Another episode of Cloud Unplugged.

We've got some good news stories.

Obviously, always things going on.

got the big super mega power

outage in portugal spain

and I think france as well

um was it cyber attack was

it not it wasn't apparently

it was a meteorological

thing maybe kind of but

maybe was cyber um no I'm

really admitting we've got

meta's standalone um ip app

launch so they've now done

their meta ai um ai system

which we'll talk about and

then we've got alibaba

And the new AI model, Qen three,

which we'll also talk about.

But I say we, probably mostly you.

And then a little bit

tangentially on just normal

things around the Amazon tariff.

Should we be showing the

extra cost to everybody?

But maybe not,

because Trump slapped their

wrists and said, absolutely don't.

But yeah, let's get into it, I guess,

Lewis.

The power outage.

What do you reckon?

Cyber or no cyber?

Well, it's interesting.

Nothing has been ruled out at this time,

I think is the most PC or

most accurate way of putting it.

But

It's just fascinating what

has been said about,

and it was interesting you

brought up a meteorological effect.

The equivalent of the Met

Office in Spain have ruled

that out as a possible culprit.

And there's a lot of social

media talking about an

anomaly causing the event.

And there's a lot of

engineering discourse

talking about solar

inverters and how they can

react instantaneously to grid events.

So at this time... Solar inverters?

What are solar inverters?

So inverters are something

that takes DC power

and make it AC by inverting

the signal to make it

suitable for transmission on the grid.

Now,

the thing with many other renewables

or otherwise,

there's a lead time for

getting new generation onto the grid.

um with wind you've got

massive turbine blades that

need to spin up and start

running the generator and

they need to turn into the

wind or be blown into the

wind but with solar it's

solid state it can all be

instantaneous and normally

they take the signal of

what they can add to the

grid or take away from the

grid by reading the grid

itself directly so there's

has taught that a highly

interconnected grid with

very similar dynamics for

how power is controlled

across a massive industry will lead to a

potential cause but as I

said that's just another

percentile and it seems

like surely this would be a

particular make or few

makes of solar um inverter

and connectivity grid

infrastructure etc etc and

there'll be many fail-safes

between that component and

a wider grid so yeah we're

all left guessing

no one knows really I guess

now I think I think what I

saw was that um they the

spanish government

basically said that the

power companies have a

deadline I believe today

this afternoon it's meant

to be this afternoon and I

was browsing and looking at

social media just before I

came on and nothing yeah no

nothing in the news but

they probably have provided

to the government I imagine

and they're probably governments

looking at it and then

there'll be pr around it

because I guess it depends

on what actually comes back

and whether they're allowed

to even tell you the truth

or not or whether we all

just forget and move on and

no one ever knows um indeed

I I tried to dig into the

cyber side of things and a

possible attack um

just because it's not unprecedented.

It's happened before.

You know,

Ukraine obviously is a high target,

but India and US have also

reported incidents in the past.

And we have seen lots of critical,

or at least large parts of

national infrastructure affected before,

even in the UK of the NHS and

So it just seems to me my

food judgment would be... Cyber.

Well,

a coordinated attack seems like

almost as audacious to be

able to affect so much of

such large portions of countries.

So then it makes you think, well,

there must be some

commonalities regardless.

So therefore,

unless it's a massive

coordinated cyber attack...

And that would be a big state act, maybe.

Yeah, I don't know.

I'm left...

slightly frazzled.

But I guess the interesting

part of the story was

really the knock-on effects

to everyday life,

and even petrol pumps being

affected because they're

electric by power outages.

So phones couldn't be charged.

People couldn't pay for things.

I mean,

I haven't used a credit card or had

one with me for a long time,

and certainly haven't used

cash for a long time.

But I'd be the first one stuck in Spain.

Although it was lovely to

see lots of people partying

in the streets and

generally just thinking, meh.

Back to real normal.

Back to human to human.

Human to human.

No more tech.

No more...

you know,

corner cutting on like tapping

and sticking at your phone

and messaging people and

being all digital communication.

Yeah.

So it was like,

do look up because there's nothing on.

Yeah.

I kind of like that.

I suppose it's probably

quite a refreshing.

Maybe we should do it every so often.

Maybe we should have a blackout.

Blackout days.

Blackout days.

Yeah.

Yeah.

I guess TBD, we'll have to find that out,

I guess.

