Big Idea

The argument in one line.

Claude's recent updates look overwhelming, but you only need a few concepts, starting with a CLAUDE.md file, one skill and /goal prompts, to get most of the value.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

A non-technical founder or operator who keeps hearing about Claude Code, Cowork and skills and wants a plain-English map of what each one is.
Someone on a Pro, Max or Ultra plan who is unsure which model and effort level to use and why their limits run out.
A team lead who wants everyone on the team using Claude consistently, through shared plugins, skills and a CLAUDE.md.

SKIP IF…

You already run Claude Code daily with skills, subagents and scheduled routines; this is an overview, not a deep dive.
You want hands-on build instructions; each concept gets about 40 seconds and the detail lives in other videos.

TL;DR

The full version, fast.

Anthropic shipped three models, in-chat docs, slides and designs, the Duet agent app and a no-five-hour-limit Ultra plan in six weeks, which people call Claude 2.0. Nate Herk explains 25 concepts in five parts. The brain covers models, tokens, effort and plans. Next is where you use Claude: chat, Code, Cowork, sessions, side chats and Duet. Then what it makes: artifacts, docs, dashboards and Motion. Then what it knows: memory, CLAUDE.md, skills, connectors, plugins and mods. Last is autonomy: permission modes, subagents, projects, /goal and scheduled tasks. His advice is to start with a CLAUDE.md, one skill and some /goal prompts.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:34

01 · Intro: why Claude feels like 2.0

Three models, in-chat docs and designs, Duet, and the end of the five-hour limit on the top plan.

00:34 – 05:46

02 · Part 1: The brain

Models, Fable, tokens and the context window, effort levels, and usage limits and plans.

05:46 – 09:33

03 · Part 2: Where you use Claude

Chat, Code, Cowork, parallel sessions, side chats and the Duet agent app.

09:33 – 12:08

04 · Part 3: What Claude makes

Artifacts, docs, slides and designs, dashboards, and Motion animated explainers.

12:08 – 17:39

05 · Part 4: What Claude knows

Memory, CLAUDE.md, skills, connectors, plugins and mods.

17:39 – 20:22

06 · Part 5: Letting Claude work

Permission modes, subagents, projects and /goal.

20:22 – 20:58

07 · Bonus: Scheduled tasks and routines

Recurring tasks, cloud routines, and creating them in natural language.

20:58 – 21:39

08 · Where to start

Set up CLAUDE.md, one skill and some /goal prompts, then build an AI operating system.

Atomic Insights

Lines worth screenshotting.

Sonnet beats Opus on tasks with clear instructions, finishing in about half the time and usually cheaper, while Opus wins when the goal is vague.
In one seven-task measurement Claude read about 160 million tokens and wrote about 800,000, so the files and instructions you load matter more than prompt length.
More effort is not better: the same prompt cost under $4 on low, about $26 on extra high and about $50 on max, and max was still buggy.
Anthropic's own docs warn that max effort is prone to overthinking, so medium is a sound everyday default.
Fable costs two and a half times Opus per use, and Anthropic's release notes say Opus 5.5 performs near Fable level on most work.
Chat, Claude Code and Cowork all draw from the same five-hour and weekly usage buckets.
Every new Claude conversation starts blind, so memory and CLAUDE.md exist to carry who you are across sessions.
Keep CLAUDE.md under about 200 lines because it loads on every single message.
If you have typed the same request twice, turn it into a skill; skills are cheap because Claude only sees a short description until it needs one.
A subagent works in its own context window and reports back a short summary, so messy research never clogs your main conversation.
A /goal runs turn after turn until a finish line is met, which makes a vague finish line the main way to get stuck in a loop.
Motion animations are built with code rather than a video model, so changing a color or size does not require a re-render of generated footage.

Takeaway

Pick a few Claude concepts, not all 25

WHAT TO LEARN

Claude's expanding feature set boils down to choosing the right model and effort, giving it durable context about you, and letting it run on its own with a clear finish line.

