YT YouTube2Text

YouTube2Text → Transcripts → Every Claude 2.0 Concept Explained for Normal People

Every Claude 2.0 Concept Explained for Normal People — Transcript

by Nate Herk | AI Automation · 5,189 words · 715 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Okay, so over the last 12 weeks,
0:01Anthropic has shipped three new models,
0:03put docs, slides, and designs inside
0:06every single chat, launched dashboards
0:08and animated explainers, released a
0:09brand new agent app called Duet, and the
0:11$200 plan doesn't even have a 5-hour
0:14limit anymore. And people are kind of
0:16calling this Claude 2.0, which is just
0:18obviously a nickname, not a new model.
0:19And if you feel a little bit lost right
0:21now, that's probably pretty normal. So
0:22in this video, I'm going to break down
0:23every Claude concept that you need to
0:25understand, 25 of them in plain English,
0:27split across five parts. And you don't
0:30need to be technical at all or know how
0:31to code for any of it. So let's get
0:33right into it. Okay, so part one is the
0:35brain, which is basically what Claude is
0:37actually running on. And concept number
0:39one is AI models. Claude isn't just one
0:42AI. It's a family of models. And right
0:44now, there are three main models. Haiku
0:47is the small one, so it's cheapest and
0:49fastest. Sonnet is kind of the everyday
0:51workhorse. And Opus is the one that you
0:53hand the big, long, messy jobs to. And
0:55right now, they're all kind of called
0:575.5 because that's the most recent
0:59version, but in the future, we'll see
1:00six and we will see seven and eight.
1:03Now, we also do have another model
1:04that's even more intelligent and it's
1:06called Fable, but for right now, you can
1:08say that the majority of work, even Opus
1:10is overkill. So like Sonnet 5.5, Opus
1:125.5, very, very good models. But the way
1:14I think about them is like a team. You
1:16wouldn't have your most senior person
1:18answering every quick email, and you
1:20wouldn't hand the whole project to your
1:21newest hire. Now, I actually tested
1:23Sonnet against Opus on seven real
1:24workflows, and Sonnet won the stuff with
1:27really clear instructions like a
1:28resource guide where I could finish it
1:30in about half the time and usually
1:32cheaper than Opus. But Opus won the
1:34tasks when the goal was vague, like
1:36building out a 90-day plan. And keep in
1:37mind, on the free plan, you get Sonnet
1:39and Haiku, but not Opus. Now, the other
1:41one that I talked about, like I said,
1:43Fable. Fable 5.1 is Anthropic's top-tier
1:46model. It's the one that the government
1:48actually took away when they first
1:49dropped Fable 5
1:50because it was too powerful. And they
1:52say that it's built for demanding
1:53reasoning and really, really long tasks
1:55where Claude works on its own. And it's
1:57also very expensive. If you pay per use,
2:00it costs two and a half times what Opus
2:02does. And Anthropic's own release notes
2:04say that Opus 5.5 performs near fable
2:06level on most work. So for most of our
2:09use cases, Opus is going to be plenty of
2:11power. Now on Max and Ultra plans, which
2:13is the subscriptions that you get on,
2:14you know, 20 bucks a month, 100 bucks a
2:16month, or 200 bucks a month, Fable can
2:18use up to half of your weekly limit. And
2:20on Pro, you pay for it with usage
2:22credits. There's also a command called
2:24slash advisor that lets Claude check in
2:26with a stronger model like Fable at key
2:28moments during a task. So you get some
2:30of the brain power without running the
2:31entire job on
2:33Fable. All right, so concept number
2:35three now is tokens and the context
2:36window. A token is basically a chunk of
2:39word and it's how everything gets
2:40measured for AI models, both what you
2:42pay and how fast you hit your limits,
2:44whether that be the session limits or
2:46the weekly limits. And then the context
2:48window is how many tokens Claude can
2:50hold in its head at one time. Every
2:52current model inside of Claude Code has
2:54a 1 million token window, which
2:56Anthropic says is roughly 555,000 words.
2:59So like a whole stack of novels at once.
