All right, welcome everyone. My name is

Amal. I'm the CTO and co-founder of

Avani, uh, one of the three hosts today

with you. And my presentation is

entirely focused on the strategy from

literally

getting to know what is an AI agent to

the level that you see an actual use

case of this. I'm using this workflow

that I'm going to show you guys next on

a daily basis. I'm also an NATO

ambassador. Um that means I am taking

the Nathan brand to the world uh

literally every single week uh in a

workshop format. Uh yesterday I was in

Lisbon, the week before that I was in

Vienna. Next week I'll be in Madrid and

it keeps going on. By the end of this

year we will be in 21 cities.

So today my

mission is to give you guys a holistic

view of what can an AI agent do for you

especially in the light of an N8

workflow. So we'll start right away from

a live demo. I hope it's not going to

break. Uh but the demo that I'm going to

show you is like this and it's something

that I am using on a daily basis myself.

So literally when I do these travels,

our accounting department really gets

mad if I don't give them these receipts,

the scan copy of these receipts. So

having AI agents work for you uh and

having like a real use case is something

that I'm going to show you today. So

what I do is I go to um these cities

every week, a new city. I get a lot of

these ones. The moment I get them, I

take a snap of it. You'll see it in a

minute. And then it goes through this

workflow which we'll build it together

um at the kind of like 70% of the

progress of this uh today's talk. And

then you'll see that it's no more a

kind of uh future technology. It's

already there and people are using it or

nerds like me are using it. I'm a a

technologist. You'll know a little bit

more about me. But let's give the demo a

try. So I have my phone attached through

this cable to my screen. You can see it

here. It's um simple Google Drive app.

You can see that um right now I'm

running Google Drive and Google Drive

has this little scan feature and I have

a folder called invoices. So what I'm

going to do is go to invoices. you'll

see all of the invoices that I have been

using lately.

And this is a question that a lot of

people ask me uh when I go and do these

workshops that what is the front end for

an A10.

The simplest answer is that the front

end can be anything. And right now

you're seeing the front end a Google

Drive mobile application that takes a

snap of this uh this receipt and then it

turns out into something that's more

usable. on the right side, you'll see

that uh this entire receipt will be

scanned and then you'll see all the data

in here. So, let's give it a try. I'm

going to scan this receipt right now.

Here, there's this little um

scan feature. I'm going to scan this

and see how the scan goes.

Work.

This is not an NA10 feature. This is an

app that is built by um

by Google. They did a good job. So you

can do like cropping only if the screen

do not go out. I touched the cable.

There you go. So now I'm doing these

cropping things in here just to give you

guys an idea of what's going on.

Fingers crossed

demos can fail. This is a real demo. So

you can see that now it's uh uploading.

The upload is right now completed.

Um now if everything worked fine, I

should be able to run this workflow. And

what it's doing is using Mestral AI's

OCR. Mal AI is a company based out of

uh France. And it just worked. Nice.

My stress is no less. So, it put all the

information from this in here that I

need with a short description of what is

it about and also a link that my

accounting department can go and find

the invoice and get all the information

that they need.

And this is an actual use case that is

being used on a weekly basis. And and

this is something uh I used to do lit

literally by hand a little bit of OCR

here and there but now that we have

things going on in this way and also

kind of demoing it it's insane like this

technology was not possible to this

extent um let's say a year ago. Now the

main topic of my talk today is that I

show you what is an AI agent and what

can you do with an AI agent. This is one

example of this and there will be

thousands of more emerging in your daily

work life. uh at Avani we are exactly

challenging this that we are reimagining

work and sometimes uh if I take workflow

as workflow orchestration NA10 it is

actually a work

building tool so NA10 is a work building

it's basically reimagining how we work

so a few words about myself I speak five

languages human languages around 20

computer languages And I have been

building companies for the past 17

years. I built six tech companies. Uh

one of which is so dear to my heart. Uh

after consulting all those big uh

clients. This is the one that I really

like the most. This voice AI agent that

I built in 2023 and 2024.

