The creator of Claude Code just revealed

seven secrets to get the most out of

Claude Code and Opus 47. In this video,

we'll not only go over every single

secret he revealed, but also demonstrate

them live as we build an app together.

By the end of this video, if you follow

along, you'll not only be a master of

Claude Code and Opus 47, but you also

have an entire amazing app that you'll

build out side by side with me, even if

you've never coded before a day in your

life. Let's lock in and get into it. So,

Boris Cherny, one of the creators of

Claude Code, just put out two things.

This amazing thread on X that goes over

all these tips and tricks, as well as an

entire blog post with a bunch of tips

and tricks. What we're going to do is

not spend the next 20 minutes reading

the blog post and thread word by word

like every other AI channel out there.

We're instead going to be about that

action, boss. We're going to go and

implement every single one of these into

a live app we're going to build

together. So, let's do this. We're going

to do this all Claude Code CLI. I know

in the last video I said Claude Desktop

is the best way to use Claude Code at

the moment. The thing is, a lot of these

new features they just dropped on our

heads for some reason is only available

in the CLI. So, we're going to do this

in the CLI for now. The first trick

before we start building this app is

we're going to go into auto mode. So, if

you hit shift tab, then shift tab, then

shift tab, you will see auto mode. And

yes, we're going to make it the default

mode. Auto mode is really cool cuz it's

going to make Claude Code go so much

faster. My one complaint about Claude

Code, to be honest with you, is it just

asks for permission way too much. If

you're trying to build complex apps like

I am, every literal 5 seconds it's

asking, "Hey, can I do this? Hey, can I

do that? Hey, can I do this?" Even when

you're in accept edit mode, it's doing

that. The issue is is now everything

requires your attention every 5 seconds.

You can't multitask and do other things.

Well, they just released this fix. They

released auto mode, which is now what I

recommend as a default mode for everyone

using Claude Code. What it's going to do

is it's going to use the AI to determine

what permissions it should have. So,

instead of asking you, "Should you do

this? Should it do that? Should I have

permission for this?" it's just going to

ask itself and kind of make good

judgment. What this will allow it to do

is work continuously for hours on end

without having to interrupt you every 5

seconds, and that's incredible and makes

Claude Code so much better. So, what I'd

highly recommend doing is hitting shift

tab until you go on auto mode, and now

Anthropic will handle all the

permissioning itself, allowing you to

give a prompt and it just goes on its

own. That's the first of the seven

tricks Boris gave. Let's go into the

second one here, and let's start

building our app out together now that

we're in auto mode. The second trick he

gave is to front-load as much

information as humanly possible into

Claude Code before you start a session.