3:01And most of your tokens go to reading,
3:03not writing. When I measured seven of my
3:05own tasks, Claude read about 160 million
3:08tokens and only wrote about 800,000. So
3:11for every token it writes, it might be
3:12reading about 200, which is why the
3:14stuff you load in, like your files and
3:16your instructions, those matter so much
3:18more than how long your prompt is. Okay,
3:21now number four, effort levels. Effort
3:23is how hard Claude has to think or how
3:25hard Claude does think before it
3:27answers. And the levels that you get to
3:28choose from are low, medium, high, extra
3:31high, and max. Now on the 5.5 models,
3:33the default is medium. And you can
3:35change it by either typing slash effort
3:38or by in the bottom right in the desktop
3:39app just choosing the effort level. But
3:42more effort isn't always better. I
3:44actually ran an experiment where I gave
3:45Opus 5.5 the exact same prompt at every
3:48single effort level. Low finished in
3:50about 17 minutes for under $4 and it was
3:52super glitchy. Extra high was the winner
3:54at an hour and a half of run, but it
3:56cost about $26. And Max in this case
3:59took two and a half hours, cost 50
4:01bucks, and was not really that good. It
4:03was still buggy. So in Tropics own Docs
4:05even say that max is prone to
4:07overthinking. So you don't need to give
4:09it so much power. So what I do myself is
4:12I keep medium as my everyday setting,
4:14and I only bump it up for really hard
4:16stuff like video editing. I'll always
4:18throw it on extra high. But if you're
4:19creating Docs and you're doing general
4:21knowledge work, you could even just use
4:22Sonic 5.5 on medium for most things,
4:24especially when you combine that with
4:25skills, which we'll talk about later.
4:27Now there is also something called Ultra
4:28Code, which isn't really a level. It's
4:30more of a setting where Claude plans out
4:31the whole workflow for each big task
4:33before it starts. So that is something
4:35that you can pull another lever as well.
4:37But anyways, number five is about usage
4:38limits and plans. So Claude has two
4:41limits running at the same time, a
4:425-hour session limit and a weekly limit,
4:45and they are shared. So your chats in
4:46the Claude app and your work in Claude
4:48Code all pull from that same bucket. So
4:50for plans, free is $0 a month, Pro is
4:52$20 a month, Max is $100 for five times
4:55Pro's usage, and then the $200 plan is
4:57now called Ultra. And Ultra gets you 10
5:00times Pro's weekly usage. And for
5:03verified Ultra accounts, there's no
5:04longer going to be a 5-hour limit, which
5:06is pretty awesome. Max and Ultra also
5:08come with $100 and $200 a month in API
5:11credits as well, which are for building
5:12your own apps on top of Claude so they
5:14don't work inside the Claude Code app.
5:17It's more about like programmatic
5:18automations that can sort of like run
5:20more in the background. Now if you want
5:21to see where you're at, you can just
5:22type /usage or you can click on the
5:24little circle in the bottom right of the
5:25desktop app, and a lot of what's in this
5:27video depends on your plan. So if you
5:29watch a video and you don't see a
5:30particular feature, then maybe it's
5:32because of the plan you're on or maybe
5:34you need to be on a weight list or
5:35something like that. And I put together
5:36a free cheat sheet with all 25 of these
5:38concepts and exactly which plan gets
5:40what, and that is inside my free school
5:42community, which is linked down in the
5:43description. So anyways, that's what
5:45Claude runs on. But where do you
5:47actually use it now that the app has
5:48three different tabs? [music] So number
5:50six is chat. Chat is the Claude that
5:52most people already know where you just
5:54type a message and you get an answer
5:55back. But since September 16th, chat can
5:58do a lot more than just talk. You can
5:59make Docs, slides, and designs right
6:01inside any conversation, and even the
6:03free plan gets web search, memory,
6:05connectors, and artifacts. And I'll
6:07explain all those things in a little
6:08bit. So chat is where you could go to,
6:10you know, kind of do a Google search or
6:11think things through really quickly or
6:13make some quick drafts. But number seven
6:16is code. So Claude code. The code tab in
6:18the desktop app is Claude code. And I
6:21know the name scares a lot of people
6:22off, but trust me, it's not scary at
6:23all. Claude code works inside a folder
6:25on your computer. So it can read your
6:27local files. It can create new ones,
6:29edit them, move them around. It can run
6:31commands. So it's less of a chatbot and
6:33more of an assistant that actually does
6:34the work. And it's not just for
6:36developers. I promise you. I don't come
6:37from a coding background, and pretty
6:38much my entire business runs out of
6:40Claude code out of one folder. Everyone
6:42on my team, no one's a formal coder, but
6:44everyone uses Claude code, and that's
6:45how we're able to move so much faster.
