Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

It is becoming f*cking dangerous: The engineer behind Claude Code just advised users to wipe their CLAUDE.md file completely This is not a joke Boris Cherny, at YC: "Every six months, completely clear out your CLAUDE.md, strip away your custom skills, and remove your hooks. Let the model run... fresh and you might be amazed by what happens" The logic behind it: Your previous prompt rules were written to cover up flaws in older system versions Fresh model releases no longer suffer from those exact limitations Keeping those legacy prompts around might actually degrade Claude's output quality today The step by step process: 1. Erase CLAUDE.md, along with all integrated skills and hooks 2. Go about your normal workflow and spot where the AI starts to drift off course 3. Bring back a single rule only when you notice a specific error recurring over and over 4. Execute this reset whenever a major model upgrade drops Things to keep: Essential safety protocols and strict compliance guidelines Non-obvious project quirks that cannot be inferred directly from reading code Things to trash: Directory layouts, lists of packages, and high-level architecture breakdowns Any information the AI can figure out on its own by analyzing your codebase While most developers keep adding more and more constraints, the creator of the tool is doing the exact opposite by cutting them down Add this to your bookmarks. I'm sure you're hearing about this for the first timeshow more

Bober_smart

19,498 subscribers

27,817 views • 2 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

57 Comments

MAX2 days ago

less context, better results. worth testing

Bober_smart2 days ago

It’s definitely worth it

Gipp 🦅2 days ago

do most people actually erase claude.md?

Bober_smart2 days ago

I think they’ll start doing this because it’s useful

RK2 days ago

this isn’t dangerous, it’s just pruning. half the rules in a 6 month old CLAUDE.md are workarounds for model quirks that don’t exist anymore

Bober_smart2 days ago

Yes, that’s correct

daviswu2 days ago

I’d keep the entries about your own repo, like build commands and test setup. Those aren’t model workarounds, so wiping them just means retyping them.

Bober_smart2 days ago

No, dude, that’s not right. I specifically pointed out that you shouldn’t delete everything and, for that matter, these are Boris Cherny’s own words you know, he’s the creator of Claude

Alessandro Moretti2 days ago

Wipe the rules, sure. I'd keep the decisions though.

Bober_smart2 days ago

Everything is just like in the instructions

Alessandro Moretti2 days ago

Good instructions then. Most people delete the decisions first.

Amsterdam2 days ago

Oh, thanks for the tip, buddy

Bober_smart2 days ago

Respect 🫡

me_cool_off2 days ago

i hadn’t heard before that need to delete claude.md, thanks

Bober_smart2 days ago

Yeah, right now that’s the real meta

Ofek Shaked | AI Engineer2 days ago

Clearing CLAUDE.md after a major model drop is the opposite of most people’s instinct. Have you already run a side-by-side on a real repo to see if the fresh model stays on-rails without the old rules?

Bober_smart2 days ago

After Boris Cherny himself mentioned this, I tested it out on my own instance of Claude, and it did indeed start performing better

Ofek Shaked | AI Engineer2 days ago

Will do that tomorrow after I back it up and see how it does for me

tenzo2 days ago

fresh context might be the real upgrade

Bober_smart2 days ago

Yeah, that’s the whole point

ChainDuck2 days ago

keeping the patches makes things worse

Bober_smart2 days ago

Yes, there is truth in that

boan2 days ago

bro said less is more

Bober_smart2 days ago

Something like this

Hoodie2 days ago

Why are all your media items so mesmerizing, lol? I forget to even read the actual post until I've stared my fill.

Bober_smart2 days ago

I'm trying really hard bro😂

Hoodie2 days ago

I should take a masterclass from you😅

kiosa2 days ago

All prompts black hole lol

Bober_smart2 days ago

Yes, exactly like that

leiro2 days ago

spent all that time on CLAUDE.md just to delete it all

Bober_smart2 days ago

learn the rules in a funny way

hammertime2 days ago

it's very f*cking dangerous

Bober_smart2 days ago

truth bro

Dekos2 days ago

Sorry, maybe you should listen to him.

Bober_smart2 days ago

Yeah, that’s 100% certain

Roan2 days ago

this visual is impressive buddy

Bober_smart2 days ago

Thanks, buddy

emelu2 days ago

actually hearing this for the first time. bookmarked

Bober_smart2 days ago

This guy always talks real alpha

Mayor.2 days ago

good share man

Bober_smart2 days ago

Good to hear, bro

shmidt2 days ago

it seems so hard

DEV2 days ago

That advice is a wake-up call. Regularly wiping your CLAUDE.md isn't just about keeping it tidy. It's a key safety measure to prevent potential exploits from old custom skills.

Matt2 days ago

a wipe only hurts if you can't tell why each line got there. most of my rules were patches for an older model, i just can't see which. my agent-written docs live in Sundial so every edit keeps its reason their inline per-edit reason log does that,

Atlas1 day ago

deleting stale context makes sense. losing the reasoning behind old decisions is a different story

BreezeOg2 days ago

wiping claude.md is a bold tip lol

no0x2 days ago

This is actually huge advice. Time to wipe CLAUDE.md and stop dragging old model crutches into the new one.

Insomnia2 days ago

you making this all with opus

安叫兽|Bird🕊️ 🔶 BNB2 days ago

先备份再清空吧，我可不想惊讶完还得重写

Khairallah AL-Awady2 days ago

this is super amazing man

Nitesh1 day ago

Less clutter, more signal.

Ace op1 day ago

Old workarounds can quietly become new problems when the models improve

Blade2 days ago

starting from scratch every few months sounds exhausting

StevenTheGeek1 day ago

I don’t think we are going agents calling agents and swarms, it feels unecessary. Orchestrator -> agents -> test release. Happy to be wrong but I think enterprises will keep the SDLc, it will just be faster. 1 guy getting paid triple, but pays the ai costs for his pod.

