250 | The No-Code Way to Use Claude Code For Creating AI agents for Business with Tim Cakir - Leveraging AI

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Isar Meitis (00:00):
Hello and welcome to another live episode of the

(00:04):
Leveraging AI Podcast, thepodcast that shares practical,
ethical ways to leverage AI toimprove efficiency, grow your
business, and advance yourcareer.
This is Isar Metis, your host,and we have an incredible
episode for you today.
Now, everybody, literally,everybody is talking about
building agents and how they'regonna change the world and so
on, but the reality is very fewpeople actually.

(00:26):
Build agents and even fewerpeople build agents that
actually do something meaningfulin their businesses day in and
day out.
Our guest today, team Ecker, hasbuilt over a hundred agents that
work in his company regularlyand do different functions in
his business.
And through this process, hebecame so familiar with what's
good and not good that evenfired more than 50 of the agents

(00:49):
that he initially created.
So he's been evolving this for awhile now, to make this episode
even more interesting, we aregoing to talk not only about how
to build the agents and showexactly how to do this.
But we're also gonna look at itwith a very interesting
platform, which is Claude Code.
Now, from an adoptionperspective, everybody knows

(01:09):
Chacha piti.
Some people know Gemini becausethey get it through their, uh,
Google membership, but a lotless people are actually using
Claude, which I personallyreally, really like.
And out of that small peoplethat uses Claude, a much smaller
percentage uses Claude Code.
And the vast majority of peoplewho use Claude Code are coders.
But the reality is there arehuge benefits that you can gain

(01:31):
by using Claude Code.
Not coding, and this is notexactly what we're gonna look at
today, is how to use cloud codenot coding because I don't know
how to code and I'm stillgaining a lot of benefits from
it.
Now, the reason most peopledon't do it yet is because
either they don't know that itexists or it has code in the
name and people assume that'sthe only thing it is good for,

(01:53):
which is not the case.
And this is one of the reasonsI'm really excited about this
episode, because you're gonnalearn a new tool that most of
you probably do not know, andyou're gonna learn how powerful
it is and what can you do withit without writing code Now.
The other reason a lot of peopleare not using Cloud code is
because until not too long ago,you had to go through a weird
install, which we're gonna showyou how to do in order to get it

(02:13):
to actually up and running.
But the reality is, as of abouta month ago, and Tim, correct me
if I'm wrong afterwards, uh, youcan now run cloud code on the
web, which means you don't haveto install anything.
There's still benefits ofinstalling it, uh, but you don't
even have to do that anymore.
So in today's episode, we'regoing to explain to you the
benefits of Running Cloud Codeand how to use it to create
incredible agents that actuallydo functions in your business

(02:37):
and provide value.
Now, our amazing guest, Tim,brings.
And really huge variety ofexperiences that makes it highly
relevant for our episode today.
He held top leadershippositions, including CEO and COO
in many different companies,which means he understands how a
business actually runs and whatit takes to run an effective

(03:00):
business.
But he was also a professor intwo separate university, which
means he's really good andpassionate about teaching, and
he's currently the chief AIofficer and the founder of AI
Operator, which is a companythat helps businesses in
implement AI in an effectiveway.
So this combination makes himliterally the perfect person to

(03:20):
teach us about cloud code andhow to create agents not being a
technical person.
So you don't need any coding ortechnical skills to do this, and
that makes me very, veryexcited.
So Tim, welcome

Tim Cakir (03:31):
to leveraging ai.

(04:16):
Well, thank you so much for thekind introduction.
It's an absolute pleasure to behere, sir.
Thanks.

Uh, you know, you and I spoke several different
times.
I, I'm a big fan of your, uh,YouTube channel.
Uh, we're probably gonna talkabout that in the end where
people can find you, but you're,you're really a magician when it
comes to doing these kind ofthings, but let's really start
at the more higher level.
Mm-hmm.
What the hell is cloud code andwhy do people need to care about
it?

Tim Cakir (04:41):
Yeah.
I think that, uh, there is oneissue about cloud code is that
they called it cloud code.
Yeah, yeah.
I think it's better than

GPT.
At least it has a name thatmeans something.

Tim Cakir (04:51):
Yeah.
Yeah.
It is just the code.
I mean, CLO is, is great, but isis, they call it CLO code and
then everybody, like I tell'em,I'm like, how have you tried CLO
code?
And they're like, I don't code.
I'm like, well, you're missingout.
Or maybe it's good for me'causeI use it and not others do.
But no, I like to share, I wanteverybody to use it.
I want everybody to get theproductivity gains and the
efficiency and the incredible.
I guess you call it magic, Icall it just obsession about,

(05:13):
uh, obsession about AI and whatit can do.
Um, and so cloud code, uh, wasbasically a, a way to interact
with the cloud models in the CLIin the command line interface,
which is the terminal in acomputer.
And that's very scary, right?
And because there is no ui,there's no user interface or the
user interface just text, it's adark screen.

(05:34):
It looks scary, right?
It looks absolutely terrifying.
I even sometimes I open, I'mlike, oh shit, am I working here
again?
Like I just work 90% of my timenow on the terminal, you know?
And it's like, it's so funny,like we've done such UI designs
and so on, right?
And then now we're going backlike to using the terminal.
Yeah.
Uh, so code code is really anincredible.

(05:55):
Let's call a product to, um, totap into, uh, large language
models that are Biro.
So code Opus 4.5 being the lathe latest one.
Um, and what's, why is it sopowerful?
Because it runs on yourterminal.
It can create files, folders, itcan run scripts, and it can
really control everything onyour computer, right?
Yep.
From including browser to theapplications to the settings of

(06:18):
your computer itself.
And because of that, it takesthe, you know, a level higher
in, in, in, um, in control andefficiency and being able to
achieve things than a chat GPT,right?
The chat, GPT is great.
You have to connect things, youknow, you, you click on
connectors, clo itself, youclick on connectors, but with
cloud code, it's in yourcomputer.
It's, it's, it's literallyeverything that you have in, in

(06:38):
your computer that you can run.
Um, and, uh.
In a nutshell, it looks scary.
It sounds scary because of thecode, if you don't code, but
what I've realized with Cloudcode is that it's right now the
most powerful product forbuilding systems, business,
operational, running systems.
Um, and, and, and, and I'm, I'mstill mind blown, you know, when

(07:00):
I speak about it, I'm like, whatthe f is it?
You know, it's just like, it'sstill incredible what we're
living and it's just, you know,the beginning of December, um,
and quote code is, is is justmind blowing.
Absolutely.

Amazing.
So, so let's really show people,first of all, how it looks like,
and then we can start divinginto examples of how you are
using it to create these systemsand processes.
Uh, and again, people Yes, I, Iagree with everything Tim said.
Like, what is the terminal?
How do we even get to it?
And it's really simple.
You both on Mac and in pc, it's,there's a button that takes you
there and, and you can, and youcan have the terminal.

(07:34):
Uh, and as it, as Tim said, itlooks like we're back in 1985.
It's like a dos prompt, kind oflike environment.
Uh.
But it works.
And it is extremely powerful.
And for those of you who'relike, oh my God, I don't want it
to mess with my computer.
Uh, you do as long as it'scontrolled.
And one of the amazing things inClaude models in general, and

(07:54):
specifically Claude 4.5, is thatit's code adherence, which is a
professional term to saying itdoes what you fucking tell it to
do versus, uh, kind of does whatyou want it to do.
It, it, it is very good at that.
So it's only going to do whatyou tell it to do, and it is not
going to do anything.
You do not tell it to do.
So the fear of like, oh my God,this thing can take over my

(08:16):
computer.
In theory, if you tell it to youand you define exactly what you
want, it will.
But if you tell it to do one,two, and three, create this
file, copy the information fromhere, bring it there, activate
this thing, and send thismessage, then that's the only
thing it is going to do.

