240 | AI Builder’s Playbook: Turn Your Idea Into a Working Solution in few hours with Pietro Montaldo - 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 Isar Metis, your host, andI am thrilled about today's
episode and you will be as well,and I'm thrilled for several
different reasons.
One, we're starting to go backto doing lives.
We used to do this every singleweek, and now I've been doing so

(00:27):
many company specific workshopsthat have been traveling
literally every single week inthe last, I don't know, six to
eight weeks, it seems likeforever.
Uh, so we were not able to dothis and we, I still have a
little bit of travel, uh, comingin the next few weeks, but after
that, hopefully we'll get backto doing this every single week.
So I'm excited because we'redoing this live, but I'm a lot
more excited because of today'stopic, because today's topic is

(00:48):
gonna solve some of the biggestproblems that you are facing and
that everybody's facing rightnow, which is deciding how to
implement with ai the solutionsthat you need for your business.
So we all have several differentsteps in our AI journey.
Step number one is we don't havea clue, and then you take.
My course or somebody else'scourse, and you start figuring

(01:09):
it out.
Step number two is like, oh myGod, what do I do now?
And then you start coming upwith ideas on how to implement
AI in your business.
But there could be 17 differentways to implement the thing that
you have as an idea.
Do you want to create a chatbot?
Do you want to create an add-onto your website?
You you wanna create anautomation?
Do you wanna build it in NA 10or in make?
Do you want to create a customGPT?

(01:30):
Does it need to have video?
Like all these kind of things.
And it's a mess for most peoplebecause most people approach
this for the first time versusour guest today that does this
for a living.
And hence he has a incrediblywell defined.
Decision tree, a process ongoingfrom idea to evaluating

(01:50):
different options, to testingthe different options, to
implementing the solution andactually using it because that's
what he does for a living.
So Pietro, our guest today isone of the co-founders of nForce
ai, which is a company that doesexactly this.
They help companies implement AIsolutions across the board,
different kinds, differentlevels of complications, and so

(02:12):
on.
And before that, in hisbackground, he has a.
Really interesting backgroundbecause he is done well, more or
less everything, uh, which isactually really, really awesome.
So he is a serial entrepreneur.
He had several business on hisown that he started, but he also
had roles as a product manager,which means he really
understands product and definingsolution.
But he's also been an analystand an investor.

(02:33):
And so he also understand how tobe very practical in looking at
data and figuring out what itmeans, which really makes him
the perfect person to now buildAI emissions for companies and
to explain us exactly how tofollow this process.
Now we're going to start on veryhigh level and go through the
steps, and we're also gonna showyou examples of the different
implementation.
So this is gonna be extremelyvaluable and hence I'm really

(02:57):
excited to welcome Pietro to theshow.
Pietro, welcome to Leveragingai.

Pietro (03:45):
Thank you so much, ISA.
That was such a niceinteraction.
Thank you for having me.

Yes.
I, I, I am, I I didn't say that.
Uh, joking.
I'm, I'm really, really excitedabout this session.
I've been following you, uh, fora while, and I know how amazing
all your content and howpractical your content is.
And again, I'll say one morething that, that I really like
about you, and I think that'sgonna help a lot of people
understand.
I think you're the ultimatetinkerer, right?
You like, I think you probablydo some of this for fun, right?

(04:10):
Just try stuff.

Pietro (04:12):
that's how I started, right?
I started'cause I wanted tolearn, and then I was like,
okay, while I learn, I might aswell make something with it,
right?
Which is definitely my f thefavorite, my favorite job I've
ever had.
You know, I literally pay myselfto experiment things, which, uh,
which is beautiful.
And one thing which I wanna addto your, uh, to your
introduction is.
One thing, which you might havenoticed that that brings

(04:34):
together all of the things whichhave done as none of them was
technical, right?
So I'm purely a hundred percentnon-technical.
So the approach you see today asan approach with, which is
especially beneficial forsomebody non-technical, also for
someone technical, which hasnever played with the ai.
It might be very interesting,but I think it's mostly
interesting for non-technicalpeople and on the side of the,

(04:56):
on the side of our businesswhere we develop a complex
solution.
What I do and my day-to-day asVictor going through this
process of experimenting withideas, but with the goal to
share it with a non-technicalaudience, right?
So what we're talking today isbasically how I go from, okay, I
had a very cool idea while I wasin the shower, while I was
working for a client.
How do I turn into somethingthat can inspire people, right?

(05:19):
I see my job a lot as.
Giving people inspiration ofwhat they can build and giving
them a first concrete step onhow they can turn it into
reality.
And I do it with LinkedIntutorials and with YouTube.
These are the two channels whereI work the most.

Yeah, fantastic.
And, and again, I think so manypeople are getting there, right?
I think we're going now from thestage of people don't have a
clue about AI to a lot of peopleknowing about ai, but don't
really know how to materializetheir ideas into practicality.
And that's why I'm reallyexcited about this.
So I'll give you the stage andjust, uh, lead us through the
process.

Pietro (05:54):
Perfect.
Okay.
I'm gonna go ahead and share myscreen for the people who are
alive, but I'm gonna drive youthrough every single steps In
any case.
Yeah.

And those of us who are joining us live, first of
all, thank you so much.
We're excited to have you.
I know we haven't beenconsistent in doing this, uh,
but I appreciate the fact thatyou're here.
If you have any questions,please write them in the chat,
either on the, zoom chat or onthe LinkedIn, uh, chat.
And if you are not joining uslive, then I suggest you try to
join us live next time.
Uh, you can be on our emailinglist and we are gonna send you

(06:23):
invites to every time we do, uh,these live, so you can be aware
of exactly when they'rehappening, and be able to join
us.
We do this almost everyThursday.
Again, if I'm not traveling,it's every Thursday at noon
Eastern time.
So you can join us, uh, thisway.
but again, back to you.