02Part 1: The brain

Match the model to the job: Sonnet for clearly specified work, Opus for vague goals, and Fable only when the extra cost is justified.
Higher effort is not always better; medium is a sound daily default, and max can cost far more while overthinking.
Most tokens go to reading, so the files and instructions you load shape cost and quality more than prompt length.
Chat, Code and Cowork share the same usage limits, so where you spend them matters.

03Part 2: Where you use Claude

Claude Code works inside a folder on your computer and acts as an assistant that does the work, which makes it useful beyond developers.

05Part 4: What Claude knows

Memory is what Claude writes down and CLAUDE.md is what you write down; keep CLAUDE.md lean because it loads every time.
Turn any task you repeat into a skill so you stop re-explaining yourself, and chain skills for deliverables.
Connectors, plugins and mods extend Claude into your apps and into its own interface, and mods can be created by describing them.

06Part 5: Letting Claude work

Choose a permission mode deliberately; plan mode is a safe place to start before trusting auto.
Use subagents and /goal to offload messy research and long runs, as long as the finish line is unambiguous.

Glossary

Terms worth knowing.

Context window: The amount of text, measured in tokens, that Claude can hold in mind at once. Current Claude Code models have a one million token window.
Token: A chunk of a word. Tokens are the unit that both your cost and your usage limits are measured in.
Effort level: A setting for how hard Claude thinks before answering, from low through medium, high and extra high to max.
CLAUDE.md: A plain text file of your instructions and rules that Claude loads at the start of every session.
Skill: A saved set of instructions for one specific job, stored in a SKILL.md file and called with a slash command or picked up automatically.
Connector: A link between Claude and an app you already use, such as Gmail or Slack, often built on the Model Context Protocol (MCP).
Mod: A type of plugin that changes how Claude Code itself looks or behaves, such as adding a panel, command or status display.
Subagent: A helper Claude sends off to do one job in its own context window and report back a short summary.

Resources

Things they pointed at.

00:00productNate Herk free AI OS course (Skool community) ↗

08:00channelSix-hour Claude Code course for non-coders

13:10toolHyperframes (free open source video tool)

16:50toolModel Context Protocol (MCP)

Quotables

Lines you could clip.

02:40

“You wouldn't have your most senior person answering every quick email, and you wouldn't hand the whole project to your newest hire.”

clean analogy for picking a model→ TikTok hook↗ Tweet quote

18:10

“If you've ever done something twice, turn it into a skill.”

one-line rule, no setup needed→ IG reel cold open↗ Tweet quote

21:00

“You definitely don't need all 25 of these on day one. If you only set up three today, start with the CLAUDE.md, one skill, and a few /goal prompts.”

gives a concrete starting point→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

Okay, so over the last six weeks, Anthropic has shipped three new models, put docs, slides, and designs inside every single chat, launched dashboards and animated explainers, released a brand new Asian app called Duet, and the $200 plan doesn't even have a five -hour limit anymore. And people are kind of calling this Claude 2 .0, which is just obviously a nickname, not a new model.

And if you feel a little bit lost right now, that's probably pretty normal. So in this video, I'm going to break down every cloud concept that you need to understand, 25 of them in plain English, split across five parts. And you don't need to be technical at all or know how to code for any of it.

So let's get right into it. Okay, so part one is the brain, which is basically what cloud is actually running on. And concept number one is AI models.

Cloud isn't just one AI, it's a family of models. And right now there are three main. models.

Haiku is the small one, so it's cheapest and fastest. Sonnet is kind of the everyday workhorse. And Opus is the one that you hand the big, long, messy jobs to.

And right now they're all kind of called 5 .5 because that's the most recent version, but in the future we'll see 6 and we will see 7 and 8. Now we also do have another model that's even more intelligent and it's called Fable, but for right now you can say that the majority of work, even Opus is overkill. So like Sonnet 5 .5, Opus 5 .5, very, very good models.