Tim Cakir (08:31):
Yeah.
And I think that it has reallygood, um, security settings.
Uh, you can set the settings sothat you can decide on what it
can do without asking you eachtime.
Um, and it's also incredible atreally asking you when it's
necessary and when it doesn'thave enough information, it's
like, Hey, I'm about to do this,but I'm not sure, like, you
know, what do you think aboutthis plan?
And you're like, oh, okay, Ilike the plan.
Go execute the plan, or let'schange the plan.

(08:52):
Right.
The planning mode of it is just,it's just mind blowing

in, in general, I find Claude to be the most
collaborative model out of allof them.
That really makes you feel thatit's a consultant and not just
an execution arm of your dreams.
And that makes Claude like anamazing partner for anything.
This as well.

Tim Cakir (09:10):
Yeah.
I think that in, you know, if wetalk about models very quickly
before I show things, you know,I use Claude, um.
4.5 opus, uh, the most at themoment.
Um, but if I'm doing a businessstrategic task, like planning,
you know, Q1 2026 for mybusiness, I will still switch to
GPT 5.1 Pro.

(09:32):
Right.
Which it's only on the propeople.
Uh, so not the plaster 20 bucks,but the 200 bucks.
Yeah.
And people tell me, oh, I havethe pro.
I'm like, no, no, no.
You don't have the, the pro howmuch you pay 20.
Yeah.
Well, you don't have the pro.
Uh, so GBD 5.1 Pro is justincredibly amazing at business,
business logic, businessstrategy, and so on.
Um, but everything else now,I'm, I'm definitely a Cloud 4.5

(09:53):
Opus fan.
Um, and it's, you know, uh, ifyou pay for Cloud Max at 20 x,
which is$200 a month, you don'thit the limits.
Uh, and I run that on my cloudcode.
So just FYI for everybody.
Be careful, you know, I'm gonnashow you things now, if you run
it on the API, the cost can goup very, very quickly.
Yeah.
So you do wanna be on asubscription package, on a Max

(10:13):
Cloud max plan, a 10 x or a 20 xplan.

Okay.
So let's dive right in.
Yeah.
Let's see how this thing lookslike, how to set it up and how
to create magic with it.

Tim Cakir (10:21):
Awesome.
Let me share my screen.
Uh, can we see my screen?
Yeah.
Yes.
Awesome.
So, so this is why I said it'sscary, isn't it?
It's just, it's it's just ablack box, right?
Uh, so let, let's

take, let's take a step back.
Yeah.
How do you get this thing?
Yeah.
How do you open terminal on yourcomputer?
Yeah.
So for people who don't knowwhat it is and

Tim Cakir (10:37):
Absolutely.
So I run a version of theterminal called I term two.
It's an application that you candownload.
Um, you can run the normalterminal.
The normal terminal doesn't havesome of the little gadgets.
I mean, you know, now you seethese as cool gadgets'cause you
can see, you know, uh, the GPUand, and, and what it's doing.
But on the normal terminal youdon't see much.
So it's a little bit cooler thanjust the terminal itself.
There's also another option thatI was into for a bill bit called

(11:00):
Ghostie with doublet ghostie,uh, which was also a terminal
tool, which is really good.
But you know, you just go onyour spotlight in your search
spotlight if you're on a Mac orin a Windows, uh, you know, to
your search and you go rightterminal.
Right?
And if you want a little bit ofbetter experience, I would
suggest I term two, uh, is agreat little app.
Yeah.

Cool.
Yeah.
Yeah.
I, I will say one more thing,not to confuse people, but I
would say anyway, if you areusing any of the vibe coding
tools, whether it is Replicateor mm-hmm.
Lovable or whatever, uh, there'sa terminal version in there that
you can run as well, and it hasslightly better user interface

(11:41):
than this thing, but overall, itdoes the same stuff.
So if, if you're starting withnothing, then just running the
terminal in the computer couldbe your easiest starting point.

Tim Cakir (11:49):
Yeah, I mean I think that the difference there is,
uh, I think we, we spoke aboutthis when we first met, but you
know, I was saying that yes, inthe Rept terminal you can run
CLO code.
Yeah.
Uh, I think you tried it since,didn't you?
Six minutes

after we close the call.
I already had it running.
Yeah.
It's one of the best

Tim Cakir (12:06):
hacks people don't realize is that you can open up
Repla and not pay too much forrep tokens and only set up your
environment and then run theterminal quote coding the
terminal and inside Repla andthen you can see the preview,
which is great.
But you know, there we're reallytalking about coding and
building a product.
Yeah.
That you wanna be in that folderand you wanna be in that
repository.
But here, what I really love.
Is that I'm on my computer.

(12:28):
Yeah.
So the first thing that I didwhen code Code came, um, and I
was like, oh, what's this crazystuff?
Right?
I was like, Hmm, okay.
And it was a couple of sleeplessnights as usual when anything
new come, uh, it's a sleeplessnights, you know, I, I love it.
Um, and the first thing waslike, oh, okay, it can do files,
folders in my computer.
And it was years that I've beenwanting to tidy up, you know, my
desktop documents and, and allthat kind of stuff.

(12:49):
And I just said, Hey, can yougimme a strategy?
And I was like, great, can youimplement that?
And it changed names of filesand images and videos and put
them in folders and it just kindof start cleaning out my
computer.
I was like, I.
Oh, crap.
This is, this is very good.
You know, I want, I've beenwanting to do that for 10 years
or so, and I never did it.
And, you know, I just piled shiton top of shit.
And then it, it was, it wasputting all the PDFs somewhere,

(13:09):
naming them, reading them,telling me what they are and
really creating differentfolders for me.
I was like, wow, this is, thisis really like, we're starting
to see real agentic, agenticsupport assistant, whatever you
wanna call it on my computer.
Um, so here's, here's theterminal, right?
I term two.
I term two, running my terminalinside my terminal.
And it's, it's a, it's a blackbox.

(13:29):
It's a black screen.
Lemme see if I can make this atiny bit bigger for people
maybe.
Yeah.
Uh, so it's just a, a black box.
Um, and then you have to installcode, code on your computer.
That's very easy.
You can actually go on, on, on,on Chrome or in a, in a browser
and write code, codeinstallation.
It's a, it's a very smallsnippet of, of some, some words

(13:50):
and some, some kind of codesthat you just input here and you
let it run and it installs it,it asks you to log in.
Right.
So I'm not gonna do that'causeI've already done that.
It's very easy.
It's, it's this five minuteprocess.
Um, and then what I'm gonna donow is I'm just gonna tease you
a little bit and then I'm gonnashow you what I've built and
then we'll go into, into how tobuild what I've built and, and,
and this type of things, ifthat's okay with you, ISAR.

That's awesome.
Sounds like a great plan.

Tim Cakir (14:12):
So, so what I'm gonna do is I'm just gonna launch
Claude by saying clo.
So I'm just gonna say, Claude,and I'm gonna enter, and as we
can see right away, I get thequestion asked if I trust this,
uh, folder, right?
Because, uh, Claude code mayread, write, or execute files
contained in this directory.
This can post security risks.
So only use files, hooks andbash comments from trusted
sources, meaning promptinjection.

(14:32):
As we, as we know, if this wassome files and folders that I
didn't know, and then there,there was some, some kind of
hacker trying to get myinformation, it could kind of do
that.
But it's my computer.
I know it's clean.
You know, I, I've, uh, I've beenworking on my computer for, for
really long, um, even if Ichange computers, iClouds, you
know, so, so I know what's in mycomputer, so I'm gonna accept
yes, proceed, right?
And then that's it.