Pietro (06:38):
Absolutely.
So, where do I start?
Everything starts from idea.
One of the coolest thing, andone of the reason why I'm most
excited about AI is that I amnot limited anymore to what I
can build.
Because no matter what I'mthinking about my, what my idea
is, I know there is a tool thatallow me to do it.
I, without bringing on ourengineering team, right?
So one of the goal, which I havewhenever I build something for

(07:01):
myself and for sharing with theaudience and for inspiring
people, is I want to becompletely independent from the
technical team.
I don't want to bring anybodyfrom our technical team on
board.
I'm focused on building microops micro solution, which helps
you with your day-to-day, mostlybusiness life.
What does it mean, right?
what are some of thesesolutions?
Just to give you an idea whatwe're talking about.
Uh, one could be a screener forReddit post, which are

(07:23):
interesting for your company anda draft of the SEO optimized
comment, which you can leave onthat specific post.
Another one could be an agentthat on a weekly basis, go and
scrape all of the ads which yourcompetitor have done on Facebook
and Instagram, and send yourreport of the ads that have
worked the best for yourcompetitor.
Another option could be.
an engine to create contentwhich respects your content

(07:45):
styles, right?
So it's simple micro solution,which can really solve one task
of your day-to-day, or canenable you to add an additional
task, which before you wereunable to, for example, a lot of
people before were unable topost on link, right?
And why do I think that link hasgrown so much in the last, uh,
six to eight months?
It's because a lot of people areempowered to do it without so

(08:05):
much effort, right?
Because they can start from anidea, from a call they had and
turn it into a proper piece ofcontent which they can share,
which has, which literally hasactual insight, right?
You don't need to make it upwith ai and that's not gonna add
value or bring you anything.
But you can turn your realinsight and what you have into
something that works.
Very quickly, right?
So everything starts from idea.
When I have an idea of somethingI wanna build, in my specific

(08:27):
case, I look through the lensesof, okay, I wanna turn it into
something.
I wanna educate and inspirepeople with, so I, I look for,
for some specificcharacteristics.
For me, an idea has to beappealing for a very large
amount of people, right?
Because that's, that's what goeson social media, right?
This won't be the same if, forexample, you are studying an
idea to apply it in your day today.
You just need to look for a painpoint and you can build an idea.

(08:49):
For me, an idea also has to beappealing for a large amount of
business people.
The second thing, and probablythis is the most important one,
it should have been firstactually, is it has to be truly
useful, right?
I'm not gonna publish somethingwhich sell smoke.
I wanna publish something wherea person try it and say, oh,
this guy has done a very cool,simple thing.
Let me see what else he can do.
Right?
And then they book a call withme and we turn that those people

(09:10):
into a customer so much easierthan anybody else.
The third thing, it has togenerally be relevant for sales,
marketing, operation, orpersonal productivity.
These are the four largestbuckets of, uh, non-technical
people and also luckily, wherecompanies are happier to invest.
because it's the one that havethe highest simple on the, on
the top line.

(09:30):
And then the other thing whichI'm looking for is solutions.
Master acquire only simpleintegrations I can explain in
under a minute, right?
I am not looking for complexintegration.
For example, if you're workingwith Salesforce and you have a
very personalized sense withinyour company, I would recommend,
okay, go and get an agency thathelps you set up a proper
automation or what directlywould sales for.
I'm looking for things which canbe integrated, quickly.

(09:53):
I can explain it for anon-technical person and you can
understand and replicate it.
And I usually aim for a maximumto integration per solution.
More than that, it becomes amore complex solution, which I
can describe for you, but Iwould wanna build it for you.
I would wanna bring an engineerto make sure it can go into
production.
And the fourth rule, and this issomething very personal for me,
every solution should take lessthan four hours to beat, right?

(10:14):
So if it has.
Been more than four hours.
I put in a bucket.
I hope one day I'll have someoneto help me in this process and
I'll, I'll get them out and I'llinvest in those.
But I want to push out minimumtwo new things per week.
That's how you have, you canstay up to date and this, and
with how fast the work move.
So I can only apply for our per,per solution and this is
fantastic.

2, 2, 2 quick questions.
Pietro, one is, how do youdocument your ideas?
Do you voice type them?
Do you write them in a document?
Do you put a, a voice note onyour phone?
So that's question number one.
And then I'll ask you questionnumber two.

Pietro (10:48):
Yeah.
So when I'm on my laptop, andmost of my idea unfortunately
comes when I'm on my laptop.
'cause I'm on my laptopapproximately, I don't know, 10,
10 hour, 15 hours per dayminimum.
Uh, I have an idea database innor, and I just add them there.
I classify if it's a leadmagnet, a build or idea for
another post.
But I have a nicely, um, anicely built, uh, knowledge
database where I put all ofthis.

(11:09):
Uh, so you do it in Notion,

so anybody can do this in Monday, JIRA, Ana,
Clickup, like any of that.
Anything, literally

Pietro (11:15):
anything.
I had it in Clickup.
I moved it to Notion.
Absolutely.
Same thing.
If there's a CanBan perfect.
you can do that.
Yes.
Okay.
That's, you don't need anythingelse.

Second question is, yeah, is, you talked about the
four hour limit.
How do you guesstimate how longit's gonna take you?

Pietro (11:28):
In general, the rules which I have described, are
thought for it to be withinthose four hours, right?
If it has less than twoconnection, if it's of limited,
uh, complication, it generallyshould respect those.
And also I get better.
I like to think I get betterover time.
So now I rarely invest timewhere I get to the four hour
mark and I'm like, I couldn'tfinish.
Right.
I don't, I don't remember thelast time it happened, and at

(11:50):
times it's just because maybethere are still some things that
require engineers to come onboard and maybe they just,
called an engineer or gave upbefore.

Okay.

Pietro (12:00):
Okay, so moving on.
The idea phase is the coolestone, and you can have them
wherever you want.
And when you have them, justnote them down.
If you're around, I use a, anote on my iPhone, and then in
the morning I have a blocker totrans transfer everything I
taught during the night to mylaptop, right?
Yeah.
Now the question is, and this isthe coolest part of this
workshop, I think, and I'venever done this, so bear with
me.
I like, I are helping make thisclear for everyone.

