a few plain rules like no em dashes, use a period. Then add 3 or four things you actually wrote because the model copies an example way better than an adjective like punchy, bad habits and all. Now any rule with one right answer, like the dashes, a script can check and the judgment calls like does this sound like me, stay in the voice file. So you're going to have Claude turn those rules into a little search script, basically control f for every rule at once. And mine's 19 lines, one search for dashes and a pretty rough one for those flip lines AI loves, like that Post's headline. Have Claude run it on every draft. And on that post, it flagged every dash plus 1 false alarm on a Blue Diamond. And the flip line search caught just one and missed that headline completely. So read the flags, fix it, or keep it on Purpose and still read the thing yourself because a search only catches the obvious ones. Setting this up is the kind of thing we get into in Cliff Notes. It's free and the links in my bio and my own videos run on the same idea, which I walk through on my YouTube in how to make AI videos that don't feel like AI slop. Oh, and I ran the script on that title. It came back clean.

You explained your whole project to Claude yesterday, and this morning it's basically a blank page again. Every chat starts with a fresh context window, so save the paragraph you keep pasting as a file called CLAUDE.md at the top of your project folder, and Claude Code reads it at the start of every session. In the ChatGPT or Claude apps, it goes into a project's instructions once. Keep it to your own rules, because researchers at ETH Zurich found these files made each run cost over 20% more and didn't raise success on average, though the rules in them still got followed. If you've explained it twice, it belongs in a file. I teach how to write one in Clief Notes, my free community. Join Clief Notes, link in my bio. #shorts #claudecode #ai #aitools #productivity

You explained your whole project to Claude yesterday and this morning it's basically a blank page again because every chat starts with a fresh context window and Claude Code's own docs say it right up top. So you paste the same paragraph back in day after day. Every morning's a first date with your laptop. All you need to do is save that paragraph as a file called Claude dot MD at the top of your project folder. And Codex reads its own version of the same file. It's actually read at the start of every session and Codex even rereads it after a long chat, gets squashed into a summary. And in the chat GPT or Codex apps, you paste it into a project's instructions once. Mine's about a page with my rules at the bottom. So when you correct it on the same thing twice, that correction goes in the file with the date so later you can clear out the old ones. And anything you learn the hard way goes into like when my numbers said opening these shorts with, hey, you cost me viewers. That's one dated line in my shorts file now. And keep it to what only you know because researchers at ETH Zurich tested these files on real coding tasks and on average they didn't get more tasks solved and made each run cost over 20% more, though the rules in them still got followed pretty well. So write down your rules and Let it read the code for the rest. So memory is a file and even cloud codes own memory feature is basically a folder of notes. Personally, I don't trust any AI memory. I can't open and read. If you've explained it twice, it belongs in a file. I teach how to write this file in cleef notes, my free community link in my bio. So you and your laptop can finally get past the first date.

An AI can draft a legal brief in ten seconds and still can't fix your printer. It learned mostly from what we wrote down, and the part of your job nobody wrote down is the part you're actually paid for. So hand AI the clicks first, and write your take on the decisions somewhere your AI reads every time, like a project's instructions. That file stays yours, and you're still the one who spots when the AI gets it wrong. I made a whole video on where you fit, called "What makes me valuable now? (Will AI replace everyone?)", and it's up on my YouTube: https://www.youtube.com/watch?v=ftojWLZ5VlE Join Clief Notes, my free Skool community with 40+ free lessons. Link in my bio. #shorts #ai #claudeai #chatgpt #careeradvice

An ai can draft a legal brief in ten seconds and still can't fix your printer. It learned mostly from what we wrote down but nobody wrote down the little tug your printer needs when it jams or that a quick note from your pickiest client means start over if you're worried about your job. That gap is good news because the part nobody wrote down is the part you're actually paid for. Spreadsheets ate the written down part of accounting and from nineteen eighty to twenty fifteen, about four hundred thousand bookkeeping and accounting clerk jobs went away and about six hundred thousand accounting jobs got added. So the work moved up a floor. So sort one normal week into clicks, anything you could write steps for and decisions, anything where your answer starts with, well it depends and hand ai the clicks first. Now past entry level, the best way to do the work is usually somebody's opinion. Ask ten senior people how to answer that picky client and you'll get ten answers that all work. So ask the ai cold and it can't guess which one you are and what you get back sounds like nobody in particular and no prompt pack fixes that and anybody whose whole value lives in one is racing the ai labs and they're going to lose. So for that decisions pile, write your take down your defaults, what you check first, the weird cases somewhere your ai reads every time like A project's instructions or a file like this one of mine, that file stays yours. A version of you that you can scale and it won't hold all of it. So you're still the one who spots when the ai gets it wrong. I made a whole video on where you fit. It's called what makes me valuable now, and it's up on my youtube, all my opinion, obviously, which is kind of the point.