(14:52):
We get this little, this, thislittle, you know, uh, I love it,
this little thing, you know.
And here we go.
This is closed, closed code isready, and I have Claude on
opposite 4.5 with a closed maxplan, right?
As I was mentioning, you don'twanna be on the API, you're
gonna pay a shit ton of money ifyou do that.
Uh, and then I'm on my user.
I'm in the.
I'm in my whole computer.
I'm not even in a folder, butwhat you could also do which

(15:14):
I'll show later, you can run itin a specific folder.
Right.
So I do run it in very specificfolders.
Like I have a sales folder, Ihave an onboarding folder, I
have a customs folder, and I runit in different places.
And what's great about this isthat I can open up multiple tabs
and I can run four or fiveClaude at the same time.
And it's like, oh.
So I had an assistant, and now Ihave multiple assistants, and

(15:34):
each can run multiple agents andeach can have different skills,
Claude skills.
It, it just gets a bit meta.
Like, it's like I runorchestrators that run other
ones and they run each other.
And it's just'cause like what?
And you know, I was readingabout hyper productivity and
it's scary because it's, it'squite exhaustive.
It's like, oh my God.
Like I have to, I have tobabysit all these agents.
You know?
So, so I think it's, it's tough.
It's not easy to get to theusage that I am, but I'll show

(15:57):
some simple tricks for everybodyto, to start getting there
because this is the world thatwe're really living in.
But in 2026, I do believe thatthis is going to be one of the
most powerful things is to buildsystems.
That can run without you.
Right?
And you wake up in the morning,this is my dream.
I wake up in the morning, whichI'm kind of doing a little bit
of that, but I wanna wake up at9:00 AM not, you know, I wanna
be on my desk at 9:00 AM I'vedropped my kids to school.

(16:18):
I'm at 9:00 AM uh, in my desk.
I open it and I have, I've doneall of this stuff.
Um, bless you.
I've, I have all of this stuff.
I've done all of this stuff.
These are things I was a bitunsure, you know, I wanna ask
you some questions.
Great.
What is, what did it do?
I it did all the follow ups, itsent the proposals.
It's, you know, it's, it'screated some, uh, contracts for
some clients and so on.

(16:38):
And then it has someclarification questions.
Like, Hey, I wasn't sure aboutthis.
You know, like, can you checkthat email?
Can you check that document?
And you check it and then yousay, okay, you know, glad you
didn't send this because here'ssome feedback, right?
And then you give that feedback.
But then that feedback the nextday, I don't want it to ask
similar feedback because itknows it.
So I wanted to learn, and I wantit to be iterative learning,
right?
Like, so that it keeps, it keepslearning every time I give it

(17:00):
feedback.
So I kind of build somethinglike that.
I'll show in a second as well,but this is the world we're
gonna live in, guys, and youknow, it's just gonna wake up
throughout your whole sleep andyou went to school, dropped the
kids and stuff, you came back,you've had multiple agents, you
know, 50, a hundred agentsactually doing work for you,
right?
This is definitely happening.
Like I can show you in a second,you know, um, it's just that I
still dream of a little bitmore, more proactivity from

(17:22):
these agents, right?
Yeah.
I still have to run them.
I can schedule them, but the,the more proactive is coming.
I know that because we saw thatwith chat, PT Pulse, right?
On the ProPlan we have somethingcalled Cha PT Pulse, which every
morning it gives me like someincredible things about me,
about my search history, aboutmy day, about my calendar, and
it kind of is proactive, right?
So, so I think we're gonna getthis proactive right now is very
reactive.

(17:43):
So I have Claude.
Claude ready.
It just, it's scary because youdunno what to do.
It's, it's asking you like, it'swaiting for me.
It's like, okay, well I'mwaiting for you.
What are we gonna do?
Right?
So let's say what, what we'regonna do is we're gonna go and
we're gonna launch, let's, let'slaunch a guide about Claude
code.
How about that?
Right?
So I'm gonna write a 10 pageguide about Claude code, right?
And I wanna write, I wanna writethis.

(18:04):
So what I'm gonna do isactually, I, I would kind of go
to the folder where I havecontent and guides, but I'm not
gonna do that.
And what I'm gonna do, hopefullythis is gonna work.
I'm gonna talk to it.
'cause I, I talk a lot to, tothese things now and I don't
type, so I'm just gonna, I'mjust gonna talk to it quickly.
Uh, go to my, uh, guides folder,uh, inside, uh, my content
folder in AR operator and run myuh, guide system.

(18:29):
And let's write a guide, um,about cloud code, how, uh, how
cloud code with Opus 4.5 and thecloud max plan.
And how cloud code people thinkit's for coding, but it's not.
You can run business operationsand you can do a lot of things
and so on.
You know, do a proper researchand make sure to run all the
agents and, and to run theactual, uh, guide system

(18:51):
workflow, right?
So you still kind of have tobabysit a little bit and you
have to tell it like, make sureto run my workflow, right?
Make sure to find the agents,you know,'cause if not it, it
still wants to please you and isgonna go and do it on its own
thing.
But I already have a system forthis.
So what I'm gonna do now, I'mdoing shift tab.
As you can see, something ischanging on the bottom left and.
Edits own so that it is kind of,it accepts like you

(19:13):
automatically accept some stuff.
I'm gonna take it to plan mode,right?
Because I don't want it to goand actually start doing things.
I want it to plan first.
It gets very powerful when weput things in plan mode, right?
So this gonna take quite a bit.
And

again, to switch to plan mode, you just used, uh,
shift tab.
Yeah.
Yeah.
So just use a, a keyboardshortcut.
There's no, again, like anythingfancy to do that you just switch
the mode that it's gonna operatein.

Tim Cakir (19:36):
Yeah.
Shift tab.
Shift tab.
And then you get to plan mode.
You see, uh, it is green on thebottom left.
It has like this, this kind ofpose button.
Is it a pose button?
So it's, it's more like, Hey,I'm not gonna take actions, you
know, I'm just gonna plan, youknow, I'm not gonna go change
anything.
Uh, and what we can see, youknow, we can open up this whole
thinking with control O, whichgets really big.
Um, you know, and it looksscary.

(19:57):
This is where exactly it startslooking scary because like, oh,
okay, what is this?
Actually it's not that scary.
'cause this is just like, oh,it's finding folder, right?
It's going to my documents andmy documents is found my AR paid
folder, right?
My company name folder.
And then there it found thecontent folder and then it
opened that up and then it'slooking at certain things and so
on, right?
And as you can see, I just askedsomething, but it wrote its own

(20:19):
it.
It wrote its own prompt.
Find the AR paper project on thesystem, look for a folder or
project called AR paper orsimilar.
We didn't find a content folderand a guide sub folder, look for
any guide system and so on,right?
So it kind of gave itself sometask and I was like, okay, this
is what I'm gonna do.
And then it keeps thinking, I'mgonna close this'cause it can
look a bit scary and let's go tothe normal look.
Right?
So that's like the detailed lookof seeing what's inside these

(20:41):
360 lines, these 376 lines.
What is it doing inside these?
You just go control, oh, it getseven more scary than what it is.
So you close it, you're like, ohmy God, lemme not see that.
Right.
You know?
But it's good.
I do recommend, you know, whenyou get comfortable, I do
recommend to read these thingsand you know, it might, yeah.
If,

if you wanna understand what the model is
doing behind the scenes, and bythe way, it's the same thing in
like all the regular tools.
Like if you're in Chachi PT or aClot or Gemini, it says
thinking.
I'm like, okay, what are youthinking about?
If you click on the thinkingBlinking thing, it's gonna show
you exactly what it's doing.
And then the, the benefit ofthat is you understand how the
model works and how it thinks,and that usually makes you a
better prompter because you knowthe steps that it's going to do

(21:18):
and you can guide it just likeTim did now.
Like, oh, go look for thisfolder.
Go look for these agents.
Why is he doing that?
Because he looked previously andsaid, okay.
It's kind of like strugglingwith that.
If I tell it what to do, it willknow what to do.
So it, looking at this actuallygives you a better understanding
of how the models work and thenyou can be a better prompter
because you can know how toguide it.