6:47It's literally like having a best friend
6:49that is the world's best coder that
6:51talks to you like a normal person, and
6:52you can ask it to do things like a
6:53normal person. It's also how I edit all
6:55my videos, like the one that I'm you're
6:57watching right now. I recorded it, I
6:58handed Claude the footage, and it helped
7:00me kind of cut out things and get the
7:02B-roll and throw it all together. Claude
7:03code comes with Pro and up when we talk
7:06about subscriptions. Now number eight,
7:08Cowork. Anthropic literally describes
7:10Cowork as Claude code for the rest of
7:12your work, meaning all the stuff that
7:13isn't coding. You can give it a goal, it
7:15can work across your files and your
7:16tools, and you can come back to finished
7:18work that you can just review. And their
7:19examples are things like reviewing a
7:21whole folder of contracts against a
7:22playbook or building a spreadsheet that
7:24reconciles a bunch of, you know,
7:26regional exports or reports or something
7:27like that. But honestly, the way I feel
7:29about it is why learn Cowork if
7:31eventually you might need the power of
7:32code? So I would just recommend that
7:33you're learning Claude code, and my
7:35channel has tons of great videos. In
7:37fact, I'll tag a six-hour Claude code
7:38course for non-coders, for regular
7:40people right up here so that you guys
7:42can check that out after this video if
7:43you need to. All right, so moving on to
7:44number nine, sessions. Every
7:47conversation in the Claude code tab is a
7:48session with its own history and its own
7:51folder. And Anthropic rebuilt the
7:52desktop app around running a bunch of
7:54sessions at the same time. So you can
7:56have one session doing research, another
7:57one writing, and another one building
7:59all side by side. It's super super
8:01slick. You can also hit command N for a
8:03new session, and you can open two of
8:04them next to each each in split view.
8:06Claude can even message and manage your
8:08other sessions for you. It's really
8:09really cool. I've had days with 18 or 20
8:12or 30 of these running in pairs just to
8:14test models against each other and
8:16actually built a little mod that warns
8:18me when two sessions are about to edit
8:19the same file. And with remote control,
8:22your computer shows up on your phone so
8:23you can check in on sessions from
8:25literally anywhere if you have your
8:26phone. Now number 10 is side chats.
8:29Let's say you're halfway through a big
8:30task and you want to ask a question
8:32without derailing that session. You can
8:34hit command semicolon or you can type /
8:37b t w for / by the way and you can get a
8:39side chat and you can see everything in
8:41your session but nothing that you say
8:43there gets added back into that main
8:44conversation which is super nice. Just
8:46keep in mind here that in the desktop
8:47app, it doesn't save your side chats so
8:49if you close one from the app, it'll be
8:51gone. But that brings us on to number 11
8:53which is Claude Duet which is
8:55Anthropic's brand new agent app. You
8:57message it the way you'd message a
8:58person by typing or talking and it does
9:00things in the apps that you've connected
9:02and it's super super proactive. It's
9:04basically it's Anthropic's response to
9:06you know Meta's Muse and OpenAI's Dots.