(12:22):
how do I turn this into aworking solution?
So all of us in general use AIin three main ways.
We use them in copilot, which iswhen AI is embedded within
another product, helps you makebetter use of that specific
product.
I'll show you a couple ofcopilot later.
we use Via Chat, for example,like use, uh, charge GPT or, or

(12:42):
CLO or any other LLM, and we usethem as agent or workflow.
In general, when we talk aboutagent of work, we are talking
about something that can do workon your behalf, right?
Yeah.
I like to think of two maincharacteristics of agent to
distinguish what, uh, what is anagent and what is just a
chatbot, right?
An agent is proactive so it canstart working on his own.

(13:05):
For example, if he, every Mondayhe starts working on the
specific top on his own, withoutme going down and telling, okay,
do this would, would beclassified as agent.
And also if it's not a singletask, but it's repeated any
performance as a set of actionsrepeatedly and proactively.
This also means, for me, it's anagent because basically the two
things work together.
He works on his own and herepeats the action on his own

(13:27):
based on when you receive acertain input.
It can be time, it can be, uh,an email received.
It can be, I dunno, taskcompleted or not right?

Yeah.
And any trigger of somethingthat happens in your tech stack.

Pietro (13:39):
Exactly any trigger.
That's something that happens ingeneral.
and I'm not sure if theory ofagents also confirms this, but
when the tool is integrated withmy, with my tech stack, I also
consider that to be an agent,right?
If it's integrated and if it'sproduct, if it's integrated,
it's repeated, it becomes anagent.
We're not talking anymore abouta chatbot, but we're talking

(13:59):
about something that isindependent and can perform work
on your behalf, right?
And this is the biggest and, andbest revolution of ai, right?
You literally can free up timebecause an agent can go and do
work on your behalf where yousit at the desk and I already
have the idea ready for, uh, myLinkedIn post, or I have a
already have my YouTube scriptdraft.

(14:20):
so once I respond this question,I know which direction I want to
go.
If I responded, okay, yes.
This is not an agent.
'cause it doesn't respectspecificity, it's probably falls
under the chat idea.
when I have a chatbot, when Iknow it's a chatbot, I have a
series of choices I can do,right?
And the most important one iswhat interface do I want to do
to this chatbot?

(14:40):
A chat can be something veryuseful, right?
It might sound diminishing nowthat I mention it this way, but
a chatbot can be incrediblyuseful.
For example, I can give all ofmy knowledge, uh, about how to
write a good hook for a, for alink in post.
And put it into, into a chatbot.
And the chatbot only performthat function.
That function is incrediblyhelpful.
I've spent, I don't know howmuch time crafting the way I do

(15:01):
the hooks and now I have achatbot, will helps me get to a
hook so much faster than before.
Yeah.

Yeah.
To, to put things inperspective, uh, we're talking
about chatbots.
We're not just talking aboutfree conversation, uh, or open
conversation, but we're talkingabout tools like a custom GPT or
a cloud project or an assistantby OpenAI that can then work.
so it's still a chat interfacethat is a dedicated chat
interface through one of thelanguage models.

(15:29):
But, it, it can haveinstructions and can have a
knowledge base and it can knowhow to follow things.
And it's, I probably have.
25 or more, either custom GPTsor cloud projects that we use
every single week in the companyto do stuff.
So yes, there's no reason to.
I'll say something else.
Anything you can solve onsimpler solutions, go to the

(15:52):
simpler solutions, right?
So don't build a more complexsolution if the simpler solution
can do that.
And there's a lot of things thatyou can build with these
chatbots today.

Pietro (16:00):
Completely agreed.
Right?
Which is very important to gothrough this step, right?
Is this an agent or not?
If it's not an agent, there's noneed to make it one, right?
It happened to me a lot of timethat I complicated things very
much.
I use different models, I dodifferent steps just to realize
that I have a knowledge base andI wanted to give it to an
interface which can think basedon that knowledge base and gimme
a response, right?

(16:20):
That makes a chat, right?
Yeah.
And you can also make a chatwith fancy, right?
Like you mentioned.
It can be a simple chat and itcan be like, I'm, I'm opening up
here on the screen for thepeople who are here live and
will see this on YouTube, onchat, GPT.
It seems absurd, but most peopledon't know that you can create a
chatbot, which you can shareexternally within inter GPT,
right?
Mm-hmm.
This is called, gpt, right?
So if you go to this section,these are all GPTs, which are

(16:43):
either rent from someone else,you don't have to pay, but you
use them and somebody else hasbuilt someone else who has
knowledge on the space, forexample.
This YouTube script generator.
It's not made by me.
It's very good.
And I use it a lot.
Actually.
This was made by me, but therewas another one, uh, which
wasn't made by me.
You can see here that this wascreated by Xigen Asset, for
example, and creates really goodhooks.

(17:04):
But then I have some other, um,some other G GPTs boats, which I
have created myself.
For example, this, this is a, aReddit blog post architect,
which works really well.
I've put everything I knowinside this, and if I go here
for example, and I go to add G,this GPT, I can see what is the
configuration I gave to thisspecific GPT, and I can also

(17:26):
edit it.
And I edit it with a copilothere, for example here.
That's a tax base that helps me.
Get to a better version of theGPT than what I will do.
Just typing myself, right?
So this is what we were sayingbefore, a copilot AI embedded in
another AI product that helps meget to better results first,
right?
So this, I have done many, manyposts and some were very

(17:47):
successful.
Where the only thing I did as Iput my knowledge into a G pt, I
tested it for a week.
If it was nice and I thought itwas giving me nice results, I
publish and tell people, okay,this is, a chat bot that does,
except on that I tested it foreach week.
These are the results.
See one result here.
And I always past the result.
And this is something that.
People are very interested withbecause they might not have the

(18:07):
time, the bandwidth, theknowledge, or three days to
invest in researching how tomake a really good red box,
right?
So if you have a specificknowledge about something, you
codify it, you put it in achatbot, and you either sell it
or you give it away for whatevergoal you want is the lead magnet
or to general leads, it becomessomething very powerful and also
a simpler possible solution youcan create for yourself.