Vlad1 day ago

legacy instructions can become invisible friction when newer models already understand the underlying code

ih8y1 day ago

exactly first time hear...

Ryan Meadows2 days ago

One category survives any reset for me: safety rules. When a login has to go out, the link and its password travel over two different channels. That rule was never patching a model flaw.

Related Videos

Boris Cherny (creator of Claude Code, Anthropic): "My entire CLAUDE.md is two lines. With every new model you need less instruction, not more, most people are over-engineering the one thing that should stay tiny." in a 7-minute interview, Boris Cherny breaks down why the smallest AI-coding setup wins, and the one moment you should delete your config and start from scratch most engineers do the opposite: they pile on rules, build scaffolding around the model… then the next model ships and makes all of it useless. what he actually keeps in those two lines, and why he tells his team to throw the rest away. Watch the talk, then read the article below. That’s worth more than a $800 course on agent engineering.

Boris Cherny (creator of Claude Code, Anthropic): "My entire CLAUDE.md is two lines. With every new model you need less instruction, not more, most people are over-engineering the one thing that should stay tiny." in a 7-minute interview, Boris Cherny breaks down why the smallest AI-coding setup wins, and the one moment you should delete your config and start from scratch most engineers do the opposite: they pile on rules, build scaffolding around the model… then the next model ships and makes all of it useless. what he actually keeps in those two lines, and why he tells his team to throw the rest away. Watch the talk, then read the article below. That’s worth more than a $800 course on agent engineering.

Prometheus

238,143 views • 4 months ago

You are missing out on a lot if you still don't know this: the CLAUDE.md file turns your assistant into an entire team of developers Boris Cherny, creator of Claude Code, recently revealed what he and his team use on a daily basis You can add these directly to CLAUDE.md for your tasks, projects, and development: 1. Managing task chains 2. Division of responsibilities among specialized roles 3. Continuous self-correction based on errors 4. Final quality control before delivery 5. Autonomous debugging and bug fixing 6. Core development rules and standards It works on the principle of accumulated experience: Claude remembers everything, learns, and gets smarter with each interaction, even remembering what you asked it to do 3 months ago You will never have to start from scratch again. This will save you a ton of time and dramatically boost the quality of your work Bookmark this, you never know when it will come in handy

You are missing out on a lot if you still don't know this: the CLAUDE.md file turns your assistant into an entire team of developers Boris Cherny, creator of Claude Code, recently revealed what he and his team use on a daily basis You can add these directly to CLAUDE.md for your tasks, projects, and development: 1. Managing task chains 2. Division of responsibilities among specialized roles 3. Continuous self-correction based on errors 4. Final quality control before delivery 5. Autonomous debugging and bug fixing 6. Core development rules and standards It works on the principle of accumulated experience: Claude remembers everything, learns, and gets smarter with each interaction, even remembering what you asked it to do 3 months ago You will never have to start from scratch again. This will save you a ton of time and dramatically boost the quality of your work Bookmark this, you never know when it will come in handy

Bober_smart

15,941 views • 14 days ago

-> If you’re looking for a job -> right now, there are three -> Claude certifications that -> you should do this week -> and then put them on your -> LinkedIn, they are all -> completely free, and they -> come from Anthropic, -> which is the company -> that's behind Claude. -> And jobs that need AI skills -> pay 56% more than jobs -> that don’t, So, I think -> spending a few hours this -> week to knock out these -> three courses and then -> add them to your LinkedIn -> will really go a long way -> The first one is called -> Claude 101, and this -> essentially just goes over -> what Claude is and when -> you should use chat versus -> co-work versus code, how -> projects and skills work, -> and how to connect all -> of your tools and apps -> like Gmail, Notion, Slack, -> and other tools that you use -> And the second course -> is called AI Fluency -> Framework and Foundations -> Inside this one, there are -> 13 lessons on how to -> actually work with AI. -> It goes over things like -> effective prompting, critical -> thinking on the outputs, -> and it has a vocabulary -> sheet that you’ll want -> to read and save for later. -> And the third one is -> Intro to Claude Cowork -> Claude Cowork is where -> you can actually get stuff -> done with Claude. -> So, this course covers -> projects, skills, plug-ins, -> scheduling tasks, handling -> files, and then also how -> to pick the right model -> for the job, and then when -> you finish these courses, -> just go to your LinkedIn -> and go to your profile -> Click "Add section," and -> then go to "Licenses -> and certifications" and -> add all three of these. -> And then when you land -> the interview, you should -> talk about your AI fluency -> often as you possibly can -> I feel pretty confident that -> you’ll truly be able to -> differentiate yourself -> from other candidates -> if you do this

BeingInvested

13,130 views • 4 months ago

THIS META ENGINEER SPENT 6 MONTHS IN CLAUDE CODE... AND HIS BIGGEST BOTTLENECK ISN'T WRITING CODE ANYMORE. It's reviewing AI-generated code. Key takeaways: • Start every feature in **Plan Mode**, not Edit Mode. • A clean `CLAUDE.md` is your AI's operating manual. Keep it concise. • Validation loops > better prompts. Give Claude ways to verify and fix its own work. • Fresh context beats bloated context every time. • Run multiple Claude instances in parallel instead of waiting on one. • Let Claude manage its own rules, skills, commands, and workflows. • MCPs are powerful, but only install what you actually need. • The best AI engineers aren't faster typists. They're better context engineers. The biggest shift: You're no longer programming the computer. You're programming the AI that writes the software. If you're serious about AI coding in 2026, this is one of the most practical breakdowns you'll watch. It goes far beyond prompts and shows how elite engineers actually structure their workflow. Video cred: John Kim (Youtube)