Yeah,

see what the rest of the world finds out.

But we weighed in.

It wasn't really cloud directly related,

but it felt news and cyber

is a big part of IT security.

And critical national

infrastructure is all

electronic based these days.

So it kind of all leads to that.

But kind of a flip side of that,

if I segue into the next story, maybe.

Well, actually,

I need to give feedback on

because I said I would also

give feedback on agent

space because I couldn't get it working.

And just as feedback,

please improve your error

messages and error handling

Google because it's so poor.

just says source effects

something something

something on the on the uh

llm stuff and then on the

vertex ai stuff the portal

for vertex ai for like the

basically like a the

catalog the app catalog

stuff so you can actually

then create um apps on

require different

permissions and then even

though I had a license as

the actual issue was it

said I had a license

assigned um but I kind of didn't

So what I did was I

unassigned and then

reassigned the license and it worked.

So who knows what happened there?

But anyway,

it just sounds like technology.

Yeah.

So I did manage to get it hooked up.

So I managed to create one of the apps.

um which is the vertex ai

application aspect of it

there's the notebook at the

lem aspect to it too um

which is like more you just

provide documents and

provide things directly but

when you use one of the

apps I've got some default

I think some of them still

in preview um like search

general search media search

got like a healthcare thing

as well um for certain

types of data types

Are these in Vertex Studio

different agents or tools for the agent?

They are agents.

So they are agents.

So the search agent, I think,

I only use the data source.

So you provide the data

source and that could be

lots of different things.

So it could be BigQuery or other stuff.

Mine was just like the drive.

So it wasn't a cloud resource.

It was Google Drive.

What was your agent called?

What was your project of one-liner?

Sell it to me.

Called it, I can't remember.

I was just fiddling around.

So whatever,

just some nonsensical name for me.

I was just testing it out.

The aim of it was for me to

ingest a lot of our Drive documentation

not just drive documentation

but obviously we've got

like customer proposals and

statements of works and

things like that so

obviously I needed to make

sure it was protected so I

needed it regional I didn't

want it to go anywhere I

wanted to make sure the

data is encrypted because

obviously it's got customer

data in it and things like

that not really customer

data but work that we've

done for the customer that

might be classified a

little bit sensitive

nothing really that

sensitive in it but um

But yeah, so then encrypted all of that.

And then I wanted to just

ask questions to see how easy it was.

Like,

can I ask questions on data in the drive?

Like, you know,

when did a project with

customer X end or when we

did this work with this customer,

what actually was the outcome, you know?

And so we'll end up reading

all the information.

So long as you're keeping

track of all of this data,

obviously you need the data.

That's obviously pretty,

it's just gonna make it up.

And also the data does need to be accurate,

as in whatever you've got

in there does have to be good data.

But if you get it right,

it's actually quite good.

So I got it to write a case study for me,

but it literally was more

verbatim of what the

statement of work was.

I haven't added it to our

existing case study drives.

We've got existing case

studies in there for customers.

So I thought that'd be quite

an interesting one.

And then you can also

integrate it into Confluence.

So I was like, well,

actually I could create a

Confluence agent and maybe

there's like other

information in our

Confluence to start to ingest.

And internally it could

become quite useful if you don't,

say you're a new starter in the company,

you don't know that much, you know,

about AppBear and the history.

You've not been privy

to all the customer journey,

then you might want to just

ask questions because you

just don't know.

So you might just go into it

and add questions.

So this allows you to hand

bake an agent with fixed

sources and provide a chat

interface for... It

provides a chat interface.

It is the agent.

It's the search agent that

provides an interface.

You have the option to toggle web UI.

And it will basically create a web UI,

basically front end thing

around the agent for you,

which is just essentially

just a search bar.

And then the results coming

up of the search underneath

that your whatever question you're asking,

whatever search is a little

bit rudimentary.

That type of thing is that

agent was very search driven.

That was the type of agent.

There are other agents like

media agents around like

graphics and images and

other things and videos.

Was there a...

Was there a user interface

to link agents so you can

have agent to agent conversations?

No, but they do have that.

I was looking.

Like a flowchart type of thing.

You can, yeah.

And you can actually add other things.

So some of the...

some of the functionality

within it was like

calendars and gmail and

some of those are in

preview still um and so you

can start to attach

additional things so

actually I think that's I

think that was actually on

the data sources side on

the integration side

there's like integration

and then you can do the

integration so technically

they probably are agents to

be a calendar agent there's

obviously a gmail agent

there's obviously on this

search so you can actually

It will be agent to agent, I imagine,

but I didn't get that far.