So, what makes Opus 47 so special is

it's a lot better at listening to your

directions. And so, what you can do is

when you first go into Opus, now that it

can run for a long time on its own, you

want to front-load as many instructions

as humanly possible because it will be

really good at parsing and following all

those instructions without having to

interrupt you at all. So, let's build

this app out together. I'll put this

prompt down below. What we're going to

do is build a really helpful app. If

you're at all familiar with Linear, it's

the number one project management tool

right now. It's like completely free and

awesome, but I personally like building

my own version of every app I use. So,

we're going to build our own version of

Linear, and if you do this with me,

you'll have an awesome app that you can

use for the rest of your life. It's

really good at helping you manage your

tasks and projects. So, before when I

was building out projects in Claude Code

or Codex, I would kind of build it out

one step at a time. Just give the first

step and the second step and the third

step and the fourth step. What we're

going to do instead is we're just going

to give it all at once. I want to build

a Linear clone out. This should be a

project management tool where I can

build projects, then put tasks and

issues in those projects. Every project

task and issue should have status and

priority. It should be a really clean,

simplistic, linear-like interface. I

should be able to assign users each task

and add any other features you think

will be helpful. Use Next.js to build

this out. Next.js is just a web

framework a lot of popular web apps are

built on. We'll make this really easy to

build and I'm going to hit enter on

that. What you're going to see now is

Claude Code will be able to obey every

single instruction I gave that I

front-loaded there and I'm not going to

have to give permission at all. It's

just going to do it all at once because

we're in auto mode. And in a second, I'm

going to give you two more tricks that

are going to be really amazing and make

this experience so much better, but

let's get the V1 of this app done. And

if you took the prompt down below and

used it, you're going to have a really

amazing app with me here. And just as

this is building, I get this question

all the time. Hey, when should I be

using Claude Code? When should I be

using Open Claude to build things? Just

as a reminder, as this is building out,

I like to use Open Claude to build

prototypes of apps when I'm on the go

since I can just hit it up in Telegram.

I also like to use Open Claude for

building out tooling for Open Claude, so

building out tools for my agent to use.

But when I'm building out deep, complex

apps, if I'm building out

consumer-facing apps, if I'm going into

a deep coding session like we're doing

right here, I'm using Claude Code. It's

just going to be a lot better at coding

by itself. All right, looks like it's

done and it's already running. Let's

take a look to see what we got here. And

boom, look at that. That's good-looking.

Anyone who's ever used Linear before

will definitely recognize this. So what

you can see here is a list of the tasks

that are assigned to each project. So we

can go in, we can create new tasks.

Let's say finish this YouTube video, put

it in to do, assign it to me, create and

boom. This is beautiful. I mean, this

looks pretty much exactly like Linear.

This is really, really nice. You even

have tags in it. This is great. If you

took the prompt from down below, you now

have an awesome project management tool

you can be doing for all your vibe

coding. This is sick. Which brings us to

the third tip that they revealed in the

thread and blog post that they just

released, which is recap. If you do

{slash} recap, this will generate a

one-line sentence recap of everything

your agent has done to this point. This

is really helpful because if you

followed my previous tips around

front-loading information and having

your agent run for hours on end, it'll

do a whole bunch of things that you're

not even aware about. If you do {slash}

recap after a long-running session or

after maybe you gone to bed and you woke

up the next morning and want to get back

to work, it will give you a rundown of

everything you've done up to that point.

This is unbelievably helpful if you

treat Claude Code like an employee,

which you should be doing and letting it

just do its thing for long periods of

time. Which brings us to our next tip,

which is effort level. They made it a

lot more configurable how much effort

your model puts in when you give it

tasks to do. So, if you do {slash}

effort, you hit enter there, you can

select your effort level. So, they have

added a couple of things here. First,

they added extra high and max mode. And

what I found based on my tests is you

basically want extra high for pretty

much everything unless you give a super

complex task. So, what I would recommend

is if you do what I said earlier, which

is give you know, front-loaded

information on every single thing you

want in app, I would go into max mode

for that. And then afterwards, as you're

configuring, adding new features,

changing things around, switch it to

extra high mode. The only reason I would

ever use high, medium, low is if one,

you're not on the $200 a month plan,

which I highly recommend. If you're on

the lower plans, then I probably would

want to get more granular and switch to

low for things like changing button

colors and things like that. But other

than that, if you're on the $200 a month

plan, I've gotten away with using extra

high for pretty much everything other

than that front-loaded max big prompt I

give to kick off apps. So, if I'm you,

use extra high for pretty much

everything unless you're on one of the

lower plans, then you can switch that.

But /effort

will help you configure how much effort

it's putting in on every single task.

The next tip they gave, and they gave

this in this blog post, I'll put the

link for the blog post in the thread

down below. this where our tip number

five here is notifications.

You should be turning on notifications

for Claude Code, so that after it's done

running these long-running tasks, it'll

ping you, put a notification here that

says, "Hey, I'm done. Check this out."