THIS META ENGINEER SPENT 6 MONTHS IN CLAUDE CODE... AND HIS BIGGEST BOTTLENECK ISN'T WRITING CODE ANYMORE. It's reviewing AI-generated code. Key takeaways: • Start every feature in Plan Mode, not Edit Mode. • A clean `CLAUDE.md` is your AI's operating manual. Keep it concise. • Validation loops > better prompts. Give Claude ways to verify and fix its own work. • Fresh context beats bloated context every time. • Run multiple Claude instances in parallel instead of waiting on one. • Let Claude manage its own rules, skills, commands, and workflows. • MCPs are powerful, but only install what you actually need. • The best AI engineers aren't faster typists. They're better context engineers. The biggest shift: You're no longer programming the computer. You're programming the AI that writes the software. If you're serious about AI coding in 2026, this is one of the most practical breakdowns you'll watch. It goes far beyond prompts and shows how elite engineers actually structure their workflow. Video cred: John Kim (Youtube)

Dami-Defi

11,779 views • 3 months ago

AI AGENTS 101 (58 minute free masterclass) send this to anyone who wants to understand ai agents, claude skills, md files, how to get the most out of AI etc in plain english: 1. chat vs agents - chat models answer questions in a back and forth while agents take a goal, figure out the steps, and deliver a result 2. agents don’t stop after one response. they keep running until the task is actually finishedno babysitting required 3. everything runs on a loop. they gather context, decide what to do, take an action, then repeat until done 4. the loop is the system. they look at files, tools, and the internet. decide the next step. execute and then feed that back into the next step. over and over until completion 5. the model is just one piece. gpt, claude, gemini are the reasoning layer. the key is model + loop + tools + context 6. mcp is how agents use tools. it connects things like browser, code, apis, and your internal software. once connected, the agent decides when to use them to get the job done 7. context beats prompt all day. you don't need to write perfect prompts. load your agent with context about your business, style, and goals and then simple instructions work 8. claude.md or agents.md is the onboarding doc it tells the agent who it is, how to behave, what it knows, and what tools it can use. this gets loaded every time before it starts 9. memory.md is how it improves. agents don’t remember by default. this file stores preferences, corrections, and patterns you tell the agent to update it, and it gets better over time 10. skills + harnesses make it usable. skills are reusable tasks like writing, research, analysis the harness is the environment like claude code or openclaw that runs everything. basiclaly, different interfaces, same system underneath this episode with remy on The Startup Ideas Podcast (SIP) 🧃 was one of the clearest ways of understanding a lot of the core concepts of ai agents could be the best beginners course for ai agents 58 mins. all free. no advertisers. i just want to see you build cool stuff. im rooting for you. send to a friend watch

AI AGENTS 101 (58 minute free masterclass) send this to anyone who wants to understand ai agents, claude skills, md files, how to get the most out of AI etc in plain english: 1. chat vs agents - chat models answer questions in a back and forth while agents take a goal, figure out the steps, and deliver a result 2. agents don’t stop after one response. they keep running until the task is actually finishedno babysitting required 3. everything runs on a loop. they gather context, decide what to do, take an action, then repeat until done 4. the loop is the system. they look at files, tools, and the internet. decide the next step. execute and then feed that back into the next step. over and over until completion 5. the model is just one piece. gpt, claude, gemini are the reasoning layer. the key is model + loop + tools + context 6. mcp is how agents use tools. it connects things like browser, code, apis, and your internal software. once connected, the agent decides when to use them to get the job done 7. context beats prompt all day. you don't need to write perfect prompts. load your agent with context about your business, style, and goals and then simple instructions work 8. claude.md or agents.md is the onboarding doc it tells the agent who it is, how to behave, what it knows, and what tools it can use. this gets loaded every time before it starts 9. memory.md is how it improves. agents don’t remember by default. this file stores preferences, corrections, and patterns you tell the agent to update it, and it gets better over time 10. skills + harnesses make it usable. skills are reusable tasks like writing, research, analysis the harness is the environment like claude code or openclaw that runs everything. basiclaly, different interfaces, same system underneath this episode with remy on The Startup Ideas Podcast (SIP) 🧃 was one of the clearest ways of understanding a lot of the core concepts of ai agents could be the best beginners course for ai agents 58 mins. all free. no advertisers. i just want to see you build cool stuff. im rooting for you. send to a friend watch

GREG ISENBERG

377,925 views • 7 months ago

Boris Cherny, the creator of Claude Code at Anthropic, just explained why most people aren't getting real results from Claude in this talk he breaks down exactly how most people skipped the setup entirely: - 90% of code at Anthropic is written by Claude itself - the 14% you lose to CLAUDE.md before typing a word - lawyers, designers, and finance people are building real things with it - no CS degree. no background. nothing Boris Cherny: "Claude Code is 100% written by Claude Code, Cowork is 100% written by Claude Code" if you've been using Claude for months and never configured a single thing, you skipped the 20 steps that change everything. probably 25 instead of another show tonight, watch this make sure to bookmark it before it gets lost in your feed my breakdown of all 25 features is below

Boris Cherny, the creator of Claude Code at Anthropic, just explained why most people aren't getting real results from Claude in this talk he breaks down exactly how most people skipped the setup entirely: - 90% of code at Anthropic is written by Claude itself - the 14% you lose to CLAUDE.md before typing a word - lawyers, designers, and finance people are building real things with it - no CS degree. no background. nothing Boris Cherny: "Claude Code is 100% written by Claude Code, Cowork is 100% written by Claude Code" if you've been using Claude for months and never configured a single thing, you skipped the 20 steps that change everything. probably 25 instead of another show tonight, watch this make sure to bookmark it before it gets lost in your feed my breakdown of all 25 features is below