9:09But every morning it can do things like
9:10scan your email and your calendar. It
9:12can flag what needs your attention and
9:13it can start drafting replies for you to
9:14review right away. And throughout the
9:16day, it follows up on threads that have
9:18gone quiet. It's designed to ask you
9:20before it sends a message or makes any
9:22sort of payment or signs a document or
9:24deletes anything. And right now it's in
9:25research preview with a wait list US
9:27only on Pro Max and Ultra plans. So
9:31that's part two. That's where you use
9:33Claude. Now let's talk about what it can
9:35actually make for you. Okay, so number
9:3712 here is artifacts. [music]
9:39An artifact is anything that Claude
9:40makes for you that opens up right beside
9:42the chat like a document, a deck, a
9:43design, a dashboard, or a small
9:45interactive tool, really whatever you
9:46want it to be. And they start out
9:48private so nobody sees them unless you
9:50actually share them. And I actually made
9:51our October All Hands deck for my team
9:53as an artifact but I've also made like
9:56internal apps and actual like dashboards
9:58that I want to come back to and revisit.
9:59I've made those artifacts as well. I've
10:01even made myself an artifact when I was
10:02planning a trip with my friends and I
10:04could send it to them as well. So very
10:06very cool stuff. Now number 13 is kind
10:08of like 13 14 15 but we're just going to
10:10call it 13. And this is docs, slides,
10:12and design, which are the three most
10:14common types of artifact. So docs are
10:16documents, slides are presentations, and
10:18design is for visual stuff like landing
10:20pages and, you know, app mockups or
10:22something like that. As of October 8th,
10:24all three of these are out of beta and
10:26on every single plan, including free.
10:28And it topic says that people have
10:29already made over 45 million docs,
10:31decks, and designs. So clearly people
10:33are loving this feature. Now I did do a
10:35full two-hour course on Claude Design
10:36back in April, and the big change since
10:38then is that you don't have to go
10:39anywhere else to use it. It used to be a
10:41separate app, but now design lives right
10:43inside your chats. And the old
10:44standalone site, claude.ai/design,
10:47is going to be closing in December of
10:49this year. So if you've got a bunch of
10:50workflows over there, just make sure
10:51you're kind of like aware of that or
10:53you're going to think that they have
10:54disappeared. But anyways, that brings us
10:55on to number 14, which is dashboards.
10:58You ask a question about your business
10:59in plain English, and Claude pulls the
11:01answer from your data. It builds a
11:02dashboard, and it keeps it updated as
11:04the data changes. It connects to things
11:06like BigQuery, Snowflake, Databricks,
11:08and Salesforce, and you can click on any
11:10number to see the exact query behind it.
11:12So you're not just trusting a chart that
11:14you can't check. Now dashboards right
11:15now are in beta on the paid plan, so if
11:17you don't see it, that's probably why.
11:19But don't stress, everything eventually
11:20rolls out to everybody, except for
11:22Mythos. We still haven't gotten that.
11:23But anyway, number 15 is motion. Motion
11:26turns a prompt into a short animated
11:28explainer that plays like a video, and
11:30then you can download it as an MP4. So
11:32it's basically Claude native video
11:34editing. Now I myself typically use
11:36Hyper Frames, which is a free open
11:37source tool for video editing, but all
11:39of this is getting really, really good.
11:40And the key about this AI video editing
11:42is that it's not using a video
11:43generation model. The animation is
11:45actually built with code, so there's no
11:46AI generated footage or fake people in
11:48it. It's in beta right now on Team and
11:50Enterprise right now, but like I said,
11:52it will come out to everybody. And this
11:53is pretty much the same idea that I use
11:55for my own videos because a lot of the
11:56motion graphics on my channel are built
11:58with code, with Claude code as well,
12:00which is why they're so easy to change
12:01because I can change the color of
12:02something or the size of something
12:04without having to re-render the entire
12:06like if it was an AI generated video. So
12:08Claude can do a ton of stuff, but it's
12:11only as good as what it about you, which
12:13is part four. [music] So, number 16 is
12:16memory. Every new conversation starts
12:18completely fresh. Like when you shoot a
12:20new message off to Claude, it basically
12:21wakes up blind and it has to read things
12:24in your project in order to know who you
12:25are. And memory is how Claude carries
12:27things over from one session to the next
12:29or from one day to the next. Now, in the
12:31Claude app, memory is on every single
12:33plan, even free. And in Claude Code,
12:35there's something called auto memory
12:37where Claude writes its own notes based
12:39on your corrections and preferences and
12:41then it reads them at the start of every
12:42session. So, literally the more you use
12:44Claude Code, the smarter it gets about
12:46you and how you like to work. Now, I've
12:47got over 150 of those notes right now
12:50inside of my projects. One of them
12:52literally just says that Grokbot is
12:53spelled with a space because I corrected
12:55it enough times because it kept spelling
12:57Grokbot without a space. So, auto memory
12:59is in fact automatic. Now, number 17 is
13:02the Claude.md.