(18:28):
So, to someone who has nevercreated anything with ai, I
would tell start by creatingGPTs or Cloud project.
This will be on any LLM you go,you create a personalized
project within A TLM, and youhave a product which works,
create value.
You can sell or you can, uh, uh,use internally, pass it to your
team, your teammate, so you workbest and it takes you a morning

(18:49):
if you already have theknowledge.

Yeah.
Sometimes less.
So I, I, great, great.
Concept here.
I want to add 2 cents.
One for the people, like, oh,how do you get somebody else's,
uh, GPTs?
if you just open the regularchat GPT interface, the first
thing in the GPT section iscalled Explore.
And if you click on that, thenit will take you to what's
called the GPT Store, eventhough it's not really a store,
because the, you don't pay foranything.

(19:11):
It's a free store.
And then you can just either goto the different sections, they
have productivity and researchand analysis and data and
marketing and like all thesekind of things.
Or there's a search bar on topand you can literally just
explain in English what you'relooking for and it's gonna show
you.
They now have multiple millionsof these gpt that other people
created.
Some are great, some are not sogreat.
Uh, but you'll have just to testthem out and see how they work.

(19:32):
So that's on that one aspect.
The other aspect, is on the wayI teach people on how to build
those custom gps or projects, Ialways tell people, start with a
regular chat.
So don't try to build a GPT.
Start with a regular chat.
Explain to Chachi, PT or Claude.
If you're in Claude, what youwant to do.
Like this is what I'm trying todo.
This is gonna be the input.
This needs to be the output.
Here's the data that I have,here's the knowledge that I
have.

(19:52):
and work within the chat untilyou get to the outcome that you
want.
So basically go back and forthand back and forth.
Try to write if, if I'm gonnafollow Pieros, uh, example, I
wanna write great Reddit posts.
So write it again and again andagain until it works correctly.
And once it works correctly,basically tell Chachi Pet
Claude.
Okay, I want to turn this into acustom GPT.
Please write the instructionsfor me.

(20:12):
These are the files that aregonna be in the knowledge base.
Uh, please write theinstructions and it writes
incredible instructions.
Way better than I can write onmy own.
Uh, and it's gonna be based onyour back and forth in trial and
error that you've done in thatspecific chat.
So this is probably the fastestand the most effective way to
create good, custom GPTprojects, whatever you want to
call them.

Pietro (20:33):
I completely agree.
and then if you wanna bring itone step farther and you wanna
give a chat put or a very simpleapp, a phase, which is nicer
than just a chat, you can do itusing.
Lovable or Base 44 or Vault.
So what are this?
These are web builder and appbuilder which allow you to
create an app just with simpletext, right?
So I have some knowledge, again,I have it codified and I want to

(20:55):
create an app that tells, let'smake a complete example.
This is something I did a fewweeks ago and it's a tax saving
calculator when you move toanother country in Europe.
And I, I collected out ofknowledge'cause I had done a lot
of research'cause I wanted tomove, uh, to a better tax
location.
I codified the knowledge.
and now I have a chatbot and aninterface, which very simple,
uh, u UX and UI that helps meunderstand and engage with

(21:19):
people better, right?
So, okay, I'll do a chatbotfirst.
I'll do a GBT project, I'll do achat that helps me with this.
I go to a very good result.
Lemme turn into something a bitmore visually appealing, right?
And then you bring it unlovableon base 44, which is what we're
seeing right, right now today.
And I turn it into something abit nicer.
This is another step and it'smore of a marketing step.
'cause the very important thingis the knowledge.

(21:41):
But it's, I have built it.
Okay?
I liked it a lot.
Lemme try and bring you one stepfurther.
At least I see it that way.
You can also start directly withbase 44 if your goal is to
rebuild your website, forexample.

Yeah, I agree a hundred percent, right?
So again, for those of you whodon't know, lovable base, 44 rep
lit, uh, there's a bunch ofothers.
There are basically what'scalled vibe coding tools, which
means they can write code and doliterally anything you want, but
they can, they're mostly used tocreate applications and you can
start with a.

(22:12):
Somewhere else, like a regularchat or a chatbot or a
something, and then give it allof that and it will build the
application for you.
And then it can help you even inthe design and look and feel.
If you wanna know more about theoverall flow.
I created an episode not toolong ago was episode 216, that
was called the Five Levels ofai, what Every Business Leader
needs to know, and it basicallystarts with a chat to then

(22:34):
these, uh, more advanced, likecustom GPTs from there to
building automation from thereto vibe coding.
So if you want to dive a littledeeper into that, you can go and
check out episode one 16.
But to summarize what we've,what we've said so far is we
said, okay, if something is nota chatbot, meaning, sorry, if
something is not an agent,meaning it doesn't need to take
action and it doesn't need to beproactive, uh, it just needs to

(22:55):
follow a flow based on theknowledge that I have, we went
down that path that can eitherbe used inside of Chachi PT or.
Build a wrapper around it withlike a vibe coding tool.
So now I assume we're gonna goto, okay, if we do need an
agent, what do we do then?

Pietro (23:08):
Correct.
Correct.
And I love how you remember theexact episode in which you
talked about that specifictopic.
that's some level of commitment.
And so now, thank you.
Now let's go to, we respond yesto the question where we, before
we have said, is it proactive,is it repeated?
Is it integrated?
If you respond yes to thisquestion, you probably want to
build an agent, which is aslightly complex system, more

(23:28):
complex system.
if we ever responded to this,what I think about next is, is
it something, is it somethingthat someone has already
productized?
and is that product really good?
At times myself, I really liketo be things from scratches,
right?
So I have to fight this urge tobe d from scratch.
And I think it's very useful ingeneral to look off what is,

(23:50):
what's already out there, right?
There might be people who havealready prototyped this to a
point where it works incrediblywell, right?
Some example of this, I dunno,uh, sending colima, selling call
emails.
It's a huge market and somepeople have been starting to be
there in the last six years,instantly, smartly, for example.
They have an incredible amountof knowledge and they build