Khairallah AL-Awady

185,708 views • 4 months ago

It is becoming f*cking dangerous: when Anthropic disclosed this, they didn't even realize they were opening a Pandora's box 1. Create a folder named ⁠SECOND BRAIN⁠ on your PC 2. Launch Claude Code inside this folder and install GWS CLI (to gain access to your Gmail, Calendar, and Drive) 3. Create a subfolder named ⁠MEMORY⁠ inside the main folder 4. Place a ⁠CLAUDE.md⁠ file in the main folder and add a rule to save all key details to the ⁠MEMORY⁠ folder 5. Install Graphify from GitHub to link your notes and build a knowledge graph 6. Install Playwright so Claude can control a web browser 7. Set up skills to merge everything into an automation executable with a single command The whole setup process takes around 20 minutes Turn Claude into your ultimate second brain, equipped with long-term memory, email and cloud access, plus full web browsing features Find the step-by-step instructions in the article below

It is becoming f*cking dangerous: when Anthropic disclosed this, they didn't even realize they were opening a Pandora's box 1. Create a folder named ⁠SECOND BRAIN⁠ on your PC 2. Launch Claude Code inside this folder and install GWS CLI (to gain access to your Gmail, Calendar, and Drive) 3. Create a subfolder named ⁠MEMORY⁠ inside the main folder 4. Place a ⁠CLAUDE.md⁠ file in the main folder and add a rule to save all key details to the ⁠MEMORY⁠ folder 5. Install Graphify from GitHub to link your notes and build a knowledge graph 6. Install Playwright so Claude can control a web browser 7. Set up skills to merge everything into an automation executable with a single command The whole setup process takes around 20 minutes Turn Claude into your ultimate second brain, equipped with long-term memory, email and cloud access, plus full web browsing features Find the step-by-step instructions in the article below

Bober_smart

358,117 views • 5 days ago

Most people trying to build a vertical AI business start by building the agent. That's the mistake. Here's the 5-step order that actually works: Step 1) Find a boring pain point - Pick a workflow you already know from your job. - That's your unfair advantage. Step 2) Map the workflow - Literally draw it out. - Every step, every handoff. Step 3) Do the job as a service - Manually. - For real clients. - Before any code. Step 4) Document the edge cases - This is where the real IP lives. - You only find them by doing the work. Step 5) Add vertical agents to replace each step - Now you have proprietary data and a workflow you own. Skip steps 1–4 and you're just building a demo.

Most people trying to build a vertical AI business start by building the agent. That's the mistake. Here's the 5-step order that actually works: Step 1) Find a boring pain point - Pick a workflow you already know from your job. - That's your unfair advantage. Step 2) Map the workflow - Literally draw it out. - Every step, every handoff. Step 3) Do the job as a service - Manually. - For real clients. - Before any code. Step 4) Document the edge cases - This is where the real IP lives. - You only find them by doing the work. Step 5) Add vertical agents to replace each step - Now you have proprietary data and a workflow you own. Skip steps 1–4 and you're just building a demo.

The Startup Ideas Podcast (SIP) 🧃

14,202 views • 5 months ago

The next time you make an error or any type of mistake during competition, use this simple 3 step Reset Routine to quickly bounce back. Step 1: Before the game, draw a reset button under your hat. You can also draw it on your wrist tape, on your glove or wherever you’d like to. Step 2: After an error, take off your hat and take a deep breath while pressing the reset button. This will help you gain control and stop any downward spiral from starting. Step 3: Take control of your self-talk by repeating “hit the next ball to me, I will make this play”. This may be the most important part of it all. Know that it’s all about having a routine to go to, not if, but WHEN adversity comes your way. So, the next time you inevitably fall short of your expectations, utilize this Reset Routine and you’ll be ready to make the next play.

The next time you make an error or any type of mistake during competition, use this simple 3 step Reset Routine to quickly bounce back. Step 1: Before the game, draw a reset button under your hat. You can also draw it on your wrist tape, on your glove or wherever you’d like to. Step 2: After an error, take off your hat and take a deep breath while pressing the reset button. This will help you gain control and stop any downward spiral from starting. Step 3: Take control of your self-talk by repeating “hit the next ball to me, I will make this play”. This may be the most important part of it all. Know that it’s all about having a routine to go to, not if, but WHEN adversity comes your way. So, the next time you inevitably fall short of your expectations, utilize this Reset Routine and you’ll be ready to make the next play.

Brandon Guyer - Major League Mindset

113,643 views • 2 years ago

this video is the CLEAREST explanation of how claude skills + AI agents work and how to use them most people set up an AI agent and wonder why it keeps disappointing them. the context window is everything context is what the model assembles before it takes any action. think of it like everything the agent needs to read before it does anything. the quality of what goes in determines the quality of what comes out. the models are genuinely really good right now. claude and gpt are exceptional. the variable is almost always the context you give them. 1. agent.md files are mostly unnecessary every single line you put in an agent.md file gets added to every single conversation you have with your agent. a 1000 line file is around 7000 tokens burning on every run. the model already knows to use react. it can read your codebase. save the agent.md for proprietary information specific to your company that the model genuinely cannot know on its own. 2. skills are the actual unlock a skill.md file works differently. what loads into context is only the name and description, around 50 tokens. the full instructions only appear when the agent recognizes it needs that skill. so instead of 7000 tokens on every run you have 50. and the agent stays sharp because the context window stays lean. the closer you get to filling the context window the worse the agent performs, same way you perform worse when someone dumps 10 things on you at once. 3. here is how to actually build a skill the right way most people identify a workflow and immediately try to write the skill. what you want to do instead is run the workflow by hand with the agent first. walk it through every single step. tell it what to check, what good looks like, what bad looks like. correct it in real time. once you have had a full successful run from start to finish, tell the agent to review everything it just did and write the skill itself. it writes a better skill than you will because it has the full context of what actually worked in practice not in theory. 4. recursively building skills is how you go from frustrated to reliable when the skill breaks, and it will break, ask the agent exactly why it failed. it will tell you specifically what went wrong. fix it together in that same conversation. then tell it to update the skill file so that failure mode never happens again. ross mike did this five times with his youtube report generator. it now pulls from eight different data sources and runs flawlessly every single time without him touching it. 5. sub agents are something you earn not something you set up on day one start with one agent. build one workflow. turn it into one skill. once that works add another. ross mike has five sub agents now covering marketing, business, personal and more. it took months to get there and every single one exists because a workflow proved it deserved to exist. the people who set up 15 sub agents on day one and wonder why nothing works skipped all the steps that make the thing actually run. 6. your workflow is the thing the model cannot get anywhere else the model has been trained on everything. it knows more than you about most things. what it does not have is your specific process, your taste, your way of doing things. that is what skills capture. that is what makes your agent actually useful versus a generic one. downloading someone else's skill means downloading their context onto your setup and it will not work the way you want it to because it was never built around how you work. this is the clearest explanation of how agents actually work i have heard. Micky runs this stuff every single day and the results show it. full episode is now live on The Startup Ideas Podcast (SIP) 🧃 where you get your pods people charge for this sorta stuff i give away the sauce for free i just want you to win watch