Check, my, email. Three words, and your AI is missing three answers. How you check things goes in a short file, which is pretty much all a skill is. Your role goes in your profile settings, written once, and the email part needs a connector, or your AI can't even see your inbox. Three words is plenty once those answers live somewhere it can read. So try this today. Paste the prompt you type most into a fresh chat and add "What would you need to know to do this well?" More on writing these down in Clief Notes, my free community with 40+ lessons. Link in my bio. #shorts #ai #claude #chatgpt #prompting

Check my email, 3 words and your AI is missing 3 answers. Check for what? Your boss, anything due today. That answer goes in a short file on how you check things, which is pretty much all a skill is. And this is my version of one. The file Claude checks my scripts against. Then the word my, that's your role and what you own. So write it once in your profile settings, the box where you tell Claude or ChatGPT about yourself. And every new chat basically starts out knowing your role and email is access. So it needs a connector, the switch that lets it into your email. And without one, it can't even see your inbox. Say those same three words to a human assistant and they just go do it because somebody set up all three on their first day. Most people hand their AI three words and no first day. So it guesses and then they blame the prompt. And honestly, 3 words is plenty as long as the answers live somewhere it can read. So try this today. Paste the prompt you type most into a fresh chat and add the line. What would you need to know to do this? Well, anything it asks that you'd answer the same way every time is something you haven't written down yet. There's more on writing these down in Cliff Notes. It's my free community with 40+ lessons link in bio. And that's three words too. But you already know what all three mean.

Sixteen bit, hand drawn, clean and modern, a blueprint. Same video, four styles, and every one came out of one folder on my computer. It's folders, a few scripts and some markdown files, with one folder per stage (voice, words, storyboard, animation, render), and I read what each stage made before the next one runs, so one bad take never turns into a bad video. Want to build these yourself? The full walkthrough is on my YouTube, and my free community, Clief Notes, is the link in my bio. #ai #claudecode #aianimation #learnai #shorts

16 bit hand drawn, clean and modern, a blueprint, same video, four different styles and every one of them came out of one folder on my computer, voice and all. And that folder can make it in pretty much any style I can describe. And it's honestly just folders, a few scripts and some markdown files. And this voice is actually my cloned voice reading a script that's sitting in that folder right now. That's it on screen. And the line lit up is the one you're hearing now. The idea behind it is almost 100 years old. At Disney from around 1930, they recorded the voices first and then timed every drawing to them. And this folder works the same way. So there's one folder per stage and each one has a short file saying what it reads, what it makes and who checks it. The voice comes first, then a free tool called whisper writes down every word with the exact time it said. Then the storyboard, which is just a text file with one beat for every time the picture changes. Then Claude writes the animation as code and then it renders the video. And after every stage I read what it made before the next one runs. So one bad take never turns into a bad video. So if you want to learn how to build these yourself, the full walkthrough is on my YouTube or hit the link in my bio for my free school Community cliff notes. And no, I'm not going to make you comment folder below to get a link that's already sitting right there in my bio.

In 1947, engineers at Harvard found a moth stuck in relay number seventy and taped it into the log book: "First actual case of bug being found." The funny word is actual. Engineers already called little faults bugs; Edison was writing about them in 1878. This time the bug just showed up in person. The part worth copying is that they wrote it down. When your AI gives you a confident wrong answer, that's your moth. Log it, turn it into a test, rerun it every time you change the prompt. Want to build checks like this? Join Clief Notes, link in my bio. #ai #computerhistory #debugging #learnai #shorts

In 1947, engineers at Harvard logged the first actual bug in a computer, a real moth. And the funniest word in that sentence is actual. Their computer filled a whole room, and this is way before chips. It ran on relays, little metal switches. A magnet clicks shut to do the math. So one afternoon, it's acting up, and at 3:45, they track it down to panel F, Relay No. 70, and there's a moth jammed in the switch so it can't click shut. They tape it into the logbook, the machine's diary, and next to it, someone writes word for word, first actual case of bug being found. Now, here's why it says actual engineers already had a name for little faults, bugs. Thomas Edison was writing about bugs in his inventions back in 1878. So the words way older than the moth. This time, the bug just showed up in person. Joke aside, the part I love is they wrote it down. At every layer of computing, from punch cards to relays to silicon, we've made mistakes. And at every layer, we've built habits to catch them right. Apps still keep a log, a running record, so someone can go find the moth. And programmers write tests, little checks where they already know the right answer. So a bug you caught once can't sneak back in. So when your AI makes up a fact or gives you a confident wrong answer, that's our moth, don't panic. Do the Harvard thing and write down every wrong answer, that's your log. Then do the programmer thing, turn each one into a check, that's your test, and rerun the checks every time you change the prompt or the model. That's pretty much how every weird new machine turned into a boring reliable one. And that page is still at the Smithsonian Moth and all because.

An AI picked the first line of this video. I wrote four openings and let a tiny model called Jev grade them: how curious would a beginner be, zero to three, and would they actually get it, yes or no. Under half a second a line, two thousandths of a cent each. The model gives opinions. A few lines of code make the call. Works for titles, thumbnails, emails, anything with more than one draft. Want to build things like this? Join Clief Notes, link in my bio. #ai #contentcreation #aitools #learnai #shorts

Here's how to get an AI to grade your ideas in half a second. Okay, confession, I didn't pick that line. I wrote four openings for this video and let a tiny model called Jev choose. I gave it a markdown file with two questions in plain English. How curious would a beginner be 0 to three? And would they actually get it? Yes or no? The weird one. This sentence just beat three others in a fight judged by a machine lost by a hair. Each line took under half a second and cost 2 of a cent. And the model only gives opinions. A few lines of code make the call. So fail the yes or no and you're out. Then the highest score wins titles, thumbnails, emails, anything with more than one draft and the line it picked to open this video.