(24:11):
really powerful agents for, fordoing that, right?
So rebuilding those, you wouldmost likely build something
which has superpower and alsothis solution are not incredibly
expensive.
Another question I ask myself,if somebody has built it and
it's really nice, is it reallyexpensive?
If it's really expensive, can Ibid something which can do the
same thing or very, very similarfor much, much cheaper so that

(24:33):
people can basically do itthemself?
If the question is yes, I willstill go ahead and bid it.
If the question as the productdoesn't exist, I will go ahead
and build it.
If the question is, yes, thereis a great product.
It's cheap and people know it ordon't know it yet even better,
but they can access it veryeasily.
I will go ahead, try the productand probably showcase six

(24:54):
tutorial about that productright when I have an idea.
And I think it's cool and Ithink it's appealing to people.
I want to just give up.
it's not like because I can'tbuild it myself, I want create a
resource on it.
I still think it's cool torecommend people to go somewhere
where there's already somethingexisting to recommend to another
team who has done a really,really good job about it.

Um, a hundred percent.
Quick question about this.
Where do you search?
Like, do you just go on Google,do you go on Reddit?
Do you go on.
hugging face and, and foul.
Like where, where do youactually search for existing
solutions that you can integrateinto your processes?

Pietro (25:27):
It's a very good question, right?
Like I, my question is verybasic and probably
disappointing, but look, on chatg, pt, right?
I tell him exactly what I wannado and I tell him to recommend
the top best product.
The next step will, before Iinvest time or especially money.
Into testing a solution as I ampart of approximately a hundred
thousand groups of AI builder,and I have two or three, which

(25:48):
I, which I really like.
It's more intimate and it'speople who I know I can trust or
I can rely on, and I ask ifsomebody has tried it.
Nobody, if somebody recommendssomething, I'll start there.
If nobody has ever triedanything, I'll just test
platform.
Most of this platform now have afree trial that allow you to
have an idea for yourselfwithout, uh, having to go to
expensive plants or have very,very cheap plants.

(26:09):
Right?
So most likely I can afford totry it.
And then usually a productthat's wider.
So for example, if I'minterested in Reddit,'cause now
I have this fixation withReddit, there are not many
product about Reddit, right?
And a product have severaldifferent features.
So I might create three, fourapps that either compete with
them or, or I can create.
Beautiful use cases based onwhat they do.
Right.
It's never only one thing.

(26:30):
Right.
So when I try something and it'sreally nice, I generally come up
with more things I can do withit or I can come up with more
things I can integrate withsomething I've already done.
Right.
So good things I think arereally great, especially in this
time where everybody can buildsomething right?
Good things for me to follow up.
They have to be really, reallygreat.
And when they are, they'reincredible and most likely use
them for, for a long time.

Awesome.

Pietro (26:52):
So if we have responded yes to the question, do we wanna
build it ourself because of allof the reason which we've
mentioned before, the next stepis it's very, very interesting,
right?
The next step is the mostcomplex one and the one I had
the hardest time also tocodifying to the small uh, uh,
graph, which is on the screenand which are now present.
The question you have to respondis, what capability does my

(27:13):
agent require to function andactually provide value to the
user right at time?
The idea I start with is not theidea I finished with, right?
'cause it was more complex, ithad more features and I, but I
need to make sure I end up withthe true nature of the idea.
I started, right?
There was one value pillarbehind the initial idea.
Did I get to the the point whereI can solve it or make it

(27:35):
exponentially better?
If, yes, okay, I will continueto build.
There are a thousand platformswhich can help you create agents
and if you search which agentsplatform, I can, uh, and I want
to create, you'll find anincredible amount of option.
All these platforms are veryversatile, but some I find
personally, and again, this is apersonal opinion.

(27:58):
That, not the, not all of themmake it exponentially easier for
you to develop a solution whilebeing a non-technical person.
Right?
So now I'm gonna talk biasedlyand I'm gonna talk as the
non-technical person I am, I canbuild on time.
I have tried a multiple timesmake, I can build very, very
simple solution.
It was very complex for me toget to a point where I was

(28:19):
confident presenting anything,to a customer with, with Anan
while doing it myself.
And at one point I just thought.
This is just not for me, right?
I can get better, I can learn, Ican do very extensive courses
about the naan.
I can fund them on online forfree.
This is just not thought for me.
And there are people whoinvested years now and
understanding how an naan worksand they got really good, even
why not being technical.

(28:40):
But now, today as a starter, Iwould never recommend a starter,
non-technical person who'sstarting now.
I would never recommend, okay,learn an A 10 from scratches,
right?
Because other things, more AInative, and when I say more,
it's because they're just bornlater.
When there were new instrument,new interface available, make it
exponentially easier to get toan equal better, uh, results,

(29:01):
which just makes a, a no-brainerfor me, right?
So when I'm talking about thisplatform, I have three platform
in my mind, right?
These are the top three, which Ialways consider when I wanna
beat something.
And it's because they brought.
The agent building activity to alevel which has, which is, can
you, you can think it, so youcan put it into practice.
I have a specific set of toolswhich you can use if you're

(29:23):
looking for these tools.
Come here, right?
I'm talking about three apps.
Again, this is my personalopinion.
I just like them and I, I amonly presenting them'cause it's
what I use and what I think Iwould recommend to some, a
friend who will come and ask me,okay, I want to bid here and I,
the first one it's relay up forme.
It's probably the simplest andeasier for starter where you
wanna get to good results.

(29:44):
I think this is the bestpossible solution for
non-technical starter users.
The second one is Lindy.
Lindy is also very simple, verybeautiful.
They have like a super powerfuluser interface and it has more
functionalities and moreintegration.
Slightly more expensive but, andnothing crazy.
And then there's a, the thirdone is relevance, which again

(30:05):
has a great set of feature.
But I find it slightly morecomplex.
Something you can get over inwith a week, right?
When someone who comes and willbe the first step, I would
always recommend relay up.
And if I have the opportunity ofchoosing between the three,
'cause all of the three can doone activity, I will probably go
for relay.
But again, this is also mybuilding philosophy, my building

(30:27):
logic, how my brain works.
I would always recommend try thethree of them.
They all have a free trial.
See where you can build the bestthing and keep them all there.
You know?
There's also no reason to chooseone of the beautiful things, not
having your own platform andinfrastructure in this time
where everything changes so fastas you can just go and use the
best technology out there thismonth and next month it might be
a different one.
And you need to adapt, but youdon't lose money invested in

(30:49):
infrastructure.