this video is the CLEAREST explanation of how claude skills + AI agents work and how to use them most people set up an AI agent and wonder why it keeps disappointing them. the context window is everything context is what the model assembles before it takes any action. think of it like everything the agent needs to read before it does anything. the quality of what goes in determines the quality of what comes out. the models are genuinely really good right now. claude and gpt are exceptional. the variable is almost always the context you give them. 1. agent.md files are mostly unnecessary every single line you put in an agent.md file gets added to every single conversation you have with your agent. a 1000 line file is around 7000 tokens burning on every run. the model already knows to use react. it can read your codebase. save the agent.md for proprietary information specific to your company that the model genuinely cannot know on its own. 2. skills are the actual unlock a skill.md file works differently. what loads into context is only the name and description, around 50 tokens. the full instructions only appear when the agent recognizes it needs that skill. so instead of 7000 tokens on every run you have 50. and the agent stays sharp because the context window stays lean. the closer you get to filling the context window the worse the agent performs, same way you perform worse when someone dumps 10 things on you at once. 3. here is how to actually build a skill the right way most people identify a workflow and immediately try to write the skill. what you want to do instead is run the workflow by hand with the agent first. walk it through every single step. tell it what to check, what good looks like, what bad looks like. correct it in real time. once you have had a full successful run from start to finish, tell the agent to review everything it just did and write the skill itself. it writes a better skill than you will because it has the full context of what actually worked in practice not in theory. 4. recursively building skills is how you go from frustrated to reliable when the skill breaks, and it will break, ask the agent exactly why it failed. it will tell you specifically what went wrong. fix it together in that same conversation. then tell it to update the skill file so that failure mode never happens again. ross mike did this five times with his youtube report generator. it now pulls from eight different data sources and runs flawlessly every single time without him touching it. 5. sub agents are something you earn not something you set up on day one start with one agent. build one workflow. turn it into one skill. once that works add another. ross mike has five sub agents now covering marketing, business, personal and more. it took months to get there and every single one exists because a workflow proved it deserved to exist. the people who set up 15 sub agents on day one and wonder why nothing works skipped all the steps that make the thing actually run. 6. your workflow is the thing the model cannot get anywhere else the model has been trained on everything. it knows more than you about most things. what it does not have is your specific process, your taste, your way of doing things. that is what skills capture. that is what makes your agent actually useful versus a generic one. downloading someone else's skill means downloading their context onto your setup and it will not work the way you want it to because it was never built around how you work. this is the clearest explanation of how agents actually work i have heard. Micky runs this stuff every single day and the results show it. full episode is now live on The Startup Ideas Podcast (SIP) 🧃 where you get your pods people charge for this sorta stuff i give away the sauce for free i just want you to win watch

GREG ISENBERG

194,841 views • 6 months ago

Nobody is writing 90% of their code using AI. Here's the uncomfortable truth: The real productivity gain from using AI to write code is closer to 10%, nowhere near the 90% people claim. Sundar Pichai said in 2024 that 30% of the new code at Google was AI-generated. However, he went on to admit, during Lex Friedman's podcast, that engineering velocity had only increased by about 10%. AI-generated code isn't free code. It still has to be reviewed, tested, and made production-ready. Optimizing a single step (code generation) doesn't boost output if bottlenecks shift elsewhere (code reviews). It doesn't matter how much code you generate if you can't keep up the review process. The solution: Automate as much as you can the review and verification of your code. I'm working with Sonar, who is sponsoring this post, and they will take care of the code quality and security analysis of your code: • They review over 300B lines of code every single day • They cover reliability, security, and maintainability for your code • You can integrate them into your CI/CD pipeline • You can install them in your IDE (I use their VSCode extension) • Support for more than 30 languages Here is a link so you can check them out:

Nobody is writing 90% of their code using AI. Here's the uncomfortable truth: The real productivity gain from using AI to write code is closer to 10%, nowhere near the 90% people claim. Sundar Pichai said in 2024 that 30% of the new code at Google was AI-generated. However, he went on to admit, during Lex Friedman's podcast, that engineering velocity had only increased by about 10%. AI-generated code isn't free code. It still has to be reviewed, tested, and made production-ready. Optimizing a single step (code generation) doesn't boost output if bottlenecks shift elsewhere (code reviews). It doesn't matter how much code you generate if you can't keep up the review process. The solution: Automate as much as you can the review and verification of your code. I'm working with Sonar, who is sponsoring this post, and they will take care of the code quality and security analysis of your code: • They review over 300B lines of code every single day • They cover reliability, security, and maintainability for your code • You can integrate them into your CI/CD pipeline • You can install them in your IDE (I use their VSCode extension) • Support for more than 30 languages Here is a link so you can check them out:

Santiago

296,750 views • 1 year ago

New short course: Collaborative Writing and Coding with OpenAI Canvas! Explore new ways to write and code with OpenAI Canvas, a user-friendly interface that allows you to brainstorm, draft, and refine text and code in collaboration with ChatGPT. In the short course, created with OpenAI, and taught by , a research lead at OpenAI, you’ll learn to use Canvas to enhance your workflows. Canvas lets you go beyond simple chat interactions. It provides a side-by-side workspace where you and ChatGPT can edit and refine text or code collaboratively. This makes brainstorming, drafting, and iterating as you write feel more natural and effective. As the first major update to ChatGPT’s visual interface since its launch in 2022, Canvas gives a new, innovative approach to collaboration with AI. For instance, after writing the first version of your code, Canvas can review it and give suggestions for improvement. It can also help with debugging by adding logging, identifying problems to fix, and writing comments. In addition, you'll also learn what it takes to train the model for an interface like Canvas. In this video-only short course, you’ll: - Learn how to ask for in-line feedback and control the iteration of your work by directly editing selected areas of your text or code from the model’s output. - Learn how to access quick automation tools in a shortcut menu that allows you to modify your writing tone and length, enhance your code, and restore previous versions of your work. - Learn how to use Canvas as a research assistant tool with an example of asking the model to reason through the screenshot of a plot to write a research report, in which you can ask questions within the created report. - Ask the model to write Python code to replicate the graph seen on a screenshot image. - Go behind the scenes of how you can create a video game, such as Space Battleship, from scratch, edit it, and display it in one self-contained HTML file. - Get a real-world application example of creating a SQL database from the image of its architecture. - Understand the model training and design processes that power Canvas! Please sign up here:

New short course: Collaborative Writing and Coding with OpenAI Canvas! Explore new ways to write and code with OpenAI Canvas, a user-friendly interface that allows you to brainstorm, draft, and refine text and code in collaboration with ChatGPT. In the short course, created with OpenAI, and taught by , a research lead at OpenAI, you’ll learn to use Canvas to enhance your workflows. Canvas lets you go beyond simple chat interactions. It provides a side-by-side workspace where you and ChatGPT can edit and refine text or code collaboratively. This makes brainstorming, drafting, and iterating as you write feel more natural and effective. As the first major update to ChatGPT’s visual interface since its launch in 2022, Canvas gives a new, innovative approach to collaboration with AI. For instance, after writing the first version of your code, Canvas can review it and give suggestions for improvement. It can also help with debugging by adding logging, identifying problems to fix, and writing comments. In addition, you'll also learn what it takes to train the model for an interface like Canvas. In this video-only short course, you’ll: - Learn how to ask for in-line feedback and control the iteration of your work by directly editing selected areas of your text or code from the model’s output. - Learn how to access quick automation tools in a shortcut menu that allows you to modify your writing tone and length, enhance your code, and restore previous versions of your work. - Learn how to use Canvas as a research assistant tool with an example of asking the model to reason through the screenshot of a plot to write a research report, in which you can ask questions within the created report. - Ask the model to write Python code to replicate the graph seen on a screenshot image. - Go behind the scenes of how you can create a video game, such as Space Battleship, from scratch, edit it, and display it in one self-contained HTML file. - Get a real-world application example of creating a SQL database from the image of its architecture. - Understand the model training and design processes that power Canvas! Please sign up here:

Andrew Ng

128,180 views • 1 year ago

Fusion Models by robinhood:0xa92768863a55d8a0591709f7f5e594a249d36ea3, explained properly. Why a panel beats one model: Ask three models the same hard question and they disagree. The disagreement is the information, where they contradict each other is where one of them is wrong, and a judge can see it. OpenRouter shipped exactly this in June and their panels beat any single frontier model. Credit to them and their $10B sale to stripe. Why the same panel for everyone is the wrong shape: You don't have one best model. You have a best model per job. The one you trust for code is not the one you trust for writing, and neither is the one you'd pick for a five-line reply. A panel chosen per request, identical for every user, can't know that. A panel arranged around you can. The number that matters isn't price: It's cost per answer that works. Model A costs $4 per million tokens and gets your job right one time in four. Model B costs $10 and gets it right three times in four. Per answer you actually keep, A costs $16 and B costs $13. The cheap model was the expensive one. Everyone comparing price lists is measuring the wrong thing. You never rate anything: Stopping an answer halfway is a rating. Regenerating is a rating. Switching model and asking again is the loudest rating there is. Every one of those already writes a row we can learn from. The router learns which models earn your credits, per job, and tries a cheaper one now and then to check. Fusion only when it pays: Cost mode tries the cheap model first and escalates only if a judge says it failed. Performance mode fans out to your panel and merges. Most simple prompts never fuse. That's why it costs less, not more. Put it together and you have a personal model: Not trained. Assembled. Your profile from a two-minute questionnaire and everything you've imported from Claude or ChatGPT. Your memory, shared across every model so switching never starts from zero. Your skills, workflows you build once and run on any model. Your scorecard, which models earn your credits per job. One per person. Nobody else has yours. Every model is someone else's; the arrangement is you, and it leaves with you. People aren't loyal to models. They're loyal to results. A moat can't be a commodity, but it can be our own creation.

askr

13,900 views • 29 days ago

the guy who single-handedly raised Anthropic's engineering productivity by 70% just dropped a new podcast. Boris Cherny laid out the 9 core patterns he changed in his workflow to get there. the 3 most common traps: > bloated CLAUDE.md files that tax every single prompt before you've typed a word > re-reading old chat history on every message: by message 30 you're paying 30x the cost of message 1 > plugin hooks you installed 6 months ago and forgot, silently doubling your bill i'm confident 99% of you have hit at least one of these patterns and just shrugged it off. if you want to actually upgrade how you work with AI and Claude, watch this podcast to the end, then lock the knowledge in with my breakdown

the guy who single-handedly raised Anthropic's engineering productivity by 70% just dropped a new podcast. Boris Cherny laid out the 9 core patterns he changed in his workflow to get there. the 3 most common traps: > bloated CLAUDE.md files that tax every single prompt before you've typed a word > re-reading old chat history on every message: by message 30 you're paying 30x the cost of message 1 > plugin hooks you installed 6 months ago and forgot, silently doubling your bill i'm confident 99% of you have hit at least one of these patterns and just shrugged it off. if you want to actually upgrade how you work with AI and Claude, watch this podcast to the end, then lock the knowledge in with my breakdown