13:04So, if memory is what Claude writes
13:05down, Claude.md is what you write down.
13:08It's just a text file with your
13:09instructions and your rules and Claude
13:11loads it at the start of every session.
13:13So, mine basically starts with, hey, you
13:14are Nate Herks, AI operating system.
13:16Your job is to save him time, blah blah
13:18blah. And then it's basically just a map
13:20of where everything in my business
13:21lives. So, if you want business
13:23contacts, you go to this folder. If you
13:24want, you know, brand assets, you go to
13:26this folder. And it's just a routing map
13:27of where I can find all my stuff. Now,
13:29one quick tip is that you want to keep
13:31this lean because it loads every single
13:33time. Remember how I said every time you
13:34shoot off a message to Claude, it wakes
13:35up fresh? It will read that Claude.md
13:37every time. So, the recommendation is
13:39really to keep that within 200 lines.
13:42Now, number 18 is probably my favorite
13:43Claude Code feature ever and that is
13:45skills. A skill is a saved set of
13:48instructions for one specific job
13:50written in a file called skill.md. And
13:52whenever you hear this .md thing, that
13:54just means markdown file. And markdown
13:56is just a type of computer language
13:58that's very very human readable. It's
14:00super easy to read, but it just gives
14:01computer more structure when it's
14:03reading something. So, you've got like,
14:04you know, section headers and bullet
14:05points and different sorts of like like
14:07bold, things like that. So, Claude will
14:09basically use a relevant skill whenever
14:12it identifies that it needs to or you
14:14can call one yourself by typing a slash
14:17and then the skills name. And skills are
14:18super, super cheap to keep around
14:20because Claude only sees a short
14:21description of each skill until it
14:23actually needs it. Now I myself have
14:25over 100 skills and the way that most of
14:27them get made is really, really simple.
14:29If I've typed the same request over and
14:31over, turn it into a skill. I typed
14:33process this YouTube video into the
14:35Google Sheet master database and the
14:37wiki. I've done that probably 15 times
14:39and then I realized, oh, why don't I
14:40just turn this into a skill? And now I
14:41have one called process video. And a
14:43skill can be anything. It can literally
14:44be two prompts or it could be an entire
14:46workflow that runs over the course of an
14:48hour. It's just when you turn some sort
14:49of process that already exists into a
14:53reference or an instruction guide that
14:54Claude can then follow for you. So if
14:56you've ever done something twice, turn
14:58it into a skill. If you ever have
14:59deliverables that you need to create,
15:01turn that into a skill. For example,
15:02once I make a YouTube video, what I do
15:04is I run my process video skill and then
15:06I run my LinkedIn creation skill and
15:07then I run my X article skill and then I
15:09run my upload skill and then I run my
15:12packaging skills. So I've turned all of
15:14these things into simple slash commands
15:15that I can ask Claude for instead of
15:17having to re-explain myself. So start
15:19building as many skills as you can. And
15:21that brings us on to number 19 which is
15:23connectors. [music]
15:24Connectors plug Claude or connect Claude
15:26into the apps that you already use like
15:27Gmail, Google Calendar, Drive, Slack,
15:29ClickUp, whatever you need. So it can
15:32actually read and do things in there
15:33with hands rather than just having you
15:35copy and paste everything back and
15:36forth. It's really the big difference
15:38between an agent and just a chatbot. Now
15:40under the hood, a lot of these run on
15:42something called MCP which stands for
15:44model context protocol. You don't
15:46really, really need to know what that
15:47means. It's sort of similar to an API or
15:49an application programming interface. It
15:51just lets two softwares talk to each
15:52other. So Claude and Gmail or Claude and
15:55Fireflies. Connectors also exist on the
15:57free plan too and my own command center
15:59pulls in my ClickUp, my Slack, and my