Tim Cakir (21:35):
Yeah, absolutely.
And I don't, it's not just aboutbeing a better prompter, it's
about this thing that we see inthe news, isn't it?
It's like, it's saying that'shumanity's getting dumber.
Right?
They're getting dumber and kidsare getting dumber because they
let Chachi BT do all the workand they just get the output and
then they go with it.
Uh, but the second you open thereasoning, the thinking, right?
The thinking I always say,because it's the reasoning and

(21:55):
you open that up and you startreading it.
And in every model that you'reusing, if you can read that, you
get smarter.
Yeah.
You get absolutely smarter.
I see things that these thingsare thinking or like reasoning
and I'm like, whoa, I didn'tthink about that.
Or Let me, let me bring moreinformation.
Lemme bring more context.
Right.
It becomes really an, uh, apartner, you know?
Yep.
And not just like, God, do thiswork for me.

(22:16):
And then I, and I, yeah, it'sokay.
Let me go.
No.
It's like, oh, well we workthrough things together and I'm
learning and the next time Iknow how it's gonna kind of
think and how it's gonna reason.
And I start understandingdifferent models.
It's, yeah, it's very geeky, butyeah, I am obsessed about all
this stuff.
So, so, uh, as you can see onthe bottom, what happened, this
is quite a new ui, uh, userinterface kind of feature is,

(22:37):
you know, it did 1, 2, 3, 4things like project type,
audience use cases, pricingfocus, these like a tabs, and
then I can submit and it'sgreat.
It's asking me clarificationquestions now, right?
So, so, okay.
Do I wanna standalonedeliverable, like top level
folder, like AI safety startuppack, which I, I've already
built that before.
So it's like a, do you want acomplete package for
distribution or do you wannaresearch the guide pipeline?

(22:58):
Right.
Do you want just a, just a, aquick guide about quote code?
Or do you want something fordistribution?
I, I potentially want astandalone deliverable, right?
So I'm gonna click that.
So I entered that.
So, so basically we tick that,so you can also tick things if
you wanna see'em, you can do aspace bar, which I've already
done, so no.
But here, who is the primarytarget audience for this quote?
Code guide, right?
Business operators, I think, uh,entrepreneurs.

(23:20):
Founders.
Yeah.
I think I would go, I'm gonna gofor entrepreneurs and founders,
right?
Because those are the peoplethat I kind of, I kind of talk
to, so I'm gonna go, yeah.
Entrepreneur, founders.
What specific businessoperations or use cases should
be emphasized beyond coding?
Right here?
I would potentially spend a bitmore time, right?
And tell it to the things, butI, I'm just gonna do something

(23:40):
quickly.
I'm just gonna tell it so thatit kind of finds my systems.
Make sure to go read my systems.
I have a, uh, this guide system,which is about content creation.
I have a sales system, which isabout, you know, running a CRM
and so on and, and how to sellin my company.
I have a onboarding system.

(24:00):
After I close the deal, how do Ionboard a client?
And then I have a trainingsystem.
How do I train people and createthe training program, the custom
training program for our 12 weekAI first mindset training
program, review all of that.
And then as a bonus, because,um, I'm on a very important, uh,
call right now.
Also, can you highlight, um, thesystem building system?

(24:23):
Oh, thanks.
Alright, so I kind of said that'cause I, I wanted to go and,
and check these things, right?
So for those

of you who are not watching the screen and just
listening the way this lookslike, again, think about the
poorest user interface you canimagine.
So dos level, so there's justthis little, there's different
rows and you can pick from themby scrolling to them with a
keyboard and selecting them.
Uh, or you can go to the bottomone, which is just like a.
Open line and type whatever youwant to type.
What Tim is doing is what I'mdoing as well.

(24:51):
We're voice typing everythingjust because it's a lot faster.
It requires a decent level oftalking skills, but not much
because these tools are prettygood at understanding what you
want.
Even if you didn't structure itas well as Tim, uh, it you get
much better at this.
The fact that Tim has a YouTubechannel and I have a podcast
makes it a lot easier to saywhat we want in the first time

(25:11):
around in an organized way.
Uh, but you will figure it outas well, and it just
significantly faster thantyping.
And to be fair, I find that Iexpress myself a lot more, uh,
with more details and color whenI speak versus when I type.
And so that's another benefitbecause then the AI gets more
context.
But now that you've done this,let's see what the system does.

Tim Cakir (25:31):
Yeah, absolutely.
I just wanna add something tothat is because I'm still kind
of practicing the, speaking tothe machines instead of typing
to the machines.
Because guys like every, youknow, guys and girls and
everybody, you know, like getready because we are going to
speak to the machines more byvoice.
You know, the, the new open AIproduct hardware is gonna come,
is definitely not gonna have ascreen, right?
So we understood that.
So, so it's not gonna have akeyboard.

(25:53):
We're gonna speak to thesethings and you're gonna have a
robot in your house, you'regonna speak to it.
Like these things aren't gonnahappen.
So I think it's a great time nowto practice speaking to these
things.
Um, so I'm still not very goodat explaining what I want and
sometimes I go, but it's okay.
'cause it understands and it'ssome, somehow it cleans it up,
you know?
And I think by practicing I'mgetting better at it.

Yeah.

Tim Cakir (26:12):
So I'm gonna accept what I just said there.
I'm gonna be like, okay, yeah,this, this is, this is the good
stuff.
And then it said, uh, pricing.
Should the guide focusspecifically on Cloud max?
No.
All right.
Uh, all tier comparison, right?
So we went all tier comparison.
I'm gonna submit all thisinformation.
So I've submitted all thisinformation.
It asks, you know, what was thefour clarification questions?
We answered them.
Um, and now four or five, right?

(26:32):
And it's now going to do thework, right?
Uh, and you'll see the minutesit's spending and the tokens
that it's spending.
But I'm not paying per token.
I have a, I have a subscriptionplan.
Um, and you know, if you'relistening and not watching, it
kind of gives you its plan.
So it's exploring the operatorsystems because I told it, go
look at my systems.
You know, about the guide, thesales, the onboarding, the
training, my system, buildingsystem, which is the brand new

(26:53):
one, uh, I just done lastweekend and I'm still working on
it.
And it was like I was buildingso many systems.
I decided to build a system thatbuilds systems like I was like.

It's kinda like, kinda like the Claude skill that
builds Claude

Tim Cakir (27:05):
skills, right?
It's the same type of concept.
Yeah, exactly.
I took that approach.
I was like, you know what, whyam I keep designing, you know,
these systems?
I should have a system that asksme question and goes and designs
the system from the first fouror five systems I've already
done.
Right?
And, and I just finished thatand, you know, I'm still working
on it, but I finished kind ofthe first draft and I'm like,
oh, this is getting meta, thisis getting cool.