The theme of a lot of these tips I'm

giving is you're now increasing your

trust of Claude Code to have it run by

itself. Instead of hand-holding it

through a whole bunch of tasks, you're

giving it trust and you're letting it

run on its own. So, what you can do to

set up notifications is actually really

simple. Say, "Hey, set up

notifications." So, you can say, "After

every piece of work you do, please send

me a notification." And just by doing

that, it's going to set up its own

automation to ping you every single time

it's done something. So, I'm going to

hit enter on that, and it's going to set

up that notification. So, after

long-running tasks, it pings me, and I

can come back and quickly get into focus

with Claude Code. So, as you can see,

it's actually loading up a skill that

can actually send you those

notifications. So, it's setting up a

skill, it's setting up a hook, and every

time a prompt is done, it's going to use

that hook to say, "Hey, Claude Code

here, all finished. Check out what we've

done." This will speed up your workflow

so much cuz you can give a long-running

task, let it work for a while, come

back, and now it will ping you every

time it's done, so you can always stay

in flow. By the way, just as a little

side tip here, what I'm doing, when I

give these long-running tasks, I'm

always switching to the chat in the

desktop app. I actually find for the

first time I love just straight-up

chatting with Opus. Previously, I

preferred Chat GPT for chatting and

bouncing business ideas, but Opus 4.7

has actually been really, really good

for me when it comes to coming up with

new ideas for my business. So, when I

give a long-running task to Claude Code

and I'm waiting for that notification to

come in, I'm going to chat mode in the

desktop app and I'm talking about

whatever it is I'm building and coming

up with brand new ideas for that app and

for that business. Highly recommend that

workflow, by the way. Before I give this

last tip, by the way, leave a like down

below if you learned anything at all.

Make sure to subscribe and turn on

notifications. Also, I do a live boot

camp every single Friday in the Vibe

Coding Academy. I also do live Vibe

Coding sessions every Thursday. So, if

you join now, you can hop in, watch me

vibe code live with Claude Code. I

narrate everything I do, show you every

single tip and trick I'm using as I'm

vibe coding. So, if you want to watch me

vibe code live and ask me questions as I

do it, join the Vibe Coding Academy.

Link down below for that, biggest

community in AI. The last tip I'll give

here is working with adaptive thinking.

So, they actually removed extended

thinking with Opus 4.7. Before in Opus

4.6, you can turn on extended thinking

and it will use a lot more tokens to

think every step of the way. Now,

instead what they're having you do is

you're actually going to use prompts to

initiate extended thinking. So, for

instance, say I want to make it so I can

invite other people to use my project

management tool, right? I want to invite

other people in and they can add task

and manage task with me. This is great

if you have a team. I would say, I want

to make a collaborative mode here where

I can invite other people to use this

project management tool with me. Think

carefully, step by step, about this. So,

you have two different knobs you can

control here. How much effort it's

putting in, and how much thinking it's

going to do. Effort is basically going

to be, in total, how long they'll spend

on a task you give it, and thinking will

be how long it spends on each individual

micro step of that task. I like to use

the extended thinking for when I give a

smaller task that might be more complex.

Like this, this is a collaborative mode.

So, this isn't some gargantuan task that

with a hundreds of different features in

it. Instead, it's one task, but it's

really challenging. So, I want to think

extra hard about this one simple task.

So, I'll just say, "Think carefully,

step by step." If this was a smaller

task, like, "Hey, just update the color

of a button." I'd go in and I'd say

something like, "Prioritize responding

quickly, rather than thinking deeply."

So, now you can control the thinking

budget of each one of these tasks. This

is the new way to do adaptive thinking.

So, just putting on some extra prompt

that says, "Think hard about this." or

"Don't think too much about this at

all." This will control the amount of

tokens that are spent on each individual

task, rather than overall. Those were

the seven tips of getting the most out

of Claude Code that was just revealed by

Boris, the head of Claude Code. I hope

this was helpful. If this was, leave a

like down below. Subscribe, turn on

notifications. I also live stream

Monday, Wednesday, Friday, 11:00 a.m.

Pacific. So, I'll see you there. I will

see you in the next video.