By show fans, who knows this picture? 40

years old. Oh, 40 years old. So this is

a picture from 1980s uh TV series. and

David Aselhov talks to this car or for

those of you who are young, you might

know this picture. It's basically the

very early uh

voice AI agents showing up as as a TV

series or or as a movie and it was not

possible until just recently and I was

building this company in 2023 and 2024

and things that are possible today was

not possible two years ago, one and a

half years ago because of technological

shift.

What I learned after all of this startup

hype and liquidation uh is this amazing

technology that literally took the

entire orchestration of my company N10.

So this in this technology was in in a

very good time uh coming to me kind of

like going down from one side from this

startup but also going up from other

side as an educator. I am now educating

the world uh teaching them NAT through

my YouTube channel and as well through

my engagement as a co-founder and CTO of

um Avanoi partnering with NATO and

taking Nathan education to the world.

But my educational content is not kind

of like here's the ABC of it and go

figure it out yourself. All of the

content that I come up with is an actual

use case as you saw.

So, as a quick icebreaker, I I would

like to ask randomly some of you one of

those questions and then I'll go on.

So, is it okay if I ask you a question?

Okay. So, what's the coolest AI feature

that you have seen recently?

voice.

>> Tell me chat GPT voice AI feature.

>> I like voice

>> to be honest. Not only that uh I mean

like the response uh how they responds.

>> Yeah. Every time you ask them even if

you give them more pressure to give me

the straight answer. Don't go uh

sideways or something else. Um, this is

what I like really because it's it makes

it more human. This is what I uh for me

it's uh pretty um

>> Yeah.

>> Yeah. makes sense. Anyone else wants to

participate? Answer one of those three

questions.

>> Come on.

>> Yeah.

>> Um let's say if AI could automate one

task in your life, what would it be?

>> Handling my inbox and managing my

calendar.

>> Absolutely. Low hanging fruits. Easy to

do it. and and also we have some slides

coming up about EU AI act regulation

wise this is the low hanging fruit. So

if I put what is an AI uh agent in one

sentence I would choose this sentence

and the keyword I would like you guys to

focus on is this digital coworker.

I think that moment is not so far away

that our bosses will introduce Lisa

and you'll say, "Hey everyone, please

welcome Lisa in a daily standup call."

But the only difference is that Lisa is

not a human intelligent. She has a face.

She can talk to you. She can learn. She

can reason. But she's a digital

coworker.

The scary part of this is if you're not

a part of this whole movement, the scary

part is that she has a PhD plus level

IQ.

She doesn't sleep.

And her salary is a€,000 per month. Good

luck competing with her.

So, but on the other positive side, if

you're a part of it, which you are,

therefore you're in this room today,

we're building these digital co-workers

so that we can save time and really have

that 10 hours working week and spend the

rest of those 30 hours with our loved

ones or do whatever we like to do. And

in order to demonstrate this, uh, I'd

like you guys to a volunteer to come and

talk to my AI agent. It sounds like me.

It's my voice clone. One volunteer,

please come here. Anyone? It would sound

really strange if I talk to myself.

Anyone wants to come here and talk to to

my voice agent? Just a few sentences.

Please come here.

>> I'm just doing it because I can't bear

the site.

>> Thank you. We need some more rebels like

yourself.

>> Uh, should work.

if the voice.

>> Hello, dear. How are you doing?

>> I'm doing fine. How are you?

>> I'm doing great. Thanks for asking.

How's everything on your end?

>> No, it's it's fine. Had a long drive

today. Um, let me ask you a question

that I was asked recently. What's the

coolest AI feature you've seen recently?

>> Uh, a long drive sounds exhausting, but

hopefully peaceful. As for the coolest

AI feature, I'd say it's the increasing

ability of AI to generate highly

realistic virtual environments in real

time. With advancements in AIdriven

graphics, you can now create immersive

virtual worlds for games or simulations

that look almost indistinguishable from

real life. It's incredible how AI can

model and render environments with such

detail and precision, opening up a new

realm of possibilities for

entertainment, training, and beyond.