So quick, uh, quick note about, about that, about
all these platforms.
first of all, they're like, youhinted, there's slightly
different in nature, so youhave.
You know, make an NA 10, whichare more flow chart oriented.
And then you got relay andIndian relevance, which are more
conversational, and at the endof the day, they do the same

(31:12):
thing, right?
There's like a flow that youdefine, but, but it, but the
user interface makes a lot moresense.
And I agree with you a hundredpercent.
I think NA 10 is extremelypowerful.
It is 100% not a tool forbeginners.
It is way more complicated.
It is way harder to connect itto some of the tools, uh, that
in, in, uh, just to give asimple example in all the three

(31:33):
tools that you mentioned, and inmake.com to connect to Gmail,
you give it your login andyou're done and in.
And then you gotta go to GoogleCloud and create a new project
and assign it, uh, and assignit, access to different things
and connect it to differentprocesses and APIs like it's a
whole.
Different other animal that youneed to dive into to get, if

(31:54):
you're doing basic stuff, noadded benefit.
because there's, like, on the,on most things, you'll be able
to do it with Relay, relay orrelevance or, or Lindy or, or
make without having to gothrough all that extra effort
and without being, uh, sotechnical.
So I agree with you with that ahundred percent.
And the last thing that yousaid, I agree with as well,
since Relay is the easiest ofthe three, if you can do it in
relay, do it in relay.

(32:15):
Just like what we said before,if you could do it in Chachi pt,
do it in Chachi pt, don't go tothe next step.
So always go to the, the paththat is the easiest for
implementation and don't lookfor really complicated, fancy
solutions just because they'refancy.
And the other thing that I willsay that you said a little
earlier that is very, veryimportant as somebody I spent
most of my career runningsoftware companies.

(32:36):
So for almost 20 years I waseither a CEO in senior position
in software companies, feature.
Creep is a, is a real deal,right?
You started with an idea aboutlike, oh, what if I had this,
what if I change that?
Uh, what if I can, uh, use itfor the, and then you forget
what you built this thing for,and then instead of investing 20
minutes and having the thingthat you wanted to build, you
spend two days trying to addfeatures that may or may not

(32:57):
work that provide 2% value, butthey're cool.
And so, sticking to the reasonwhy you wanna build the thing is
a very, very big deal.
And I cannot even tell you howmany projects I've started and
didn't finish because I wastrying to make them fancy versus
building it in 30 minutes anddoing the one thing that it
needs to do.
And then if you want, build theextension of that the next day,

(33:18):
like build something else thatdoes second step instead of
trying to merge them into oneapplication.
great points.

Pietro (33:25):
Thank you.
and then going to the next step,trying to make this a bit more
concrete.
The, like, my goal here withthis last part where we'll
finish, is how do I make thischoice a bit more concrete,
right?
I could narrow it down and theseare not the only point of
consideration, but are the mainones the most important one,
right?
So the first question which youwanna ask to yourself when you
were to pick any platform, be itanything which you have

(33:46):
mentioned, or anything which youhave tried that we haven't
mentioned is, does this platformhave an integration which I
need?
For example, say you need anintegration with Yahoo Finance,
right?
Who has an already madeintegration with Yahoo Finance.
No matter like how complex,complex the integration actually
is, it will be exponentiallybetter and easier for you to get

(34:07):
to a good results if theintegration has already been
built rather than if you have toset it up yourself, right?
So for example, for a YahooFinance, no other platform,
which we have mentioned has it.
So what you want to go for isstuck ai.
I was, I was trying to do thisand the only platform I could
find, maybe there are others,but it was stuck ai.
Personally, I haven't used it,right?
I had to set this up manually'cause I, I didn't wanna open a

(34:29):
new platform, but it would beeasier to go there for somebody
really nontechnical who doesn'thave access to engineering.
If I want to integrate it withLinkedIn, Reddit, I've tried a
lot of integration.
The one that really has, it'sincredible, right?
Anything that has to do withrelay edit, I was like, it's so
exponential, easier to go torelay than anything else.
And then, for example, if youneed integration with Epi
Appify, which allows you toscrape a lot of pages and access

(34:52):
a lot of very interestingthings, the best integration I
could find and the easier it'sbeen made.
It's irrelevance, right?
So the point is, wherever youcan find the best integration or
the existing integration withwhat you're looking to do with
the specific agent, go there.
The second point is.
If you want to speci perform avery specific action, and when I

(35:13):
talk about a very specificaction is, for example, doing
phone calls or sending SMS orscraping complex websites,
right?
A very specific action, which isa bit outside a normal workflow.
You also want to do the sameactivity, right?
So who can do phone calls?
Really can't.
Relevance?
I'm not sure.
I'm gonna say can't, but don'tquote me on that.
It might.
Lindy is very good at makingphone calls, right?

(35:35):
So you wanna look for a platformfor beginner, a platform that
has integration with YahooFinance and a platform who can
make phone calls.
You'll probably end up, okay, Ionly have one option here
really, on what to build.
And there's always thealternative, okay, we're getting
too complex.
There is two specific things.
Lemme go to an agent that bidson an A 10 or any other more
complex system, and I have a bidfor me, right?

(35:56):
But here what we're trying to dois we're trying to imagine a
world where we have no contentwith engineering and no budget
to hire, right?
So this is the work we would do,which is what I do on my day to
day to create content and forexample, what do I wanna say
anyway, um, the third thing.
Which we wanna do.
And the third question which wewanna ask, ourself is kind of
similar to this, the home onethat I see it written.