But the way I think about them is like a team. You wouldn't have your most senior person answering every quick email, and you wouldn't hand the whole project to your newest hire. Now, I actually tested Sonnet against Opus on seven real workflows, and Sonnet won the stuff with really clear instructions like a resource guide where it could finish it in about half the time and usually cheaper than Opus.

But Opus won the tasks when the goal was vague, like building out a 90 -day plan. And keep in mind, on the free plan, you get Sonnet and Haiku, but not Opus. Now, the other one that I talked about, like I said...

Fable. Fable 5 .1 is Anthropic's top tier model. It's the one that the government actually took away when they first dropped Fable 5 because it was too powerful.

And they say that it's built for demanding reasoning and really, really long tasks where Claude works on its own. And it's also... very expensive.

If you pay per use, it costs two and a half times what Opus does. And Anthropix's own release notes say that Opus 5 .5 performs near Fable level on most work. So for most of our use cases, Opus is going to be plenty of power.

Now on Max and Ultra plans, which is the subscriptions that you get on, you know, 20 bucks a month, 100 bucks a month, or 200 bucks a month, Fable can use up to half of your weekly limit. And on Pro, you pay for it with usage credits. There's also a command called slash advisor that lets Claude check in with a stronger model like Fable at key moments during a task.

So you get some of the brainpower without running the entire job on Fable. All right, so concept number three now is tokens and the context window. A token is basically a chunk of word and it's how everything gets measured for AI models, both what you pay and how fast you hit your limits, whether that be the session limits or the weekly limits.

And then the context window is how many tokens Claude can hold in its head at one time. Every current model inside of Cloud Code has a 1 million token window, which Anthropix says is roughly 555 ,000 words. So like a whole stack of novels at once.

And most of your tokens go to reading, not writing. When I measured seven of my own tasks, Cloud read about 160 million tokens and only wrote about 800 ,000. So for every token it writes, it might be reading about 200, which is why the stuff you load in, like your files and your instructions, those matter so much more than how long your prompt is.

Okay, now number four, effort levels. Effort is how hard Claude has to think or how hard Claude does think before it answers. And the levels that you get to choose from are low, medium, high, extra high, and max.

Now on the 5 .5 models, the default is medium and you can change it by either typing slash effort or by in the bottom right in the desktop app, just choosing the effort level. But more effort isn't always better. I actually ran an experiment where I gave Opus 5 .5 the exact same prompt at every single effort level.

Lowe finished in about 17 minutes for under $4, and it was super glitchy. Extra High was the winner at an hour and a half of run, but it cost about $26. And Max, in this case, took two and a half hours, cost 50 bucks, and was not really that good.

It was still buggy. So Anthropic's own docs even say that Max is prone to overthinking. So you don't need to give it so much power.

So what I do myself is I keep medium as my everyday setting, and I only bump it up for really hard stuff like video editing. I'll always throw it on extra high. But if you're creating docs and you're doing general knowledge work, you could even just use Sonnet 5 .5 on medium for most things, especially when you combine that with skills, which we'll talk about later.

Now, there is also something called Ultra Code, which isn't really a level. It's more of a setting where Claude plans out the whole workflow for each big task before it starts. So that is something that you can pull another lever as well.

But anyways, number five is about usage limits and plans. So Cloud has two limits running at the same time, a five -hour session limit and a weekly limit, and they are shared. So your chats in the Cloud app and your work in Cloud Code all pull from that same bucket.

So for plans, free is $0 a month, pro is $20 a month, max is $100 for five times pro's usage, and then the $200 plan is now called Ultra. And Ultra gets you 10 times pro's weekly usage. And for verified Ultra accounts, there's no longer going to be a five -hour limit, which is pretty awesome.

Max and Ultra also come with $100 and $200 a month in API credits as well, which are for building your own apps on top of Cloud so they don't work inside the Cloud Code app. It's more about like programmatic automations that can sort of like run more in the background. Now, if you want to see where you're at, you can just type slash usage, or you can click on the little circle in the bottom right of the desktop app.

And a lot of what's in this video depends on your plan. So if you watch a video and you don't see a particular feature, then maybe it's because of the plan you're on, or maybe you need to be on a wait list or something like that. And I put together a free cheat sheet with all 25 of these concepts and exactly which plan gets what.