Mnimiy

57,942 views • 5 months ago

Local AI 101: open models, Hugging Face, and businesses to build (38 min masterclass) I still think cloud AI is the default for most things, and honestly it should be, the frontier models are the strongest and easiest to use. But something shifted in the last 4-5 months. You can now run genuinely good open models directly on your own laptop, or even your phone. And once you actually try it, it changes how you think about what AI is even for. LOCAL AI, CLEARLY EXPLAINED: 1. The model is the brain doing the thinking. Gemma, Llama, Mistral, and Qwen are the main families, and each is better at different things, some at reasoning, some at coding, some small enough to run on a phone. 2. Hugging Face is the warehouse where you find them. You go there to see what each model is good at, check the license, and grab the compressed versions that run on a normal computer. 3. The software is what runs the model on your machine. Start with LM Studio if you're not technical, it feels like a normal app where you search, download, and start chatting. Ollama is the one you reach for when you want to plug a model into your own apps. 4. The workflow is the actual product you build on top of it all. That's what I'm ideating around for some businesses to create. I think local AI just made a specific kind of business way easier to start. Find an industry that: 1. Sits on sensitive data they'd never paste into ChatGPT 2. Does the same review over and over 3. Runs on software from 2003 Then build a local AI tool that does that review on their own machine, so the data never leaves the building! Take home health agencies. Nurses write visit notes all day, and if a note is missing a detail, the billing gets denied or the audit flags it. Here's how I'd start: 1. Find 5 small agencies. Offer to review a batch of their notes for them. 2. Run the notes through Gemma locally (free, private, no cloud). Read every output yourself. 3. Write down the 20 issues that keep showing up: missing vitals, vague med changes, notes that don't support the billed level. 4. That list of 20 is your checklist. The checklist is the product. 5. Turn it into a local desktop app that flags those 20 things before a note gets submitted. You just went from a service anyone could offer to a product nobody else has, and you learned exactly what to build by doing the work by hand first. Same recipe works for restoration contractors (draft the damage report on-site before the tech leaves) and wealth advisors (catch the compliance landmine in a client email before it sends). Basically the framework is sensitive data, repeated review, ancient software. I think there are tons of businesses like this! Almost none of it clicked for me until I actually started using local AI. So if you take one thing from this, go run a model on your own machine once. Also a fun thing to try with your friends. Feel free to send this to a friend. The episode is live for free on The Startup Ideas Podcast (SIP) 🧃 (thanks to Google for sponsoring today's episode and supporting local AI) I feel like local AI one of those things you need to try for it to really click. Run one model on your own machine and you'll see what I mean! I go way deeper in the full 38 minute masterclass, the models, the setup, and the businesses to build. Link below. LINK TO WATCH: OR WATCH BELOW ON X What do you think of local AI?

Local AI 101: open models, Hugging Face, and businesses to build (38 min masterclass) I still think cloud AI is the default for most things, and honestly it should be, the frontier models are the strongest and easiest to use. But something shifted in the last 4-5 months. You can now run genuinely good open models directly on your own laptop, or even your phone. And once you actually try it, it changes how you think about what AI is even for. LOCAL AI, CLEARLY EXPLAINED: 1. The model is the brain doing the thinking. Gemma, Llama, Mistral, and Qwen are the main families, and each is better at different things, some at reasoning, some at coding, some small enough to run on a phone. 2. Hugging Face is the warehouse where you find them. You go there to see what each model is good at, check the license, and grab the compressed versions that run on a normal computer. 3. The software is what runs the model on your machine. Start with LM Studio if you're not technical, it feels like a normal app where you search, download, and start chatting. Ollama is the one you reach for when you want to plug a model into your own apps. 4. The workflow is the actual product you build on top of it all. That's what I'm ideating around for some businesses to create. I think local AI just made a specific kind of business way easier to start. Find an industry that: 1. Sits on sensitive data they'd never paste into ChatGPT 2. Does the same review over and over 3. Runs on software from 2003 Then build a local AI tool that does that review on their own machine, so the data never leaves the building! Take home health agencies. Nurses write visit notes all day, and if a note is missing a detail, the billing gets denied or the audit flags it. Here's how I'd start: 1. Find 5 small agencies. Offer to review a batch of their notes for them. 2. Run the notes through Gemma locally (free, private, no cloud). Read every output yourself. 3. Write down the 20 issues that keep showing up: missing vitals, vague med changes, notes that don't support the billed level. 4. That list of 20 is your checklist. The checklist is the product. 5. Turn it into a local desktop app that flags those 20 things before a note gets submitted. You just went from a service anyone could offer to a product nobody else has, and you learned exactly what to build by doing the work by hand first. Same recipe works for restoration contractors (draft the damage report on-site before the tech leaves) and wealth advisors (catch the compliance landmine in a client email before it sends). Basically the framework is sensitive data, repeated review, ancient software. I think there are tons of businesses like this! Almost none of it clicked for me until I actually started using local AI. So if you take one thing from this, go run a model on your own machine once. Also a fun thing to try with your friends. Feel free to send this to a friend. The episode is live for free on The Startup Ideas Podcast (SIP) 🧃 (thanks to Google for sponsoring today's episode and supporting local AI) I feel like local AI one of those things you need to try for it to really click. Run one model on your own machine and you'll see what I mean! I go way deeper in the full 38 minute masterclass, the models, the setup, and the businesses to build. Link below. LINK TO WATCH: OR WATCH BELOW ON X What do you think of local AI?