I don't know if I trust it

yet to have access to my

email and my calendar and

my drive and everything.

I mean,

they do have access because it is

their services,

but yet another service

indexing into a model.

I guess it's a very interesting journey,

and it's going to be a few...

you know,

bits that we can come back to

throughout the series to see like, ah,

okay, it's not just new user onboarding,

but it's a sales flow when

we ever need to create a

report or case study or

whatever it may be.

And in tech,

you know,

in software development is going

to be a whole bunch.

We're talking about documentation,

like how do we ground it?

How do we use software one

point to have tools that do

concrete things and provide

very exact validation of

things like schemas and

parameters and blah, blah, blah.

and how do we use software

two point O in agent space

to then come up with words

to describe this concretely

or reliably or whatever.

It's fascinating.

Then we'll all use more.

Yeah.

So it was cool.

I kind of liked it.

It solved kind of ish the

problem that I was going to look.

So obviously it's a bit of

just playing around.

So nothing too interesting.

But yes, so agents, talking about AI,

I guess,

because I did need to give an

update on that, and I will keep,

like you're saying,

keep giving an update.

The meta AI, over to you,

because you were looking

into this a little bit more than I have.

I haven't really seen this, but...

Yeah.

Well, interesting.

I was going to do a segue

from the other story

because I started using

Meta AI in WhatsApp and on

WhatsApp web to start

fielding some of the

research and comparing it

against Gemini and other bits and pieces.

So I asked it about the

cause of blackouts and stuff.

And it was okay.

It was okay.

It...

kind of made up um got lost

down a particular trailer

fort as ais do but it's a

llama for a mixture of

experts multimodal model

but fundamentally open

source the version doesn't

know which particular model

is used in whatsapp

He goes, I'm just Lama IV.

Which one?

Are you the Scout?

Are you the Maverick?

Which one are you?

And he couldn't quite answer that.

Sorry, just a quick one.

You're saying WhatsApp.

I know that obviously Meta

bought WhatsApp.

Do you mean that you were

using this in WhatsApp

itself or in the Facebook app?

No, in WhatsApp.

So it's available in Facebook.

I don't have a Facebook

account or an Instagram account,

but I do have WhatsApp.

And WhatsApp is even more

pervasive maybe than Facebook.

um or I don't know about

instagram maybe it's up

there um apparently the

tool meta ai is available

on on on many of the

platform many bits of parts

of their platforms and

tools but it's in whatsapp

And you can have a little

chat and you could forward.

You have to copy and paste

things to it and you could

forward chat you've had

with it to people.

But the context is lost.

It doesn't.

What I really wanted was a simple button,

a way of saying fact check

this nonsense that people

are sharing with me.

That's I think the killer.

Yeah.

People being,

I think I WhatsAppped you today.

I was, I literally was WhatsAppping you.

So I feel you're referring to me.

I don't know.

That's your own sensibilities.

Right.

Okay.

Okay.

Okay.

It does feel like a killer use case for,

you know,

if you could get past the snide

sort of worry.

John,

have you seen someone about your

paranoid delusions?

Or is it AI fed?

Every message I send to you

gets fact-checked.

I just get an auto-response.

Yes.

You don't even read it.

You don't need to anymore.

I basically have an agent

spam John's specific filter.

I mean, most people it comes direct,

but with you, there's unknown.

They just,

they go through a filter and get

screened.

There's known,

and they generally come from in,

there's known but John.

It's like a, you know, like, you know,

I guess some repellent.

It doesn't,

it doesn't actually kill you or

block you.

So you don't block,

but what you do is you repel

Through the art of the agent

chatting a load of shit

back to the other person to

the point that they just

don't bother messaging you anymore.

It's kind of genius.

Yeah, I like it.

I like where you're going with that.

It's clever.

Yeah, I mean,

it's integration and possibility to,

you know, thwart misinformation.

I think that will become...

Become, you know, make it useful to me.

Maybe give access.

I know lots of people that I

know that are not in tech.

Lots of my wife's friends

are in personal training or

healthcare or whatever.

The first use of an LLM

wasn't ChatGPT and it

wasn't any of the others.