Software is the small number. Every business on the planet spends almost a trillion and a half dollars a year on software. The paychecks in just the US come to about thirteen trillion. That's roughly nine times bigger. Software buys the part of a job you could put in a manual. A paycheck buys the judgment nobody wrote down. That's why investors are now saying stop selling the tool, sell the work. Want to learn how people are packaging their judgment? Join Clief Notes, link in my bio. #ai #business #futureofwork #learnai #shorts

All the money businesses around the world spend on software in a year, the apps, the subscriptions, the licenses, comes to almost a trillion and a half dollars. And that's the small number, which is massive, right? It's just not the big one. The big one is paychecks and just the American ones. Add up every wage and salary in the country and you get about $13 trillion a year. So the paychecks in one country are about nine times what every business on the planet spends on software. Now here's why the gap is that big. A program is just steps somebody wrote down ahead of time. If this, then do that. So software buys the part of a job you could put in a manual, send the invoice, run the payroll and a paycheck buys everything else like knowing which customer to call back first and which one gets a very, very polite email. Haha. And AI cracks that old deal, open chat, G P t, Claude. These models Learned from a mountain of stuff people wrote down so they can handle way more of a job than if then code. And that's got investors like Sequoia and Foundation Capital basically saying, stop selling the tool, sell the work, don't sell a small business tax software, sell them their taxes filed. It's called service as software. Software as a service flipped around and it goes after the paycheck money, not the software money. But here's the catch, most of What a paycheck really pays for the judgment, like knowing who gets the call and who gets the polite email never got written down. So AI mostly never got to learn it. My bet is the ones who come out ahead write theirs down first and sell it themselves. Because the second any judgment gets written down, it pretty much turns into software. And software is the cheap part. So even with everyone buying AI right now.

You're a programmer. No, really. In 1804 the Jacquard loom read punched cards: hole or no hole, thread up or thread down, and out came a picture. That's a one or a zero. Then machine code, then languages, each layer hiding the mess below it. The newest layer is your sentence. A programmer takes logic and attaches it to symbols, and your words are symbols too. Want to write better ones? Join Clief Notes, link in my bio. #ai #programming #computerhistory #learnai #shorts

you're a programmer the jacquard loom from eighteen o four can prove it it reads punched cards where a rod finds a hole that thread lifts where it doesn't it stays down thousands of times and out comes a picture so those cards are pretty much the first program we really see at scale hole or no hole that's a one or a zero that's machine code just faster then languages turn symbols you can read into something a machine can run so nobody types ones and zeros right each layer hides the mess below that's abstraction and the newest layer is your sentence every programmer i know is about to rip me apart fair enough you're standing on every layer somebody built but a programmer takes logic and attaches it to symbols that's the whole job and your words are symbols too which technically means

Ever heard of a draw boy? For centuries, fancy silk took two people: a weaver, and a draw boy pulling cords by hand to lift the right threads, every single row. In 1804 Jacquard's loom put that job on punched cards. By 1839 someone wove a portrait from about twenty-four thousand of them. A program is a pattern written down so a machine can run it. If your work only lives in your head and you walk your AI through it every time, you're the draw boy. Want to write yours down? Join Clief Notes, link in my bio. #ai #history #computerhistory #learnai #shorts

Ever heard of a draw boy? It was a real job for centuries, and there's a decent chance you're doing it right now. A loom makes a picture by lifting some threads and not others, row after row. And fancy silk took two people, a weaver and a draw boy, pulling cords by hand to lift the right ones every single row. So it's slow, it's expensive, and it can only get so complicated. Now, in 18:04, Jacard's Loom put that job on punched cards, one card per row. A hole means that thread lifts, so one weaver can run any design, just swap the cards. So yeah, the draw boys job mostly went away, and silk weavers fought these looms hard, scared for their work. Fair enough, right? But the work moved. Somebody still had to punch every design into cards once, and then anyone could weave it again and again. And on cards, there's pretty much no ceiling. In 1839, somebody wove a portrait of Jacard in silk from about 24,000 cards. And Charles Babbage, a guy designing a computer made of gears, hung one at home. And he says two Royal Academy artists took it for an engraving, a print. Real artists fooled by cardboard with holes in it. Ha ha. Four years later, Ada Lovelace, often called the first programmer, wrote that Babbage's engine weaves algebra the way the loom weaves flowers and leaves. And that's all a program is, a pattern written down. So a Machine can run it and your AI is a loom too. It follows the pattern. You write down a few plain sentences on how you do the job and anyone on your team can run it. So if your work only lives in your head and you walk your AI through it thread by thread every single time, then quick question.