16:01Gmail through these connectors. Now
16:03number 20 is plugins. A plugin is a
16:05bundle that you install in one go and it
16:07can include skills, commands,
16:09connectors, agents, all of these things
16:11packaged together. And by the way, in my
16:12free school community that I mentioned
16:14earlier, I've got tons of free plugins
16:15and skills and docs, tons of free stuff
16:17in there. But Anthropic has a directory
16:19where people can also submit other
16:20plugins and companies are shipping their
16:21own, like Salesforce which has a sales
16:23plugin and beta. And your own company
16:25can use the same plugins together, so
16:26everyone's kind of creating the same
16:28sort of docs and resources. Like I have
16:30a plugin for my team called AIS branded
16:32doc. Whenever anyone's making a document
16:34that might be externally facing, they
16:35have to run that skill so that all of
16:37our docs feel consistent and feel on
16:39brand to anything external. Now number
16:4121 is mods. And it's one of my favorite
16:44things in here. It's kind of recent and
16:45it's so so cool. A mod is a type of
16:47plugin that literally modifies or
16:49changes or hacks how Claude code itself
16:52works. You can add new commands, you can
16:54add new tools, you can even customize
16:55the interface or you can change just
16:57little pieces of the interface, like a
16:58panel beside your chat or you can add a
17:00little pet in the top right. And you
17:02don't have to code one. Anthropic's docs
17:03say that you can literally just describe
17:05the mod you want and Claude wires it in.
17:07So if you've ever been like I wish the
17:08Claude desktop app had this, just ask it
17:10to do it and it will build a mod for it.
17:12Now I've got seven of these myself. One
17:14of them saves me money by keeping my
17:16session cached so Claude doesn't have to
17:18re-read everything when I come back from
17:19[music] a break. One other one shows a
17:21progress bar whenever Claude's working
17:22towards some sort of goal and it shows
17:24me the drill down of all the tasks. I've
17:26also got a little pixel cat that I
17:27renamed Bite and just hangs out on my
17:29screen and tells me when other sessions
17:30are done. So really get creative with
17:32those mods. They're pretty fun but they
17:33can also be super super super useful
17:35too. Like the one I mentioned earlier
17:37that make sure that two sessions don't
17:38overwrite the same file. But anyways,
17:40now Claude knows who you are. The last
17:42part is about letting it actually work
17:44on its own. So number 22 is permission
17:46modes which control how much Claude asks
17:48you before it does something. And there
17:50are five of them. Manual asks before it
17:52edits files or runs commands. Accept
17:55edits lets it change files but still
17:57asks before most commands. Plan has it
18:00look around and propose a plan without
18:01changing anything. Auto mode runs
18:03without the routine prompts but a
18:04background check makes sure that each
18:06action lines up with what you actually
18:07asked for. And then bypass permissions
18:09or dangerously skip permissions skips
18:12almost all the prompts. Now for me, I
18:14always just leave auto on and this has
18:16been the default for new sessions on
18:17Pro, Max, and Team plans since August.
18:20And if you're nervous, then just start
18:21in plan mode, read the plan, and then
18:23let it go. But auto mode seems to do the
18:25trick for most cases, and it always will
18:27take the cautious side. You'd rather be
18:29safe than sorry. And then number 23 we
18:31have sub agents. Now a sub agent is a
18:33helper that Claude can send off to do
18:35one specific job, and it works in its
18:36own separate context window, if you
18:38remember when we talked about context
18:39windows earlier, and it only reports
18:40back a small summary to the main
18:42session, so it helps us save those
18:44tokens. So all the messy research and
18:45searching doesn't clog up your main
18:47conversation. Now I have this skill
18:49called roast room, which you guys can
18:50get if you want in my community, which
18:52is built on this idea of sub agents.