And that is inside my free school community, which is linked down in the description. So anyways, that's what Claude runs on. But where do you actually use it now that the app has three different tabs?

So number six is chat. Chat is the cloud that most people already know where you just type a message and you get an answer back. But since September 16th, chat can do a lot more than just talk.

You can make docs, slides, and designs right inside any conversation. And even the free plan gets web search, memory, connectors, and artifacts. And I'll explain all those things in a little bit.

So chat is where you could go to, you know, kind of do a Google search or think things through really quickly or make some quick drafts. But... Number seven is code, so Cloud Code.

The code tab in the desktop app is Cloud Code. And I know the name scares a lot of people off, but trust me, it's not scary at all. Cloud Code works inside a folder on your computer.

So it can read your local files. It can create new ones, edit them, move them around. It can run commands.

So it's less of a chatbot and more of an assistant that actually does the work. And it's not just for developers. I promise you, I don't come from a coding background and pretty much my entire business runs out of Cloud Code, out of one folder.

Everyone on my team, no one's a formal coder, but everyone uses Cloud Code and that's how we're able to move so much faster. It's literally like having a best friend that is the world's best coder that... talks to you like a normal person, and you can ask it to do things like a normal person.

It's also how I edit all my videos, like the one that you're watching right now. I recorded it, I handed Cloud the footage, and it helped me kind of cut out things and get the B -roll and throw it all together. Cloud Code comes with Pro and Up when we talk about subscriptions.

Now, number eight, co -work. Anthropic literally describes co -work as Cloud Code for the rest of your work, meaning all the stuff that isn't coding. You can give it a goal, it can work across your files and your tools, and you can come back to finished work that you can just review.

And their examples are things like reviewing a whole folder of contracts against a playbook or building a spreadsheet that reconciles a bunch of, you know, regional exports or reports or something like that. But honestly, the way I feel about it is why learn co -work if eventually you might need the power of code? So I would just recommend that you're learning Cloud Code and my channel has tons of great videos.

In fact, I'll tag a six -hour Cloud Code course for non -coders, for regular people, right up here so that you guys can check that out after this video if you need to. All right, so moving on to number nine, sessions. Every conversation in the Cloud Code tab is a session with its own history and its own folder.

And Anthropic rebuilt the desktop app around running a bunch of sessions at the same time. So you can have one session doing research, another one writing, and another one building all side by side. It's super, super slick.

You can also hit Command N for a new session, and you can open two of them next to each other in split view. Cloud can even message and manage your other sessions for you. It's really, really cool.

I've had days with 18 or 20 or 30 of these running in pairs just to test models against each other. And I actually built a little mod that warns me when two sessions are about to edit the same file. And with remote control, your computer shows up on your phone so you can check in on sessions from literally anywhere if you have your phone.

Now, number 10 is side chats. Let's say you're halfway through a big task and you want to ask a question without derailing that session. You can hit command semicolon, or you can type slash BTW for slash, by the way, and you can get a side chat and it can see everything in your session, but nothing that you say there gets added back into that main conversation, which is super nice.

Just keep in mind here that in the desktop app, it doesn't save your side chats. So if you close one from the app, it'll be gone. But that brings us on to number 11, which is Claude Duet, which is Anthropic's brand new agent app.

You message it the way you'd message a person, by typing or talking, and it does things in the apps that you've connected. And it's super, super proactive. It's basically it's Anthropic's response to, you know, Meta's Muse and OpenAI's Dots.

But every morning it can do things like scan your email and your calendar, it can flag what needs your attention, and it can start drafting replies for you to review right away. And throughout the day, it follows up on threads that have gone quiet. It's designed to ask you before it sends a message or makes any sort of payment or signs a document or deletes anything.

And right now it's in research preview with a wait list, US only on pro max and ultra plans. So that's part two. That's where you use Claude.

Now let's talk about what it can actually make for you. Okay. So number 12 here is artifacts.