(27:25):
I even tag, I think tropic on,on one of my LinkedIn posts.
They didn't come, but I waslike, guys, you gotta see this.
You know, it's like I'm taking abit next level now.
Yeah.
Um, so, so what it does, itgives you the tasks and it has
four tasks here.
It's gonna explore, then it'sgonna design the guide
structure, then it's gonna writefinal plan for quote, quote
guide, and then it's gonnaexecute the guide workflow with
all agents.
Right.

(27:45):
So even before going to my guidecreation, uh, system, it's going
to do these three tasks, whichis the explore, the design, and
the write the final plan for thequote quote guide, which is
great because this is not AIslop, this is exactly what I do.
These are my systems and this ismy context.
And so the guide is gonna beabout exactly what I do.
And people are like, oh yeah,but you got AI to do your guide.

(28:07):
I'm like, yeah, well, AI looksat my context, asks me
questions, and is gonna ask memore questions.
And I build these systems forthe last, I don't know, couple
months or whatever.
Right?
It's definitely very, verycustom content to the context
of, of Tim, the AR operator andmy systems.

Let me ask you a pause, you and ask you a
question.
You are sending it to all theseresources, like your systems and
your agents.
Yeah.
Uh, can you talk a little bitmore about what these are?
Yeah.
Like what are these systems andwhat are the agents that it's
actually using?
Uh, because that's a big part ofthis Yeah.
This secret sauce, right?
It's like the clot itself isactually using these additional
resources in order to do thething that it's doing.

Tim Cakir (28:42):
Yeah.
Yeah.
So, I mean, this is a perfecttiming.
I just wanted to launch thisbecause it's gonna take, you
know, a few minutes anyways, andI, I'll, I'll change my, uh,
screen share.
I think you can see my browsernow, right?
Yep.
So I had closed code open foryou guys.
You know, you just go to cloclo.com product cloud codes, you
can find it.
This is like the kind of, uh,the kind of codes, uh, that you
have to, um, put it in theterminal to, to, to download it.

(29:02):
And then, you know, you canlearn a little bit about it.
It might be scary, but here'sthe plans that I was mentioning
and so on, uh, which is cool.
Um, and now before we get intothose systems, now on cloud
itself, you actually do havehere code, right?
So if you click on code nowhere, it opens another tab,
which is code code, but it runson the cloud.
Right.

(29:23):
Uh, uh, it can run on anenvironment that you give,
actually can start now runninginside your also system, uh, and
your GitHub re repository, whichis another tool where, uh, we
put code in.
So this is quite newish.
So we know that what's happeningis that they've realized that
the usage of cloud code and nowthey're like, okay, well let's
offer it on a browser.
Let's start offering indifferent things.

(29:44):
So there is gonna be a productthat is gonna be better in UI
and it's not gonna be theterminal.
So you're not gonna have to bescared, but I believe that the
biggest power still is in thatterminal.
So, so, you know, I'm stillthere.
Um, so because you asked thatquestion and I had prepared
this, of course, ISAR, uh, sothis is my, I never doubted you,
Tim.

(30:05):
So I prepared this quick thingwith cloud, with cloud code.
I said, Hey, you know, can youbuild something that shows my
systems.
And it is like, oh yeah, ofcourse I can build you a whole
thing and then build me a wholebeautiful thing that I'm gonna
talk about.
So, AR operator systemsarchitecture overview, right?
So all of this, that what you'regonna see runs in my computer in
folders, right?

(30:25):
They have agents, they have uh,skills, they have cps model
context protocols, they have allthese things, but it all runs in
my computer, right?
Um, so this is my new metasystem, which I'm very proud of
myself, like I'm proud ofmyself.
Finally, like I'm taking a wholeDecember, like almost on just
focusing on this, right?
So we've closed all our cohorts,uh, Q1, uh, clients are ready.

(30:47):
I have a few more training, uh,to finish before Christmas, but
I'm like focused.
I told my team and my guys, I'mfocusing on systems, you know,
'cause it's just like.
It's gonna help me scale in2026, you know, so I build a
system builder.
So this system builder is quitesimple.
It's the meta system that buildsthe systems.
It has five core agents, it hasan orchestrator, a
questionnaire, a matcher, agenerator, a validator, which

(31:09):
we'll look into it.
There's five simple agents.
You know, the orchestratoractually does the whole work.
The questionnaire comes and asksquestion to the human, which is
me, to make sure that it doesn'tkind of create its own, uh,
thing, but it's, it's from mybrains and stuff like that.
And then it matches things like,um, I created blueprints from
the existing systems that arelike templates, blueprints, and

(31:29):
it matches the new system.
That I need to a blueprint andit says, oh, you already have
this type of blueprint so that Igo even faster.
And then it generates the wholesystem and then it validates it.
So these are five core agentsthat work on these.
And then there is like theselittle factories, right?
So I have an agent factory, Ihave a skill factory, a closed
skill factory, a folder factorythat has to create the foldering

(31:50):
system.
I have a, a workflow factorythat creates the whole workflow,
that has a different agent sothat it knows how to run.
I have a script factory so thatanything that it needs to be
code, right?
So that it's deterministic andnot non-deterministic.
Because as we know, uh, largelanguage models are
non-deterministic.
And that's why you don't get thea hundred percent each.
Uh, same time, uh, same, um,each time the same input, the

(32:11):
same output, sorry.
And that's the problem thatwe're all facing with large
language models.
But you can fix that by runningscripts and scripts that are
already defined and, uh, thedeterministic.
And then I have one that createsall the read me files, which is.
Files that explains the, thethings, right?
And then a, a one, uh, a factoryfor schemas, right?
Building schemas of, ofdifferent things.
CPS codes and stuff like that,right?

(32:33):
So, and then this is theblueprints that I was saying
there called the archetypeshere.
Uh, you know, which is one is apipeline, one is a journey, one
is a production, one is aprogram.
So the pipeline is my salespipeline.
So I made that a blueprint.
The journey, uh, the journey isthe onboarding, is the
onboarding one.
Uh, the production is the, um,content one, the guide one,
right?
And the program is my trainingsprogram.

(32:55):
And for, for archetypes, um, Ithink it's, it's quite good
because, you know.
Most things are gonna be likekind of a pipeline or kind of a
journey, kind of a productionkind of program.
If something doesn't match, Imight create a fifth one, you
know?
But these are my blueprints atthe moment, and I have, you
know, a three phase process.
Discovery is about 20, 30minutes.
And then, so it's not this, youknow, people think, oh, I'm

(33:17):
gonna build an agent.
It's gonna be, you know, and ifyou do that, you're gonna get
crap agents because you justbuild something that is very
generic.
But I spend, I spend hours onbuilding these systems and then
people are like, yeah, but youspend so many hours.
Yeah.
But I spent 15 hours buildingthat training system, or maybe
now about 30 hours that I'vespent.
And it's, you know, I've servedabout, you know, about was it

(33:38):
250,$300,000 of clients withthat system?
You know?
And it's like, it's okay.
I'll spend the, the 30.
Yeah.

Yeah.
The, the ROI is very obvious.
I have, I have a couple ofquestions.
Yeah.
Just diving a layer deeper toget people to understand what it
is so we understand it has.
Uh, agents and it has factories.

Tim Cakir (33:57):
Yep.

Can you walk us through what is an agent and how
you built it and what is afactory and how you built it?
I think prototypes are prettyobvious, and I think the phases
of production are prettyobvious.
Uh, I think the things thatpeople are scratching their
heads and like, okay, how didyou create this, uh, yeah.
Organization or questionnaireor, or matching agents.
Uh, and where, what did youcreate them with?

(34:19):
How did you create them andwhere do they live now that
you've created them?