What do you find cool in the world of

AI?

>> Yeah, also that.

>> Also that. Thank you so much. Thanks a

lot. Thank you.

>> Thank you.

>> So that

Thank you for actually breaking that ice

which I couldn't not do it with the

icebreaker I had in coming here and

talking. So this digital version of

ourselves is inevitable. As Mark

Zuckerberg said, everybody will have an

AI agent and I think this is the early

sign of that. Almost everyone will have

one of these things. And instead of

sending your Calendarly link to to

people, you will send your AI agent to

talk to them and then you get the gist

of it. Or if the AI agent realizes that

it's the time to call you, it will call

you on your phone. Then you jump over

and take over the line. And and that

digital coworker is the key term that I

want you guys to remember. It's coming.

Some examples of this is lowhanging

fruits. Invoice reconciliation. There's

a lot of information flowing in an

invoice and invoice reconcilation is

very very easy to to build and AI agents

ace at this right now. I I don't know

why that is not doing it in a massive

scale although I know they are doing it.

They invited me in that headquarters. I

give a talk about a project I did about

nan and dative and AI agents. Um,

another one is cash flow prediction.

Like big businesses really can suffer

from cash flow, but this is an simple

one like all the data moving in

different spreadsheets and different

accounting systems. This can easily be

automated.

Compliance uh is another big chunk that

we can do because a lot of texts move

around and I don't know how many audited

reports have audit errors because there

was a human in the loop like a human

auditor did it but an AI agent capable

of doing uh tax audits will do almost

zero mistakes.

Um, I worked for or or I was a partner

to a $300 million scaleup company, an

e-commerce uh company. And there I

learned that you optimize little bit of

your uh your payments, especially for

your suppliers. There is like 200 pages

of each supplier contract and you find

one small little clause that you can pay

them 5 days in advance. You end up

saving thousands or tens of thousands of

euros every day. That's not possible

with human brain but an AI agent can do

that for you. My uh beloved feature of

AI is this voice AI. Imagine this uh

kind of a real state company

implementing a a call system, a phone

call system that all the tenants who

rent get a phone call on the first day

of the month. Your your rent is due. You

didn't pay. Can you please pay? and have

all the integration with the bank APIs

and so on to make sure that they paid or

not and call only those who didn't pay.

So

where do you see someone a little bit

audience engagement? Where do you see

that AI agents making the most impact?

There's one very visible

area.

>> Yes.

>> Pictures. This gentleman says pictures.

Anyone else? every everything where you

have a lot of very small details which

is hard for a human to write

>> small details text or whatever content

it is okay picture content text

>> making appointments in German uh doctors

>> making appointments in German do with

German doctors very very true what else

>> there's a very visible one nobody's

saying that right now in this crowd it

disrupted this industry it made a huge

impact

coding.

I have been coding for 23 years

non-stop. The last lines of code I wrote

was like literally this afternoon when I

was doing a podcast.

So coding was immersively disrupted.

Right now we have this concept of vibe

coding and and this is happening because

AI agents now can run in the so-called

YOLO mode in 3 four hours build the

entire work for you autonomously and

that's that's amazing. So there are a

lot of these platforms like the most

awesome one NA10 uh relevance AI you can

I like their LinkedIn scraper uh Lindy

an end toend integration vapia voice AI

agent 11 Labs uh an EU based polish

company that allows you to do voice

cloning voice technology stuff crew AI a

Python library that you can build AI

agents orchestrate multiple AI agents

together. Mastra for typescript

engineers, AI SDK agnostic model uh

selector and open AI and also Google are

getting into this software development

kit game bringing their own

technologies. But there's another corner

of AI agents that uses computer. This is

very interesting. Like if you have an

API that's fine you can automate almost

everything with NA10 but if you don't

have an API nowadays AI agents can also

use computer they can m move the mouse

take a screenshot understand the picture

on the screenshot and then click

somewhere type something execute your

wish for example you say can you please

buy me some um uh some bananas from I

don't know flink or or or whatever

grocery it goes on that check list and

then buys you this stuff and then does a

full checkout. Uh one from Antropic is

called computer use uh the first one and

then operator from open AI and from

China Manus AI. These are projects that

are doing this. And I'm this dude uh so

deep into Linux and

kernels and Docker containers and so on

and virtualization and I say that I want

to poke things around and think and see

how do they do these things. It's not so

complicated to put a virtual machine.