It was Meta.

um into whatsapp because

it's right there it's in

the thing they use every

day so yeah so that that is

interesting but it's also

interesting how good the

model is and llama four and

you would imagine they

would use a fairly sizable model

There's a whole different

variety of the flavours,

incantations of Llamaphore.

And I couldn't find out

which particular model the

one in WhatsApp was.

But I'd imagine they'd want

to fairly put a good foot

forward and sort of

showcase something fairly good.

But it couldn't count

characters in a sentence

back from the end.

So you say,

what's the thirteenth letter in

this sentence with some

typos and stuff in there?

It couldn't do that.

Gemini can, Gott can.

There are the latest sort of

large models can do that sort of stuff.

But this one couldn't.

And then it went for a crazy

arc of reasoning,

sort of just talking absolute nonsense.

And this is great to sort of

see the thinking behind

what an LLM's doing and saying, right,

E minus O minus three minus cent

Is that five?

I don't know.

And then it sort of goes

back and then it comes out

with a number or a letter

that says it's S. Oh, no, no, no,

it's five in its chain of form.

And it says one.

And then it said, did you count spaces?

Because it was just one off

apart from a space.

And it said, yes, I did count the space.

And it said, well,

let me check that again.

And then it comes back and it goes,

was it Q?

It's like there wasn't a Q

in the sentence.

What are you talking about?

And then it goes, well,

how did you come to your answer?

And I did a bit of bash to use rev,

pipe a sentence for our rev

to get a reverse,

and then pipe it for head

and tail to get the characters.

And then pasted the result

back to it and said, I did this.

And it goes, ah,

I should have trusted the algorithm,

was its answer.

So I'm just like,

so now I'm the algorithm.

Is it because I'm wearing a

Skynet t-shirt?

I don't know.

Maybe.

That is so funny.

Though I do relate because

that is sometimes how I

feel when I'm trying to talk to you.

I mean, just in general, you know,

sometimes, you know,

my brain does act like that.

It does go a bit, what one might say,

senile.

I don't know.

There is moments of that that occurs.

But anyway, yeah, I did.

I mean,

totally a little bit related to this.

I can't remember what I was listening to.

Some of them might be a

Proficy Markets podcast.

Going back a while.

They were saying that the

reason Meta wanted to get

into the phone business is

because obviously the

phones will own the AI, right?

So to go into another app to then do AI,

like WhatsApp or Facebook,

is a little bit weird

because you'll have AI on

your phone natively.

So obviously,

that's why they were trying

to like the devices

themselves are the prime in

terms of like market share of AI.

So they've got their work

cut out really meta, you know,

to kind of leverage into

the user behavior like they'll want.

Obviously, they want it.

That's all about, you know,

the TikToks and the

Instagrams and the Facebooks are all.

all around,

the usage and the nature of the usage.

But it'll be a hard thing

because once you've adopted

a certain AI chat

functionality that you're happy with,

and once they start

integrating into the things

that you're saying,

the context and MCP stuff, then yeah,

the probability of you

using another AI to do kind

of a similar-ish thing,

probably unlikely.

So it'll be an interesting

thing over time.

It is.

I think if you provide a

compelling enough set of context, I mean,

the meta AI angle in all

the blurb is it will form a

history of its chats with you.

And remember, you know,

to be able to use that in

the context so it doesn't

need to go from scratch every time.

And if you're using it for

I don't know,

looking up holidays or whatever,

then it will remember where

you've been and what you've done.

Whether that means it's got

privy to all your chats.

Unless you're engaging with

it all the time.

That annoys me, though.

I hate it when...

Netflix or Spotify start to

pretend to homogenise your

behaviour into something

that they can predict.

You're like, well, actually,

I want it something different.

I don't want the same thing.

Play something different to me.

John, you are one of those people.

That's all you do, isn't it?

Yeah, no,

I just want... That's the thing

that actually generally frustrates me,

because I'm like,

I like different things.

I don't know what I want to listen to,

really.

Surprise me, and you're like...

I have a fourteen year old son, Charlie,

and the doom scrolling,

he's very good at not doing that.

But there was a period,

maybe a year or two ago,

where doom scrolling became

a phrase in our family.

And I think we've all got to

catch ourselves at it.

And it is toxic.

And you want to not have

that as an example.

And you want to make sure

that when you engage,

you are the one that's choosing

It might be a bit more work.

Give me the bloody unlearn button.