An artifact is anything that Claude makes for you that opens up right beside the chat, like a document, a deck, a design, a dashboard, or a small interactive tool. really whatever you want it to be. And they start out private.

So nobody sees them unless you actually share them. And I actually made our October all hands deck for my team as an artifact. But I've also made like internal apps and actual like dashboards that I want to come back to and revisit.

I've made those artifacts as well. I've even made myself an artifact when I was planning a trip with my friends and I could send it to them as well. So very, very cool stuff.

Now, number 13 is kind of like 13, 14, 15, but we're just going to call it 13. And this is Docs, Slides, and Design. which are the three most common types of artifact.

So docs are documents, slides are presentations, and design is for visual stuff like landing pages and app mock -ups or something like that. As of October 8th, all three of these are out of beta and on every single plan, including free. And Anthropic says that people have already made over 45 million docs, decks, and designs.

So clearly people are loving this feature. Now I did do a full two -hour course on cloud design back in April, and the big change since then is that you don't have to go anywhere else to use it. It used to be a separate app.

But now design lives right inside your chats. And the old standalone site, cloud .ai slash design, is going to be closing in December of this year. So if you've got a bunch of workflows over there, just make sure you're kind of like aware of that or you're going to think that they have disappeared.

But anyways, that brings us on to number 14, which is... dashboards. You ask a question about your business in plain English, and Cloud pulls the answer from your data, it builds a dashboard, and it keeps it updated as the data changes.

It connects to things like BigQuery, Snowflake, Databricks, and Salesforce, and you can click on any number to see the exact query behind it. So you're not just trusting a chart that you can't check. Now dashboards right now are in beta on the paid plan, so if you don't see it, that's probably why.

But don't stress, everything eventually rolls out to everybody. Except for Mythos, we still haven't gotten that. But anyway, number 15 is Motion.

Motion turns a prompt into a short animated explainer that plays like a video, and then you can download it as an MP4. So it's basically clawed native video editing. Now, I myself typically use Hyperframes, which is a free open source tool for video editing, but all of this is getting really, really good.

And the key about this AI video editing is that it's not using a video generation model. The animation is actually built with code, so there's no AI generated footage or fake people in it. It's in beta right now on Team and Enterprise right now, but like I said, it will come out to everybody.

And this is pretty much the same idea that I use for my own videos, because a lot of the motion graphics on my channel are built with code, with cloud code as well, which is why they're so easy to change, because I can change the color of something or the size of something without having to re -render the entire like...

if it was an AI generated video. So Claude can do a ton of stuff, but it's only as good as what it knows about you, which is part four. So number 16 is memory.

Every new conversation starts completely fresh. Like when you shoot a new message off to Claude, it basically wakes up blind and it has to read things in your project in order to know who you are. And memory is how Claude carries things over from one session to the next or from one day to the next.

Now in the Claude app, memory is on every single plan, even free. And in Cloud Code, there's something called Auto Memory, where Cloud writes its own notes based on your corrections and preferences. And then it reads them at the start of every session.

So literally, the more you use Cloud Code, the smarter it gets about you and how you like to work. Now, I've got over 150 of those notes right now inside of my projects. One of them literally just says that Grokbot is spelled with a space because I corrected it enough times because it kept spelling Grokbot without a space.

So Auto Memory is, in fact, automatic. Now, number 17 is the Claude .md. So if memory is what Claude writes down, Claude .md is what you write down.

It's just a text file with your instructions and your rules. And Claude loads it at the start of every session. So mine basically starts with, hey, you are Nate Hurk's AI operating system.

Your job is to save him time, blah, blah, blah. And then it's basically just a map of where everything in my business lives. So if you want business context, you go to this folder.

If you want, you know, brand assets, you go to this folder and it's just a routing map of where it can find all my stuff. Now, one quick tip is that you want to keep this lean because it loads every single time. Remember how I said, every time you shoot off a message to cloud, it wakes up fresh.