And then I end up building this one. Uh

this is datative accounting software on

the right side running inside a web top

docker container and a nexjs application

here literally taking my commands here I

upload multiple invoices it goes there

and then uploads it one by one and

runs some OCR fills all the fields

basically um creates like a a

full-fledged um reconciliation of an

invoice system. Uh this is a technical

diagram but just to mention the the

project's called uh it's an open source

project I did. It's called blind clicks.

Uh it basically blindly clicks until it

can't find the click and then the moment

it doesn't find the click it basically

wakes up the dragon which is an LLM with

a project from Microsoft called

omniparsser that takes a screenshot

understands in what xy coordinate what

icon is there. Can I click it? Can I

mouse over it? This is a drop down and

so on and so forth. and then the LLM

understands it and then it create

creates a new sequence and that sequence

is executed inside this docker container

that's running the entire UI very simple

fun project I did imagine if big

companies like that if they start

building something like this it changes

the way uh accounting is done and when I

did this this post went viral that dudes

found it and then they invited me they

were so nice and that of HQ and I give

this presentation uh to that of uh

200ish engineers in the room. Very

pleasing experience.

Uh in EU, we need to kind of like it's a

little bit controversial. What I'm going

to say is kind of like think about these

guys like North American guys like Satya

and end of uh 24 said like SAS is dead.

Microsoft is a SAS company and Satia is

saying like SAS is dead because they're

seeing this this new wave coming in.

Mark Zuckerberg says there will be

billions of AI agents and that's we we

need to kind of get to that ambitious

level and true almost every single one

of us will have an AI agent and and

that's the most convenient way to kind

of augment ourselves put it out there in

the world scroll feed give us a summary

when we are taking a shower or compose a

LinkedIn uh post when we we're driving

these things that I am currently doing

I'm a nerd I'm super computer geek I

want to build things but this is going

to come to the masses and it's

inevitable. It's coming. Our Nvidia CEO

gives this beautiful

diagram that we are right now here

before hitting that physical world. We

are going to build these agentic AI and

this arena is just starting and

Salesforce CEO says this is bigger than

the internet. It is because it's a

tectonic shift. It's it's the way

muscle was uh kind of like productized

or commoditized with machineries. Now

intelligence is commoditized. So a Lisa

that takes a salary of a€,000 per month

a PhD plus IQ never sleeps is going to

take our jobs. And that's how big it is.

It's bigger than the internet.

So a quick question from you guys.

Do you believe 2026 will be this the the

breakthrough or the changing year for AI

agents especially especially in Germany

or in Europe considering all the

regulations GDPR EU AI act and

everything else. We have some EO act

slides coming in anyone

>> please.

>> I believe it might well be the case. Um,

I have hit upon madness earlier this

year and I was having this like wow

moment and um I thought

>> So you're optimistic it's going to it's

going to be there hopefully. Yeah.

>> Yep. It's going to be there.

>> Yeah. Because ROI is explosively

positive.

>> It's the same as an Uber killed the taxi

company.

>> Exactly. Exactly. It's it's coming.

Anyone else wants to please?

>> Technically technically definitely but I

see problems with itself ad

yes advisor and I see many

management

>> mentality mindset

>> fear and I think the biggest problem

especially in Germany is that people are

not ready

>> yes

>> I've been developing with GPS and agents

since years

>> and I know what is possible but the

problem

adoption, change management. Exactly.

>> Is the biggest issue,

>> right?

>> And if you try to transport it, you

always see a war.