18:54Basically, I ask it a business question,
18:56and each seat on the board, like Alex
18:58Hormozi, Jeff Bezos, or Warren Buffett,
19:00is its own custom sub agent that I
19:02built. And all of these custom sub
19:03agents argue about my decision before
19:06they give me a verdict and before they
19:08give me some business advice. Now number
19:1024 is projects, and this one's a little
19:12confusing because the word means two
19:13different things right now. In the
19:15Claude app, a project is kind of like a
19:17folder for related chats with its own
19:19files and instructions. But in Claude
19:20code, a project is one ongoing
19:22conversation or like a folder that you
19:24have locally where Claude coordinates a
19:26whole system of related work for you.
19:28You tell it what needs doing and it
19:29starts a separate thread for each task,
19:31and those threads usually run in the
19:32cloud and in parallel, so they keep
19:34going after you close your laptop and
19:36you can check on them from your phone.
19:38Now the difference from sub agents is
19:39that a sub agent is a quick helper
19:40inside one session, while a project can
19:43run a full session for every task. So
19:45Claude code projects are in beta right
19:46now on Pro Max and Ultra and they're
19:48still rolling out, so if you don't see
19:49them yet, that's why. But once again,
19:51they get rolled out to everybody. And
19:52that brings us on to number 25 which is
19:54slash goal. Normally, Claude does one
19:57turn and then we'll kind of stop and
19:58wait for you to think. But with slash
20:00goal, you give it a finish line and it
20:02keeps working turn after turn until that
20:04condition is actually met. So just make
20:06sure the finish line is really really
20:08clear, so there's no question about when
20:09it's done because then you might get
20:11stuck in this loop where it's like I
20:12can't clear the goal, I can't clear the
20:13goal, and it just keeps saying that. But
20:15this has really become how I use Claude.
20:16It's about 20% of everything that I send
20:18and a lot of the research behind this
20:20video ran on a single slash goal. And
20:22I've got one bonus for you guys, which
20:23is number 26, which is scheduled tasks.
20:26This is how Claude works even when
20:28you're not there. In the desktop app,
20:29you can literally set up a scheduled
20:30task to run on a schedule like every
20:32Monday or every morning. And routines
20:34are basically scheduled tasks, but they
20:36just run in the cloud. And these can
20:37once again be triggered on a schedule or
20:39by events from your other tools. I've
20:41got one that updates my stats on the
20:42website every Sunday and another one
20:44that posts carousels every single 4
20:45hours. And the cool thing about
20:47scheduled tasks is you can also just
20:48create them with natural language. Hey
20:50Claude, build me a scheduled task for
20:51this time. And when the scheduled task
20:53goes off, you run this skill and then
20:56this skill. And it literally will just
20:57set it up for you. So you definitely
20:59don't need all 25 of these on day one.
21:01If you only set up three today, I'd
21:04start with the Claude MD, one skill for
21:06the thing that you do most, and then try
21:07out a few slash goal prompts. And then
21:09once you've started getting the hang of
21:10this, what I really want you guys to do
21:11next is build your AI operating system.
21:14So if you want to check out a full free
21:15course where I show you guys exactly how
21:18to build out an AI operating system with
21:19Claude Code, then join the link in the
21:21description, which is my free school
21:22community. I've got the full free course
21:24in there for you guys with a free plugin
21:27that helps you get set up. But that is
21:28going to do it for today. So I hope that
21:29you guys enjoyed the video or you
21:30learned something new. And if you did,
21:32please give it a like. It definitely
21:33helps me out a ton. And as always, I
21:34appreciate you guys making it to the end
21:35of the video. And I'll see you all in
21:37the next one. Thanks everyone.

About this transcript

This page contains the full transcript of Every Claude 2.0 Concept Explained for Normal People by Nate Herk | AI Automation, generated from the public captions YouTube serves with the video. The transcript has 5,189 words across 715 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.