(36:18):
And it's when you want to have aspecific feature inside the
platform, like building abrowser extension or expanding
memories.
Now that we talk about it, Ithink this is not much
different.
Second point, right?
You're always talking about somespecific capability, some
specific actions, some specificinterface with which, the person
interact with the, with the end,with the end agent, or that the

(36:38):
end agent has to have, right?
So for example, here I wasthinking more about memory.
Okay.
We need the agent to havememory.
That's something I couldn't getto a good level with, uh, with
really, but I could get to areally good level with the, with
relevance, right?
Or I was looking to do a, createa browser extension, uh, that I
could talk with based on aspecific set of knowledge.
The only thing I could find wasMind Studio.

(37:00):
Mind Studio is very good at thatand they're focusing exactly on
that, right?
So basically here what you'resaying is you wanna ask for what
integration we have and whatspecific action it needs to do,
And are must have for thisagent.
And based on those, you'll pickthe platform that better suit
you.

Now, I'm not sure.
Awesome.
I want, I wanna do a quick recapfrom beginning to end of the
entire, kinda like decisiontree, uh, because I think that's
gonna help people put it alltogether.
So we started, we had an idea,and then based on the idea we're
saying, okay, is.
Does it require an agent or not?
If it does not require an agent,create a chatbot, which chatbot
you choose doesn't reallymatter, uh, because they're all

(37:39):
kind of the same.
Not exactly.
Again, this goes back to if youneed specific functionality that
exists in, Claude and does notexist in chat PT, as an example,
go to Claude and not to chatt,similar to what we discussed
right now.
Uh, and then if you wanna wrapthe chat with some other user
interface, you can do that aswell with the Vibe coding tools.
So this is number one.
If you do need an agent, then aswe said, the first thing you

(38:02):
wanna check this, somebodyalready built this.
Does that functionality exist ona third party tool that I can
use either the fullfunctionality or even partial so
it knows how to process this andthat data.
It doesn't gimme the output inthe format that I want.
I can still use that first stepto do all the data analysis and
collection and I can create.
My layer of output on top ofthat, it will still save me a

(38:22):
lot of time from doing theentire effort.
So look for something thatexists.
Uh, and Pietro said you can justuse your Chacha PD to ask what
exists and what would be bestfor your specific scenario.
Just define your scenario verywell.
And then if you decided todevelop it on your own, then it
comes down to picking the righttool that you wanna develop this
on.
And then you pick the tool by a,does it have the integrations to

(38:44):
the tech stack that you want toconnect to?
So does it connect to Gmail?
Does it connect to your CRM?
Does it connect to LinkedIn?
Doesn't matter, whichever thething that you want to connect
this to.
The second question is, does ithave a built-in functionality or
capability that I need in orderto make my life easy?
And I'll give you an example of.
And I'm actually gonna give anexample and add something, but

(39:06):
let's say, you wanted to, uh, beable to do data analysis.
So some of these tools do notknow how to do data analysis.
Like you give it an Excel filethat doesn't know what to do
with it.
Some of them do.
And so if you need somethinglike this, then find a platform
that knows how to do that, andthat's gonna save you a hell of
a lot of time.
Now, the last thing that I willsay for people who are not total
beginners but have been playingwith this for a while, but still

(39:28):
struggle to go to the slightlymore technical level, and I'll
give you a little bit of a trickand we're not gonna dive into
that because I do want Pietro toshow you a few examples, but I
think it's gonna help a lot ofpeople who are like in the
transition phase between a totalbeginner to a more advanced
user.
All these platforms have theability to connect to more or
less anything.
And they do this through eitherWEBHOOKS or HTTP calls.
Web hooks are like triggers thatcan connect to any other

(39:50):
platform.
And HTTP calls basically allowyou to transfer data back and
forth that was not availablethrough the built in solution,
inside the tool.
Now, some of you, most of youprobably don't have a clue what
is an HT TP call mean?
And to be fair, I don't either.
So the way I used to do this isto copy data back and forth from
whatever tool to Chachi PT orcloud or whatever.

(40:12):
And the way I do this now is Igo to Comet, which is an agentic
browser by perplexity, and I runwhatever tool inside of that.
And whenever I get to the pointit's like, oh, you can solve
this with an HTTP call.
I literally ask the browser, theagent within the Comet browser
to write the code for the httpcall for me.
And then it actually opens therelevant components inside the

(40:33):
application.
So inside relay or relevance ormake, or NA 10, whatever you're
working on, and it writes thecode for you.
And so even going the next stepor like, oh, I need to call my
technical team, is becoming moreand more available to
non-technical people.
It's still not perfect, but it'san option that you have at your
fingertips without talking toany other person because now
the, the browsers themselves aregetting to the point that they

(40:56):
can do that.
I'll stop now.
Give it back to you, Pietro.
let's show a few examples sopeople know how this looks like.

Pietro (41:02):
Super interesting and I definitely try what you, what
you said iza, the last bit ofthis, uh, of this podcast.
We were thinking to show yousome of the things which higher
built.
I'll just be very high level andexplain why I picked this
specific tool for this specific,uh, build AI micro up.
Call it however you want, right?
Yeah.
So we said before that Relay,for example, has great

(41:24):
integration with, uh, withReddi, right?
And if I go here and I am on myrelay interface, I'm sharing on
the screen, let me make itactually a bit bigger.
Uh.
I can see this is the agentwhich I have built.
And for example, this wentreally, really vital.
And I had at least a hundredcalls who came outta this flow.
'cause people thought, wow, thisis so cool, this is so simple.

(41:46):
And they didn't need me to setit up right.
I created a video, YouTube whereI explained how to set it up,
like the most advanced one, theycould set it up even without,
right?
And what this tool does is everyFriday monitor a specific set of
sub subreddit, which I define.
So it goes to, uh, subredditabout startup and, uh, product
marketing.
And it compiles the informationabout me, my own company, so he

(42:08):
knows who I am and what do Iwant to comment and what I look
for.
And then get all of the postsfrom those sub, which respect
some specific, instruction,which I gave, which is the top
post, which are in fit with whatI do.
Right?
So I was looking for posts,which are specifically
interesting for what I do,right?
I was people looking forautomation, for example, and
then the agent categorize ofthis comment, select the top 10

(42:31):
because I don't have time toanswer to everyone and drafts a
comment, which is.
Optimized to be picked up by,LLM, right?
So it's, a, it's an AO optimizedcomment and this works really,
really well.
It allows you, number one tofind organic, clients.
'cause Right is great forfinding organic clients, even
without being picked up by LM.