It will read that cloud .md every time. So the recommendation is really to keep that within 200 lines. Now, number 18 is probably my favorite Claude Code feature ever, and that is skills.

A skill is a saved set of instructions for one specific job written in a file called skill .md. And whenever you hear this .md thing, that just means markdown file. And markdown is just a type of computer language that's very, very human readable.

It's super easy to read, but it just gives computer more structure when it's reading something. So you've got like, you know, section headers and bullet points and different sorts of like bold, things like that.

whenever it identifies that it needs to. Or you can call one yourself by typing a slash and then the skill's name. And skills are super, super cheap to keep around because Claude only sees a short description of each skill until it actually needs it.

Now, I myself have over 100 skills and the way that most of them get made is really, really simple. If I've typed the same request over and over, turn it into a skill. I typed process this YouTube video into the Google sheet, master database, and the wiki.

I've done that probably 15 times. And then I realized, oh, why don't I just turn this into a skill? And now I have one called process video.

And a skill can be anything. It can literally be two prompts or it could be an entire workflow that runs over the course of an hour. It's just when you turn some sort of process that already exists into a reference or an instruction guide that Claude can then follow for you.

So if you've ever done something twice, turn it into a skill. If you ever have deliverables that you need to create, turn that into a skill. For example, once I make a YouTube video, what I do is I run my process video skill, and then I run my LinkedIn creation skill, and then I run my X article skill, and then I run my upload skill, and then I run my packaging skills.

So I've turned all of these things into simple slash commands that I can ask Claude for instead of having to re -explain myself. So start building as many skills as you can. And that brings us on to number 19, which is Connectors.

Connectors plug Cloud or connect Cloud into the apps that you already use like Gmail, Google Calendar, Drive, Slack, ClickUp, whatever you need. So it can actually read and do things in there with hands rather than just having you copy and paste everything back and forth. It's really the big difference between an agent and just a chat bot.

Now under the hood, a lot of these run on something called MCP, which stands for Model Context Protocol. You don't really, really need to know what that means. It's sort of similar to an API or an application programming interface.

It just lets two softwares talk to each other. So Claude and Gmail or Claude and Fireflies. Connectors also exist on the free plan too.

And my own command center pulls in my ClickUp, my Slack, and my Gmail through these connectors. Now, number 20 is plugins. A plugin is a bundle that you install in one go.

And it can include skills, commands, connectors, agents, all of these things packaged together. And by the way, in my free school community that I mentioned earlier, I've got tons of free plugins and skills and docs, tons of free stuff in there. But Anthropic now has a directory where people can also submit other plugins and companies are shipping their own like Salesforce, which has a sales plugin in beta and your own company can use the same plugins together.

So everyone's kind of creating the same sort of docs and resources. Like I have a plugin for my team called AIS branded doc. Whenever anyone's making a document that might be externally facing, they have to run that skill so that all of our docs feel consistent and feel on brand to anything external.

Now, number 21 is mods and it's. One of my favorite things in here, it's kind of recent and it's so, so cool. A mod is a type of plugin that literally modifies or changes or hacks how cloud code itself works.

You can add new commands, you can add new tools, you can even customize the interface, or you can change just little pieces of the interface, like a panel beside your chat, or you can add a little pet in the top right. And you don't have to code one. Anthropix docs say...

that you can literally just describe the mod you want, and Claude wires it in. So if you've ever been like, I wish the Claude desktop app had this, just ask it to do it, and it will build a mod for it. Now I've got seven of these myself.

One of them saves me money by keeping my session cached, so Claude doesn't have to reread everything when I come back from a break. One other one shows a progress bar whenever Claude's working towards some sort of goal, and it shows me the drill down of all the tasks. I've also got a little pixel cat that I renamed Byte, and just hangs out on my screen and tells me when other sessions are done.

So really get creative with those mods. They're pretty fun, but they can also be super, super, super useful too. Like the one I mentioned earlier that makes sure that two sessions don't overwrite the same file.

But anyways, now Claude knows who you are. The last part is about letting it actually work on its own. So number 22 is permission modes, which control how much Claude asks you before it does something.