>> Exactly. This is this

>> we can and I think there will be some

companies that will adapt the technology

and use it and companies that don't will

get

>> right. Absolutely. Absolut I I am 100%

with you. If we do not shift our

mindset, therefore the previous slide

was about these North American leaders.

Now, if we do not shift our mindset to

kind of like let's take the risk, there

is a 7.5 million penalty that we could

end up paying, but the ROI is probably

20 million. So, let's take the risk.

Let's let's go and implement this

despite the fact that this might not be

100% compliant. And who knows that these

big players like Mr. AI and Alfala and

those dudes are also not compliant. And

so I am the CEO of this 500 employee

company. I know that I'm going to either

go bankrupt in two years or I'm going to

take this risk and and go build

something at Gentech and automate

something that will give me 20 million

euros every year and I'd risk 7.5

million of that. So I think that

mentality needs to be developed more.

I I want to talk about EUA act. I'm not

a lawyer or a liar or neither of them. I

want to talk about it because it's

important. I think everyone needs to

have at least one or two slides about

this this act. The first question I ask

myself is that why is it there? I have

two kids, one and a half years old and

six and a half years old. They're going

to grow up here. Uh I want a safe place

for them. I wouldn't go in a in a

country not having

traffic lights.

And if I ask myself, what does EU act

for? It's just like traffic lights. It's

a regulation that prevents harm to human

beings and it protects protects our

framework called my EU50

which is a very simple framework that um

that is built I think when EU was built

that protects one of the 50 elements

that are in these six pillars human

dignity freedom democracy equality

rule of law and human rights and any of

those are threatened by any AI

application then you need to go through

one of these four layers that are

unacceptable in EU that's kind of like

literally no go or it has high risk

therefore you need to go through a very

high uh degree of regulations or it has

limited risk therefore the number of

regulations you need to fulfill can be

limited or it's also called transparency

risk and there is minimal risk this low

hanging fruit like literally things that

you can do with email filters this is

minimal risk things that you can do with

deep fake this is limited risk for

example an AI recommendation system

doing the wrong recommendations for

example for a child

Iris can be anything to do with

healthcare or credit scoring system if

there is based in data and certain

people do not get a good credit scoring

for their mortgage loan and so

And prohibited is anything social

scoring or real time biometric

surveillance or predictive policing. For

example,

if John Doe is someone who has been

committing the same traffic uh violation

throughout multiple years and we can

write a model that can predict with 96%

um accuracy that John Doe will do

another like go through another uh red

light next Tuesday at around 3 p.m.

which AI can do if it gets enough data.

And why don't we send uh a pace like a

like a ticket a week in advance to John

there and say like you can do it or

please pay this in advance. This is

productive policing. It's it's something

that's prohibited though it's really

black mirror-ish idea but uh it's

prohibited in Germany. So now it's the

time to build. I didn't want it just to

be a talk talk. Although I'm not very

good in talk. I am this engineer who

likes dark basement and dark mode to

code. So I'm going to build something in

front of you and I will show you things

that normal engineering time will take

forever. Just by show of hands who is an

engineer in this software engineer in

this room. Okay. So therefore I will

keep the numbers the the the vocabulary

to non-technical

uh caliber. Uh I want to show you guys

that this technology NA10 has

revolutionized this this world of

software development. I've been building

software as a full stack in total 23

years. Before that I was a desktop

engineer. For those of you who knows

what Notepad++ is or what Visual Basics

6 means, I am someone from that time and

I've been continuously building. Now,

what I'm going to show you is probably I

can't explain how hard it is to build

this thing that you're going to see in

the old school way of building things.

But when I keep building things, I will

try to explain it at least with words

that how many hours would it take for a

normal software engineer to build such a

thing. All right. So, I'm going to go to

NA10 and explain a few things uh as I

go. So anything is this um very simple

orchestration layer where you can build

things and and those things are called

workflows and any workflow can have

certain nodes as a tiny node can be

doing certain specific uh thing for

example here I'm going to add a

so-called chat node and this chat node

does one thing it allows you to do chat

when you do chat it triggers the

workflow So you can see here there's a

trigger. If I say hi, it just puts a

check mark there. That means like this

node was executed. It's done. It's

executed. Now I want to add an AI agent

to this. How do I do this? Very simple.