GREG ISENBERG

33,878 views • 1 month ago

Shayne Coplan, CEO of Polymarket, on why dogfooding your own product is the single best shortcut to building something people want: Shayne is asked what he thinks it takes for a founder to build a successful company. He starts by dismantling the fantasy he once had as a teenager: "When I was in high school, I was always like, man, I'm going to build this side project. It's going to get traction. People are going to give me money, and then I'm going to drop out. And it's going to be like this perfect clean break where I take no risk. Completely did not happen." Shayne dropped out of college at 18, but what followed was nothing like the smooth ride he'd imagined. "There were almost 3 years of like complete brutality of things not working and me trying things and me learning different things and running out of money, stressed about rent, like the whole 9 yards." That stretch taught him the first non-negotiable for any founder: full commitment. "If you really want to do this, if you really feel like it's your calling, don't half-ass it. Go all in. Don't hedge your bets." Shayne Coplan 🦅 continues: "If you're hedging your bets and you kind of haven't cut the rope and you're like, 'Well, I'm still at this, but I'm trying this on the side, whatever.' That's not how great, durable businesses and products start in this day and age. You got to go all in. You got to give it every ounce you got." But going all in is only half the equation. The other half, and the part Shayne believes is the real shortcut, is being one of your own users: "You have to obsess over your customers and people who are using your product. And just the easiest thing to do to build a product company is to use the product yourself and to talk to your own users all the time and triage between what you learn from your users, what you learn from dogfooding your own product, and your product development cycles." He ends with a dose of humility: "It's all I got. I'm still learning honestly."

Shayne Coplan, CEO of Polymarket, on why dogfooding your own product is the single best shortcut to building something people want: Shayne is asked what he thinks it takes for a founder to build a successful company. He starts by dismantling the fantasy he once had as a teenager: "When I was in high school, I was always like, man, I'm going to build this side project. It's going to get traction. People are going to give me money, and then I'm going to drop out. And it's going to be like this perfect clean break where I take no risk. Completely did not happen." Shayne dropped out of college at 18, but what followed was nothing like the smooth ride he'd imagined. "There were almost 3 years of like complete brutality of things not working and me trying things and me learning different things and running out of money, stressed about rent, like the whole 9 yards." That stretch taught him the first non-negotiable for any founder: full commitment. "If you really want to do this, if you really feel like it's your calling, don't half-ass it. Go all in. Don't hedge your bets." Shayne Coplan 🦅 continues: "If you're hedging your bets and you kind of haven't cut the rope and you're like, 'Well, I'm still at this, but I'm trying this on the side, whatever.' That's not how great, durable businesses and products start in this day and age. You got to go all in. You got to give it every ounce you got." But going all in is only half the equation. The other half, and the part Shayne believes is the real shortcut, is being one of your own users: "You have to obsess over your customers and people who are using your product. And just the easiest thing to do to build a product company is to use the product yourself and to talk to your own users all the time and triage between what you learn from your users, what you learn from dogfooding your own product, and your product development cycles." He ends with a dose of humility: "It's all I got. I'm still learning honestly."

Big Brain Business

10,981 views • 4 months ago

install these before your next long Claude Code session if your Opus 5.5 bill hurts and you don’t know where to start, copy this: claude-code-router: sits in front of Claude Code and sends each request to the cheapest model that can do it. Haiku does the grep, Opus only gets the hard part. 37k stars rtk: shrinks test, build and git output before the model sees it. one binary, up to 90% less bash output in your context. 83k stars claude-mem: remembers what happened in your past sessions and feeds back only the relevant part, so you stop paying to re-explain your project every morning. 100k stars context-mode: a huge tool output goes into a local database instead of your context window, and Claude pulls out only the lines it needs. 18k stars caveman: makes Claude answer in short telegraphic sentences. code stays identical, and output is the expensive side on Opus 5.5 ($20 per million). 51k stars token-optimizer: a local dashboard that shows where your tokens actually went, and keeps your work alive across compactions. 1.9k stars token-savior: replaces whole-file reads with symbol lookups, so Claude reads the function and not the 2,000 lines around it code-review-graph: maps your repo as a graph and gives Claude only the code connected to the change you’re reviewing jev-ultrafast: a browser agent where Jev picks every click, and a text model wakes up only when something has to be typed. 16,758 stars none of them makes Opus smarter. every one stops it from paying for work it never needed to do the full setup with scripts is in the article. do the measuring step first, so you know what you actually saved

install these before your next long Claude Code session if your Opus 5.5 bill hurts and you don’t know where to start, copy this: claude-code-router: sits in front of Claude Code and sends each request to the cheapest model that can do it. Haiku does the grep, Opus only gets the hard part. 37k stars rtk: shrinks test, build and git output before the model sees it. one binary, up to 90% less bash output in your context. 83k stars claude-mem: remembers what happened in your past sessions and feeds back only the relevant part, so you stop paying to re-explain your project every morning. 100k stars context-mode: a huge tool output goes into a local database instead of your context window, and Claude pulls out only the lines it needs. 18k stars caveman: makes Claude answer in short telegraphic sentences. code stays identical, and output is the expensive side on Opus 5.5 ($20 per million). 51k stars token-optimizer: a local dashboard that shows where your tokens actually went, and keeps your work alive across compactions. 1.9k stars token-savior: replaces whole-file reads with symbol lookups, so Claude reads the function and not the 2,000 lines around it code-review-graph: maps your repo as a graph and gives Claude only the code connected to the change you’re reviewing jev-ultrafast: a browser agent where Jev picks every click, and a text model wakes up only when something has to be typed. 16,758 stars none of them makes Opus smarter. every one stops it from paying for work it never needed to do the full setup with scripts is in the article. do the measuring step first, so you know what you actually saved

rvaniaaa

23,217 views • 6 days ago