(42:51):
But the most interesting part,it's also if the post goes very
well and it gets a lot of views,it usually gets queried by large
language models.
And you will be recommendedbecause you appeared in that
comment, which is somethingsuper, super cool, right?
Why did I decide to build ithere?
The idea was I want to beatsomething already and I want to
make it related to answer engineoptimization because it's a
topic which is really hot and alot of people are very

(43:12):
interested in, right?
So I went to the platforms whichI, which I like, which I
mentioned before and I checked.
Who worked the best with Redditand where could I get to this
result first?
And in this case really has agreat connection with red, uh,
Reddit and the set of action,which I needed, were already
there, which is really, reallycool.
Another really cool stuff.
And this is another co-pilot,which is something which I
mentioned at the beginning ofthis call.

(43:33):
I can tell him here, if I don'tknow how to do something, I can
ask him or I can tell him toinsert a new step.
For example, I dunno, changestep 10 to adjust the prompt
for, I dunno, the most recent,uh, Reddits.
And he does it for you.
So something very, very cool.
Yeah.
Which I really like and it'sgetting better by the day,
right?
so this is one thing which I,for example, I beat him, which
was very, very popular.
Another thing which will be torelevance.

(43:55):
As we said before, relevance isreally, really good describing
content and finding information.
'cause it has already a presetof tools which are really good
at doing that, right?
So for when I wanna do somethingdeep, very deep and which has a
lot of study research, my go-totool would be, would be
relevance.
In this case, for example, Iwanted to create an N engine
that, uh, that tells me the fullbackground of one person.

(44:16):
Basically scrape all of thedigital footprint, which is
public, right?
You won't go into privatewebsite, but everything which is
public about a person, right?
So you see here I created, veryextensive research that use
LinkedIn data, Twitter datascrapes, uh, a lot of different
websites.
Uh, does research on G twoCrunch based, similar web built
with, right?
And I could get a really goodresults first here because of

(44:39):
the scraping capability of thistool, right?
And then I built, somethingwhich is called How to Sell to
Anyone, which is basically atool you would use before a
sales call.
To, to get all of the possiblebackground about the person and
then use a framework, which isdone by, by a Harvard, Harvard
professor, which based on theinformation helps you
understand.
How you should approach a salescall with this person, right?

(45:02):
It won't sell for you, but itwill put you in a better
position so that you can sell.
Yeah.
Uh, and then I don't haveactually an account on Lindy.
I only have on the free plan.
But if I were to build somethingwhich has a car wash, voice
assistant for a local business,I would definitely go to Lindy.
I 20 to, it's a bit com morecomplex'cause you have to
register a phone number or rentit.
And it's mostly for the UnitedStates, but if you're working

(45:23):
with the United States, you canget to really, really good
solutions with very, very goodvoice model by being something,
here on Lindy.
And I guess that's it.
We could go on and I could showyou, I have approximately a
hundred flows built in, uh, in,uh, in notion.
But, uh, this was meant to giveyou, first introduction to how I
would approach buildingsomething if I were an encoder.

(45:45):
Like I actually am

petro.
This was incredible.
As expected and even more, Ithink most people are dabbling
with it and are not at yourlevel.
And because you build so many ofthese across so many platforms,
your, your process is, isinvaluable.
If people want to work with you,follow you, learn from you, what

(46:08):
are the best ways to connect orfollow you?

Pietro (46:10):
So, uh, I, I am every day on LinkedIn, so please send
me a connection request theresaying You saw this, uh, I
receive a lot of connections, sosend, put a note in the
connection.
You saw this podcast, you wannachat And that would be the best,
probably the best way.
It's Pietro, Aldo, my Art.
I'm the first person who comeout luckily and I have just
started my journey on YouTubewhere I post mostly tutorial on

(46:31):
how to get to beat these things.
For example, I posted tutorialhow to Build the Red did Flow,
and I would really appreciate afollow there.
You can definitely engage withme there'cause I'm much more
active.
This is the channel.
It only has 80 followers, butafter this session, I hope you
will get at least to 90.
And I'm very, so in a year fromnow, if we'll talk, this channel
will be significantly larger.

(46:51):
And I'm really trying totransmit all of this knowledge
and make a also a more, moreexpanded sessions and micro,
micro, micro videos about whatwe talked about today.
Because either you push me tothink about it and now I have it
codified and now I can turn intomuch more content.

Fantastic.
so again, thanks to all thepeople who joined us live.
Uh, people are writing.
This was awesome and peoplewrote that you were awesome
several times, both on LinkedInand on the Zoom.
Thank you so much.
This was really, really great.
I think it will help a lot ofpeople, build a process in their
head and hopefully go andexperiment more than they're
doing right now.
I, and obviously I really,really appreciate you.
This was absolutely fantastic.
I'm glad that I twisted your armto put this into a, a well

(47:29):
structured process.
Uh,'cause I think it's gonnahelp a lot of people and I think
it's gonna help you and youalready said that.
So again, thank you.
Thank you.
Thank you.

Pietro (47:35):
Thank you so much, guy.
It was really a pleasure.
Thank you so much, Isaac, forhaving me, Joyce, for
organizing.

Thanks everyone.
Bye-bye.
Have a great rest of your day.

All Episodes

240 | AI Builder’s Playbook: Turn Your Idea Into a Working Solution in few hours with Pietro Montaldo

Episode Transcript

Popular Podcasts

Stuff You Should Know

Medal of Honor: Stories of Courage

Dateline NBC

.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;}240 | AI Builder’s Playbook: Turn Your Idea Into a Working Solution in few hours with Pietro Montaldo