And there are five of them. Manual asks before it edits files or runs commands. Accept edits lets it change files, but still asks before most commands.

Plan has it look around and propose a plan without changing anything. Auto mode runs without the routine prompts, but a background check makes sure that each action lines up with what you actually asked for. And then bypass permissions or dangerously skip permissions skips almost all the prompts.

Now for me, I always just leave auto on, and this has been the default for new sessions on pro, max, and team plans since August. And if you're nervous, then just start in plan mode, read the plan, and then let it go. But auto mode seems to do the trick for most cases, and it always will take the cautious side.

You'd rather be safe than sorry. And then number 23, we have subagents. Now, a subagent is a helper that Claude can send off to do one specific job.

And it works in its own separate context window, if you remember when we talked about context windows earlier. And it only reports back a small summary to the main session, so it helps us save those tokens. So all the messy research and searching doesn't clog up your main conversation.

Now I have this skill called roast room, which you guys can get if you want in my community, which is built on this idea of sub -agents. Basically, I ask it a business question and each seat on the board, like Alex Ramosi, Jeff Bezos, or Warren Buffett is its own custom sub -agent that I built. And all of these custom sub -agents argue about my decision before they give me a verdict and before they give me some business advice.

Now, number 24 is projects. And this one's a little confusing because the word means two different things right now. In the Cloud app, a project is kind of like a folder for related chats with its own files and instructions.

But in Cloud Code, a project is one ongoing conversation or like a folder that you have locally where Cloud coordinates a whole system of related work for you. You tell it what needs doing and it starts a separate thread for each task. And those threads usually run in the cloud and in parallel.

So they keep going after you close your laptop and you can check on them from your phone. Now, the difference from subagents is that a subagent is a quick helper inside one session, while a project can run a full session for every task. So Cloud Code projects are in beta right now on Pro Max and Ultra, and they're still rolling out.

So if you don't see them yet, that's why. But once again, they get rolled out to everybody. And that brings us on to number 25, which is slash goal.

Normally, Cloud does one turn, and then will kind of stop and wait for you to think. But with slash goal, you give it a finish line, and it keeps working turn after turn until that condition is actually met. So just make sure the finish line is really, really clear.

So there's no question about when it's done, because then you might get stuck in this loop where it's like, I can't clear the goal. I can't clear the goal. And it just keeps saying that, but this has really become how I use Claude.

It's about 20 % of everything that I sent. And a lot of the research behind this video ran on a single slash goal. And I've got one bonus for you guys, which is number 26, which is scheduled tasks.

This is how Claude works. Even when you're not there in the desktop app, you can literally set up a scheduled task to run on a schedule like every Monday or every morning. And routines are basically scheduled tasks, but they just run in the cloud.

And these can once again be triggered on a schedule or by events from your other tools. I've got one that updates my stats on the website every Sunday and another one that posts carousels every single four hours. And the cool thing about scheduled tasks is you can also just create them with natural language.

Hey, Claude, build me a scheduled task for this time. And when the scheduled task goes off, you run this skill and then this skill, and it literally will just set it up for you. So you definitely don't need all 25 of these.

on day one. If you only set up three today, I'd start with the cloud .md, one skill for the thing that you do most, and then try out a few slash goal prompts. And then once you've started getting the hang of this, what I really want you guys to do next is build your AI operating system.

So if you want to check out a full free course where I show you guys exactly how to build out an AI operating system with cloud code, then join the link in the description, which is my free school community. I've got the full free course in there for you guys with a free plugin that helps you get set up. But that is going to do it for today.

So I hope that you guys enjoyed the video or you learned something new. And if you did, please give it a like. It definitely helps me out a ton.

And as always, I appreciate you guys making it to the end of the video. And I will see you all in the next one. Thanks, everyone.

The Hook

The bait, then the rug-pull.

Six weeks of Claude releases landed so fast that people are calling it Claude 2.0. Nate Herk slows it down into 25 concepts anyone can follow, no code required.

Frameworks