Tim Cakir (34:23):
Yeah.
Yeah.
Um, just before that, maybeI'll, I'll just finish off on
the system.
Yeah, sure.
And then we will dive deep,because, because I absolutely
wanna show you that, but I wantpeople to understand that, you
know, in, in the front end, in anutshell, in an overview, it's
much simpler than, you know,then we might understand because
I, I think it's, it is quitesimple.
At the end, you're gonna see, ina second you're gonna be like,

(34:44):
oh my God.
He, he, he just made it sound socool.
And it's simple things.
They are simple things, youknow, it's processes, SOPs, and
then job descriptions.
You know, it's like I createprocesses and all these
processes go into a system andthat system, like, who's gonna
run the system?
Well, I need 10 people.
And those 10 people are agentsand each agent has a job
description.
Yeah.

(35:05):
It's that simple.
You know, like, I mean, itsounds simple when I say it like
that, but.
So my sales system, same thing.
You know, it's about sales andthen, you know, you can see
where things live.
I'm gonna show you the backendin a second.
My onboarding system, you know,and, and then, you know, I know
how to do these things and itknows it's a 14 day engagement
journey.
The onboarding, my training's a12 week, it has three phases.

(35:26):
It's all mapped out.
The guide system that we're justrunning, it's, you know, five to
10 page guides, two to threecase studies, five LinkedIn
posts, three email templates.
It's gonna do all of this aboutthat guide that we ask about
quote code.
Um, and this is only sevenagents running this thing.
Uh, it was about 36 agentsyesterday, and I, I, I kind of
optimized this and I fired, youknow, 29 of them again, you

(35:47):
know, uh, because I added a lotof new ones, I added these ones,
I added these five core agents,and, and then I was like, you
know what?
This is all my systems are muchsimpler and efficient, and this
was.
Kind of cumbersome.
It, it had 30, I don't knowwhat, and, and they were like,
it was a bit too much.
And I, I was getting 40 pageguides, which were, were very
good, but maybe it was, it wasoverkill, you know?
Yeah.
So I that, yeah.

(36:08):
Um, and then, you know, I builtthat guide system that can build
guide systems.
You know, this was one of mymost exciting thing, but I'm
gonna show you now, um, thebackend.
Right?
Um, alright, so let's go in thebackend again.
Uh, but this time, if you don'tmind, I'm gonna open, I'm gonna
open a visual studio code.
So I'm gonna open an IDE, right?
So, uh, where people code again,right.

(36:29):
But I'm not gonna code.
Yeah.
Right.
And why I'm gonna open this isbecause you can read files a
little bit easier, right?
Um, you know, and, and I want, Iwant us to read some files.
So I, I went into my.cloudfolder, right?
This is a folder that is in mycomputer.
And inside this folder I haveall my cloud code things, right.

(36:50):
So I have, you know, I have myguide system that we just talked
about, right?
And I have all the agents here.
So you were asking about theagent.
So let's, let's get the firstagent that we actually running
in the backend right now.
Right?
Which is the, the, the guideorchestrator.
Right?
So I just opened this up.
I'll close this to make it a bitcleaner for us and maybe, let me
see if I can make this a littlebit bigger.
Yeah.
So this is, this is an agent.

(37:13):
And, and people are like, what?
This is how, how, how does anagent supposed to look like
anyways?
Nobody knows, right?
Um, nobody knows.
So, yeah, so I think it lookslike this, you know, it, it has
a name, it's a guideorchestrator.
It has a description.
It's a meta agent thatorchestrates guide creation with
five phases and seven agents.
Creates focused five to 10 pageguides.

(37:33):
No fluff, never creates contentdirectly launches specialized
agents.
So this agent's job is to be theorchestrator.
It has some tools, right?
It has some tools like task Readthe Globe.
Like these are different, uh,tools, uh, that you can have in
your terminal.
So they can kind of, you know,create tasks, can read stuff,
you can find stuff, you cansearch for stuff and all that.
And then this, actually, thisone runs on sonnet.

(37:54):
So Opus decided that Sunnet wasmore than enough for this,
right?
Uh, and in some, you might seethat I have Haiku.
Haiku is, is a good fast modelby, uh, Tropic again.
Um, and, um,

yeah, again, for those of you who don't know, uh,
every time Op, uh, philanthropiccomes up with, with a new model,
so now it's 4.5, they areeventually, they, they don't
always come with them at thesame time, but eventually they
have three models.
Haiku is the smallest, cheapest,fastest one.
Then there's Sonnet, which isthe middle one, which is the one
they always released first.
And then there's Opus, whichOpus 4.5 was just released last

(38:26):
week, uh, which is theirbiggest, more capable mm-hmm.
Uh, and also most expensive, uh,model to run.
But because team is running on aprepaid plan, he doesn't care
how much it costs, because aslong as you don't run to your
end of your allowance, then youare fine.
And then you can run Opus asmuch as you want.

Tim Cakir (38:43):
That's exactly it.
That's exactly it.
And you know, sometimes I go tothese agents folder that I'm
like, oh, okay, let me switchthis to Opus, or things like
that.
But, but Opus helped me.
Opus 4.5 helped me, especiallyor not in the guide system,
sorry, in the latest system Ibuilt.
Opus 4.5 helped me in this one,I think it was Opus 4.1 that
helped me, or it was SUNNET 4.5when I did this, you know, and
it decided that was Sonet.
I'm fine with it.

(39:03):
The output is great.
Yeah.
So, and then as you can see, ithas a little bit about the
description, what it's supposedto do, as we said.
So this, by the way, this isquite important, um, because
this is like the tag, themetadata about this agent,
right?
You know, if you just, uh, if ana, if a model has to read
something, it will just readthis to understand if it needs
to use it or not.
Sadly, these gets loaded in thecontext window, but with closed

(39:26):
skills, which is a new feature,you can have skills like this
and you have this metadata thatit doesn't read anything else.
It just reads this and it knowsthat, oh, okay, I need to use
this skill.
Let me go read the other stuffso it loads.
I'll, I'll

pause you.
I'll pause you for one minute.
Uh,'cause I don't want to jumpinto cloud scale.
I probably need to record awhole episode about cloud scale
because it's, yeah.
Invite

Tim Cakir (39:43):
me again.
Yeah,

yeah, yeah, yeah.
But, but I be aware of what youwish for.
But, but Claude skills are,think about every automation
you've built, basic automationslike a custom GPT.
But think about packaging it ina way that it knows how to call
it just when it needs it.
So you don't need to go and copythe files and send it to your
custom GPT Uh, Claude Code says,oh, this requires this kind of

(40:07):
data analysis.
I have a skill that knows how todo it.
And for me, as an example, Ihave one that knows how to query
my podcast if I wanna knowsomething.
When did I talk about this thingin my podcast?
Uh, it knows how to do that,which otherwise it, it wouldn't.
Uh, or it has one that doesthings based on my, uh, brand
guidelines, or it has one thatdoes very specific data analysis

(40:27):
for me.
And like Tim said, the, theimportant thing here is instead
of.
Blocking your context window,which means it's consuming part
of the memory you can use inorder to do the actual ask.
You wanna do.
It does not unless it isactually required.
And then it goes and opens,looks, basically opens the book,
reads what it needs to do, andthen does what the book tells it
to do.

(40:47):
So these are cloud scales andit's very, very powerful and the
fact that you can run theminside of anything Claude,
including cloud code is, isreally, really cool.

Tim Cakir (40:55):
Absolutely.
Absolutely.
And I think that this is stillthe beginning, but you know, we
do have the problem of thecontext window, right.
The memory that I can remember.
Yeah.
It's very small still, you know,and, and you know, Gemini is
doing 1 million, uh, tokens and,and you know, uh, we do have,
uh, cloud models that are 1million, but they're not served
to everybody.
Uh, you know, uh, but we'resupposed to get, hopefully,
bigger contact windows verysoon.
We

have about 14 minutes, so let's, uh, let's go.
I will shut up.