I put an AI agent node.

And there we go. An AI agent is here.

Now let's take a step back and do a very

simple definition of what is an AI

agent. Anyone wants to participate

before I go on and describe it?

What's an AI agent if you want to

describe it?

>> Anyone please?

>> Agent has a goal for example.

>> A goal.

>> Yeah.

>> Goal actions. Uh those are the terms I'm

I'm hearing. Goal actions

>> um autonom

>> autonomous aut um agency.

That's the term. Okay. Anyone else? What

is an AI agent? Give me some terms and

then I'll put the definition there. Yes.

>> Maybe humanike.

>> Humanike. I like it. Somebody else uh

used the term non-human intelligence. I

really like that term as well. Somebody

else defined it as uh uh alien

intelligence or something like that.

Yeah. Anyone else? What's an AI agent?

>> You mean just like something

large?

>> An LLM could be a part of this. An AI

agent.

Nobody talked about tools or tasks. So

an AI agent can also do have access to

our tools or execute tasks.

The simplest definition of an AI agent

that I can come up with is I s I show it

to you guys earlier but a little bit

more technical could be something like

this. An AI agent is a software that

have access to your tools, understands

your natural language, and it can

perform tasks on your behalf just like

your digital coworker. I really want to

coin this digital coworker thing. So if

you and see this this definition

contains certain keywords that I really

want you guys to kind of like

remember it when you go home especially

this digital coworker

this digital coworker is is is

inevitable. It's it's coming. This the

work is going to be disrupted first

before even the robots come and then

disrupt the muscle part. The

intelligence part will be disrupted

first and the most important part is

performing tasks. So this performing

tasks these digital beings or these

digital co-workers will perform tasks on

what? On the actual tools that you give

access to them on your tools. So like

these little tools that you give access

to them be it your Gmail be it your Jera

or uh Google Sheets or Microsoft Teams

or whatever it's going to disrupt that

and the beauty of it is that it is going

to go with this natural language.

One of the 90 languages that these LLMs

today speak

this can change. It's it's amazing like

it it will increase the number of

languages.

All right. So now I have this AI agent

in here. It needs an LLM. I'm going to

give it an open AI LLM access. That's

nice. Now if I type here something,

you'll see what happens. So I'm going to

type

hi. This time it's going to respond

something to me. You see here it says

how can I assist you today. This is

basically a chat GPT integration. I just

click click click and then everything is

done. The term that I really liked about

anything is Lego.

An is like a Lego is like smaller

pieces. You put them together and then

it works.

So now there is a concept in um

in programming you might have heard is

called cache caching. Look how easy it

is to implement a cache layer here. Why

do I need this cache layer? I'll explain

it here. So I'm going to use this

workflow to do this. Hi, my name is

Amal.

See what happens. This workflow

says

this thing. Hi Amal, nice to meet you.

How can I assist you today? The next

question I'm going to ask is what is my

name? If I ask this, look what happens.

Previously it says I amal. Now it says I

don't have access to your name. What is

the problem here? There is a big

technological problem here that every

single time that we send this traffic

this this request to Chad GPT we

basically send plain first fresh

information now somehow we need to cache

it somewhere we need to save it

somewhere now I'm running on NA10 cloud

and N10 cloud is super scalable you can

imagine what $200 million can buy you

the recent funding of NA10.

Now if I click on simple memory, click

on simple memory, click on back, those

were just three clicks I did. It

literally added a cache layer. Now for

those of you who are a programmer and

you know what cache invalidation means,

it's one of the complicated things in

computer science.