Tim Cakir (41:20):
Yeah.
No, no, no.
I love, I love I love it.
And we'll give value, we'll givevalue very quickly now, so, so
this is an agent, right?
Uh, it has some principles, youknow, these are forbidden.
Don't not write contentyourself, you know, and, and we
don't get the pink elephant inthe room problem anymore, right?
'cause prompt adherence, youknow, if you say, don't do that,
it doesn't do it.
But before, if you said, youknow, don't do that, it's like,
oh, I'm gonna go do that.

(41:40):
You, she's like, oh, he

talked about it.
He probably wants

Tim Cakir (41:42):
you to do, but it's, it's, it's like my four and a
half year old daughter, youknow, you tell her not to do it,
she'll do it, you know?
But I think that AI now, likelarge language models that are
now maybe 7, 8, 7 years old,maybe not, you know, they, they
act a little bit more like theyknow that they'll get in trouble
if they do it, so they don't doit.
Yeah.
You know, a little bit better.
And so this agent also is awareof the other agents, right?
Because it's the orchestrator.

(42:02):
It's like, Hey, these are yourbuddies that you're gonna come
in and then you're gonna usethese buddies, you know, the
guide, researcher, the guide,the outline, and so on.
And then it has a phase, it hasa total time.
Yeah, it has a whole thing thatis like, it's like a process,
right?
It's like a, yeah.

So again, to make it simple for people who are not
looking at the screen, the, it'ssimple English, like there's no
magic.
No.
And it's just describing in awell structured way that I'm
sure you used AI to help youstructure to explain what this
agent needs to do, how to do it,and what it shouldn't do.
That's basically what it is.

Tim Cakir (42:37):
Just in plain English.
Like, like a job description.
Like you're hiring someone.
Yeah, you're gonna give'em a jobdescription.
You're gonna tell'em, Hey, whatyou can do, but you can't do,
what are the tools that you haveaccess?
Here's the logins, right?
This is your job.
Go do your job.
Right.
Uh, so this is the same thing,and I, and I recommend
everybody.
Um, you know, when you'regetting agents and so on, write
a job description, you know,like onboard your job, your
agents, you know, give them thesame quality that you would give

(42:58):
to a good team member thatyou've hired for a lot of money.
Right?
Um, that's important becausethat's, that's what sets you for
success.
Um, and, and then it has, youknow, the actions and so SOP,
right?
So a standard operatingprocedure, it knows exactly how
to create a project structure,discrete folders.
It has, you know, it tells itwhat to say when it's done.
And some decision trees, somequality focus and some things

(43:18):
that it should remember, right?
This is an agent Question.
Question about

this.
Yeah.
Just to verify, I assume youhave a, either a Claude skill or
something like that, that helpsyou write these in a well

Tim Cakir (43:30):
structured way.
Yeah, so it, it's not evenskills in most of the things in
writing, it's, my agents do areally good job because I'm
breaking things into different,uh, agents.
So I don't have one agent thatwill outline and write the whole
thing.
Right?
Yeah.
I have an agent that will do theresearch, will write it to a
context file.
Then the next agent that isgonna create the outline will
look at the, uh, output of theresearcher agent, right?

(43:53):
And then it's gonna create hisown context saying, Hey, like I
looked at that.
Here's the outline that I think,and then the next agent is gonna
take that outline.
That's that outline that is,yeah.
So the instructions

don't get too complicated and too cumbersome
because something very

Tim Cakir (44:06):
much more specific and focused.
You know, one is a good writerand then the writer I will say,
this is how I write.
Here's some examples.
Right?
And then I'm not eating up onthe context window neither.
And by another question, because

we're in a.
Id.
Mm-hmm.
If you wanna reference a file,so I said, this is how I write.
You wanna give it threeexamples?
Where do these example live?
Or you just give it a link toyour folder inside your
computer?

Tim Cakir (44:29):
Yeah.
So in, in, in a lot of cases, Ido have, uh, the path name to
the file Yeah.
Where it is, because it knows inthe computer.
Uh, but if, if you also wannaput an example, you could put
another file and say, here's thefile.
Uh, but also, uh, you can, youcan kind of embed it inside
here.
You could say, here is my tone.
You know, you could kind of puta section.
Yeah, here are the words I liketo use.
Here are the words I like.

(44:50):
Yeah, yeah.
Understood.
You can kind of do that.

No, but I was thinking more of, let's say you
create one that's writingproposals and you want to give
it examples of two of yourwinning proposals, then you'll
put a file to give it a pass tothe file.
Yeah.
Got it.
Okay.
Exactly.

Tim Cakir (45:01):
Yep.
Um, so, so this is an agent atthe end of the day, and let's
just go back, uh, a stage, youknow, so this is an agent inside
my guide system.
As you saw, there's a bunch ofother agents, like this is the
enhancer, right?
This is the one that is gonnawrite some case studies and
examples for the guide, right?
Uh, and is gonna ask a bunch ofquestions.
So this is the guide system,it's, it lives here, has agents,

(45:22):
uh, has some other things like aknowledge base, as you were
saying.
So, you know, here's my program,here's my company training
program.
Here's our blueprint, here's thecompany methodology, here's my
style guide.
Actually, as you can see,there's a Tim Checker style
guide.
There's a Tim Checker writingstyle, right?
And it knows, because in theagent you can say, Hey, I do
have some style guides.
Go read those, you know?
Uh, so you're keeping this in aknowledge base.

(45:43):
So the guide system has aknowledge base.
It has a memory as well, right.
So it writes in the memory, ithas some schema, which is empty
at the moment, has sometemplates, you know.
Uh, so workflow is statetemplates and project metadata
template.
So to write those things and soon.
Each time the same, these are,these are adjacent files, right?
Uh, and has, you know, here's,you should have the year, month
date, the topic, and so on.

(46:04):
So it kind of gives very smallkind of template structure.
Yeah.
Structure.
Yeah, that's a good word to putit.
Thank you very much.
Uh, and at the end of the day,the, the more important thing
here is the workflow, right?
So I showed you kind of, youknow, the, the beautiful thing
earlier, but at the end of theday, this, it's the five phase
workflow using the seven agents,right?
So this is why, if you rememberwhen I spoke, I said make sure

(46:25):
to use my guide creationworkflow.

Yeah.
Because

Tim Cakir (46:27):
then it does, it reads this first it says, oh,
okay, well I have this, or Ihave an agent that I'm gonna
first work with, right?
And then it's guide researcher,and it's the orchestrator that
works with it.
Okay?
Then I, it works with theoutline, then it works with the
right, but then the polish, thenthe market, then the
enhancement, right?
So this is basically

the handshake agreement between the different
agents, right?
That's what makes the contentmove from one agent to the
other.
Is this workflow.

Tim Cakir (46:51):
Yes.
Because it knows this step.
I'm gonna work with this guy.
Yeah.
The step two, I'm gonna do this,but I do something very, uh,
different than a lot of people.
And I think people really lovethat on my LinkedIn.
And the other day I showed is Idon't let an agent as, as we
were just saying earlier, Idon't let an agent speak to the
next agent because the onlyspeak

to the orchestrator,

Tim Cakir (47:09):
they only speak to the orchestrator, but they also
only write files to contextfiles.
Right.
Because the output, it would betoo much, and then the context
window would go, but because itwrites it to a context file, and
then the next agent, instead oflooking at thing, it just picks
up that context file that waswritten from the previous
agents.
Yeah, yeah, yeah.
Right.
And I'm able to keep muchbetter, uh, again, in

I I we're short on time, but I wanna explain what
you said because it's important.
The context window is yourlimiting factor, right?
This is the amount of data thatcan go in and out, out of a
single conversation.

Tim Cakir (47:40):
And it gets worse.
And it gets worse, right?
Yeah.
And if, if you take.

Yeah, let's say now it's the fifth agent and he
wants to know what happened inthe first four steps and has to
read everything that happened ineach one of previous agents.
It's basically gonna run out ofmemory before it can do the
thing that it needs to do.
And what Tim is doing, it'shaving, the only thing that gets
transferred is basically theoutput versus the entire thing,
and this is what he's callingcontext files obviously go, Hey,

(48:06):
here's what I created.
This is all you need to know.
Like you don't need to know howhe created it, what the data
that is.
You don't need to know any ofthat.
Yeah,

Tim Cakir (48:11):
just read this.
Here's what I created.

Use this as your entry point to do your task,
which saves a lot of space andefficiency and tokens and a lot.
There's a lot of other reasonswhy this is a good idea.

Tim Cakir (48:22):
Brilliant.
Exactly.
Yeah.
So that's a little hack thatI've been doing.
So that's basically one system,you know, it has, it has a
master workflow, it has agents,it has some memory, it has
different things, it hastemplates.
And this runs in this folder,right?
And I have a bunch of these, youknow, I have a memory system,
which this, this is basicallyevery agent and every time I run
quote code, this is somethingnew that I, because I read the

(48:43):
article a few weeks ago and Iwas like, oh my God, I need
systems that get better becauseI can't just keep, you know,
working on my system.
But instead, if I get feedback,this memory system takes memory,
takes things and saying, oh, Timsaid this, Tim, you know, gave
me feedback on here.
And then I'm building thismemory system because then the
next, uh, phase it's looks atthe memory system and takes the

(49:06):
learnings and does thelearnings, right?
Yeah.
And, and it kind of updates theagents as well as we go along,
right?
Yeah.
And so, yeah, so every system,as you can see, has same agents
bin schemas skills, you know,uh, and so on, right?
So, so they all kind of work onthe same thing.
All my systems, you know, havesimilar, similar things, and the
agents are, are just there.
Um, it's, you know, it, it'ssimple.

(49:28):
I say, yeah.
So

I, I wanna, I wanna really summarize this and then,
and then let people know wherethey can find you and follow you
and work with you.
But at the end of the day, wewent from, I know I want to do
something.
Mm-hmm.
To breaking this down to, okay,if I had to hire people to do
this, what would I do?

(49:49):
I will hire a person that willbe their manager because I need
five people and I don't want tomanage them.
Yeah.
And then I will need a personthat does this, a person that
does this, a person does this,and then you literally go in and
explain what each and every oneof them does.
Just like Tim said, it's a jobdescription.
A very detailed job description,because it's gotta follow that.
Exactly.
It's not gonna take anyinitiatives.
Uh, so you gotta tell it exactlywhat it needs to do.

(50:11):
As Tim said, the smaller youmake it, the better, because
then it's very specific and thenhand it off to the next agent
and it's not gonna get confusedwith what it needs to do.
And then create the plan of howdoes the handshake work once
you're done, who do you handthis over to?
And then when you run theprocess, you ask quote unquote,
the process to use this plan andthen use the agents in order to

(50:33):
complete the task.
And then what happens is.
Then the magic happens, right?
Because when then you give it atask, it already has all these
job descriptions and all thehandshake functions on how to go
from step one to step two, tostep three, to step four.
So if you think about everyprocess in your company, you can
mimic this.
If you can clearly describe whateach of the steps are and how

(50:55):
the steps lead from one another,uh, which makes it so powerful
and brilliant because you canapply this to literally anything
in your business, and it's all100% of it.
Simple English.
There's nothing here that youcannot do without writing any.
Code.
Yep.
Tim, this was nothing short ofincredible.

(51:15):
I think what you built is mindblowing, but as you said, it's
also very simple and I know thatpeople are gonna listen to this
first, like, okay, we, we lostyou in 30 minutes in.
Listen to this again, and you'llsee that there's nothing here
that is rocket science.
It's literally simple English,explaining what you want.
That will happen and breaking itand structure it in a very

(51:35):
well-defined way.
If

Tim Cakir (51:38):
I can add something.
I think it's about, yeah, yeah,a hundred percent.
It's about zooming out.
I was, you know, since, sincechat g pt, you know, which has
been three years and four dayscame out, I was very zoomed in
and I was kind of, you know,doing all these things, prompts
and so on.
And then I realized if we zoomout, I, and then we start
looking at this, as you said,it's much simpler as just a
business with processes, a bunchof process, system job

(52:00):
descriptions, and put'em inthere and then you'll give them
tools.
The model context protocol, MCPthat I mentioned are like APIs.
Imagine it's like a login.
You know, I can give it an MCPto HubSpot to my CRM and it can
update my HubSpot.
It can do these things.
And it's like getting asalesperson.
I hire a salesperson, I'm like,here's your HubSpot login.
So it's the same idea.
It is just we, I think,overcomplicate things, but when

(52:22):
we simplify things again, and AIis an incredible colleague now,
uh, for you to, to build thesesystems, you just have to very
much think about.
What do I do, right?
Yeah.
What does this team member do?
How do we do what we do?
You know, what do we do?
Right?
Because if you actually startasking these questions and you
can write them down, the AI willhelp you build it anyways.

(52:44):
You know, you don't have to, youknow, you don't have to be a
rocket scientist.

Amazing.
Tim, if people wanna follow you,work with you, learn from you,
hire you, what are the best waysto do that?

Tim Cakir (52:52):
Yeah, I think that's the best way is LinkedIn.
You know, Tim, check here.
Uh, T-I-M-C-A-K-I-R is LinkedInor my email tim@arbeta.com.
Um, you know, we do have, um.
Uh, training programs.
We build stuff for people and soon.
But we usually work withcompanies of at least 50 people.
So we work with, with kind ofmid, you know, SMBs, uh, nothing

(53:13):
under 30 or so people.
Um, because that's where wereally get, get, see, return on
investment.
Right.
Just FYI, for everybodylistening.
Um, uh, and my YouTube, yeah, myYouTube is a, is a new project.
I mean, it's been a year, it'sbeen something where, how I'm
sharing what I'm sharing today.
I share everything that I know,every failure, everything I, I
fix, I try.
You know, and then from thereyou'll find from my LinkedIn,
you'll find newsletter.

(53:33):
There's a lot of free content Iput there.
It's our, I wanna, I wanna teacheverybody for free, but I have
some clients so that they canpay, so that I can kind of also
share things for free foreverybody.
Yes, same thing.
Same thing that

I do.
Gender is absolutely amazing.
You are absolutely brilliant.
I appreciate everything thatyou're doing and what you're
sharing and sharing with ustoday.
Thank you so much, and thankseverybody who joined us live,
uh, and, and was in this, uh, inthe backend and listening to
what we're doing.
And obviously if you'relistening to this after the
fact, come and join us in thislive so we can ask questions as
well.
Uh, thanks again.

Tim Cakir (54:05):
Thank you very much.
Thank you for having me.

All Episodes

250 | The No-Code Way to Use Claude Code For Creating AI agents for Business with Tim Cakir

Episode Transcript

Popular Podcasts

Stuff You Should Know

Dateline NBC

The Burden

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}250 | The No-Code Way to Use Claude Code For Creating AI agents for Business with Tim Cakir