And this just implemented a cache layer

for me. Now if I go back and do the

following,

I say, "Hi, my name is Amal." It goes

and stores it here in this tiny database

on the servers of NAN. And if I say,

"What is my name?" Now, it goes and

pulls it back from that. And then it now

have access to my name. And this is just

a tiny example of how cool this

technology is and how cool the 1.1

million lines of code that was stacked

upon each other through human brain. At

the time there was no vibe coding. There

was no AI agent code generation. It has

happened throughout six years

almost 15,000 so-called pull requests. A

pull request is a bunch of code that is

get together in very close collaboration

of multiple developers

and then it ends up being building such

products. I will demonstrate one small

thing here just to to make the point on

how cool this product is. And I I'm in

Germany. I can't really brag about this

but when I go to other countries I brag

about this German perfection or or made

in Germany thing. It's uh it's like

this. For example, here I'm adding a

so-called trigger. And when you add a

trigger, look what happens. If I want to

add a Google Drive trigger

for the workflow that I just

demonstrated earlier, I need to have a

Google Drive trigger like this, right?

If I add a Google Drive trigger,

do you see a tiny little feature here

that must have gone through a feature

request, a product owner worked on it, a

programmer implemented it is this thing.

This little dude is collapsed by default

because it's a trigger time. You don't

you can't add an action here. So it's a

very tiny little uh example of this. But

there are others like this. So for

example, if I add this particular

workflow that I have created here that

enabled me to do this thing, you'll see

what happens.

I add this invoices folder

and I look for files to be created. And

when I execute this, it goes and pulls

out all the metadata of that newly

updated invoice. Not the most

complicated thing. The the thing that

really sparked this idea of German

perfection I want to keep talking about

is coming in two steps. If I now want to

download this file, I download it.

That's fine. Um, and I want to download

it from this particular ID that's

sitting somewhere

in here. I will need to find it. Here

you go. It's just coming in a few

seconds. That perfection, that Lego

blocks kind of like clicking together.

And and that could have not been

possible without going through all that

evolution of that product across six

years. So here I'm adding a mistral AI

integration because now I have this this

particular file with me on this canvas

which is literally a file that can be

downloaded. It's a PDF file with 556 KBs

and it's available. You can see it just

downloaded and if I open it it's exactly

that PDF file that I scanned earlier.

Now if I integrate this with Mr. AI. Mr.

AI is a French company that's building

LLMs and also their document extractor

is very nice. Now, see, I opened it. I

don't do anything. It just clicks. Why?

Because the

programmer sitting in Berlin said,

"We're going to call our our field

data." And the programmer sitting in

Paris said, "We're going to call it

data." So, that German perfection like

literally everything clips together. I

don't change anything. And then I

execute this. It just goes on and and

then it basically uh when you put the

credentials it basically executes and

then it just works. I was giving this

advanced example advanced course earlier

um and and showing them if something

fails because you provided an incorrect

account how can you trace it? Therefore,

that error that you saw is from there.

Now, it literally three steps and I'm

send downloading some data from Google

Drive, uploading it somewhere else and

Mr. AI is giving me back all this

information. This level of

product development with without having

such a tool would take me forever. Leave

alone the fact that I could literally

demo it here. This this would have been

almost impossible almost.

I want to make sure that I have enough

time left for the Q&As's.

But if I want to conclude this

presentation, I would like to conclude

it with this that from

this basics agents basics

uh to summarizing emails and everything

uh where we are going next is is

amazing. It's going to be this agentic

workflows, enterprise focused wrap-ups.

And the bottom line is that AI agents

are no longer experimental. It's it's

there to stay and and we are going to

see in 2026

major shift towards adopting AI and and

NAN is going to be this technology that

people adopt and then automate things

with AI

and this the the the areas where AI

agents fail are very important to also

know. the ambiguity in your inputs. This

geigo garbage in garbage out, the

missing data or context, the

elucination. Nathan recently announced

something called guard rails to kind of

prevent AI agents to elucinate and to

prevent doing things that they're not

supposed to do, evaluations a couple of

months ago. These little tiny builds and

whistles around agentic workflows are

going to change the way it's going to

do. But the lesson to learn here is that

success of an AI agent is not just an AI

automation. You put it in the while it's

going to work. The human in the loop is

very important and also the clear

constraints and instructions.

Thank you so much. Any questions?