I'd be like, no, actually, unlearn me,

thanks.

And the idea of YouTube Shorts.

For me, I put things in Watch It Later.

And I find them on a

different feed on X or something.

So I am being fed.

It's just around the house.

We're all being fed.

It's just where it arrives.

We believe we've got agency, but not so.

And what do you think about

Quen three then?

Alibaba, new AI model.

Is this something you've done?

I didn't even know.

I mean,

it obviously makes sense that

they're going to come out with a,

um a new model but I didn't

get that far uh digging

into it um the headline

headlines um alibaba um

creating their next model

and obviously they have a

cloud and large organization

china yeah and their latest

model I mean it just stands

to reason anyone who's got

access to lots of cloud

compute will provide their

own model or enter this

space and with llama

three, four now available.

There's foundation models

that have been pre-trained

as your starting point and

whether you use that to

refine and produce your own

foundation model from

scratch or shortcut it and

go straight to producing a

model that's got some

grounding or some prior knowledge,

some prior foundation.

I don't know,

I guess it's just a highlight

of the speed to which

anyone with enough compute

can enter this market and

how there's no ring fence

from tariffs or other things.

In fact,

there's probably more of a lever

to accelerate development.

if you know things are

constrained so I don't know

I I I just thought it was

it's a good um a good bit

of uh bookmarking about you

know book ending um yeah no

that's good I'll have to

have a look at the the old

Is it Quenthrie?

You do pronounce it Quenthrie, do you?

Quenthrie, indeed.

Yeah,

double-check it in case I was

pronouncing it wrong.

Cool.

I guess the other thing that

was quite interesting is

the use of models and where

they're hosted.

is kind of fascinating.

If you're in meta land,

then you've got access to

meta models built into certain tools.

If you're using some tools,

you can put API keys and

get all the tools.

And then you use SAS-type services.

I mean, I use CURSO a lot.

It's come up before.

And it's got access to a

certain amount of models

built in so it's got access

to gemini and it's got

access to our open iei um

oh three and four models um

gpt model um and anthropic

most importantly because it

seems to be the one that

work but you can put in

your own keys from the

cloud and use the cloud

hosted versions of all the um

all the open ai models for

instance because um you can

use azure and microsoft to

go direct and get per cost

a cost based price rather

than the reseller sas price

so it's all changing super

quickly and yeah they're

moving target isn't it at

the moment it was there was

I was at the amazon summit

today in London and there

was somebody who was telling me,

this is secondhand

information because I

didn't speak to them myself,

that somebody was talking about another,

I don't know if they were a

booth or maybe they were

just at the event where

is a bit of another model

that's a competitor to Anthropic.

But the way they've trained

the model was actually

through not just learning the code,

but actually learning the compilation,

the errors of the code as

it's been written and

actually training it

through usage as opposed to

just training it through

informational aspects of code.

So actually knowing what good code is,

what bad code is.

and therefore is apparently

more effective.

And I think what they also then do is,

They'll then come in with

that model and then

basically support training

on your specific code and your standards.

So if you've got specific

standards you expect and other things,

therefore it will inherit

the standards through an

additional model.

Like you're saying,

you take that model and add

your context into that

model more and train the

model a little bit more.

Then you don't have to have

such specific context.

You've trained the model for

the context instead and

therefore you've now

It kind of like it

simplifies then the usage.

But yeah.

Yeah.

A foundation model, I guess,

is a highly compressed

version of Tinternet is

like relation in the model

and it can power fashion or

give a reflex answer

immediately without any

deep think and without any use of tools.

But then you want that bit

to be super fast,

but with a company or a

specific bit of extra knowledge,

just so it's super fast at

the really important bits

for you that are unique to you.

Yeah,

it's definitely the way things are going.

Cursor made a big announcement as well.

Oh, did they?

Yeah.

They were about, well, no,

it was a pre-announcement.

The developers that I follow

on Cursor have been

answering people saying,

it would be super useful if

it could use my browser.

And I tried it with

automation for the MCP

plugins to use the browser,

and it was fantastic.

But I did that in a sandbox

for a test account.

You mean like a Selenium thing, as in like,

actually, I'm writing some code.

But to find a trustworthy

thing that has access to your browser.

It was tricky.

I didn't manage to bottom

out on the research,

and the LLMs weren't very

useful at doing deep

research to tell me which

ones were trustworthy or had enough stars,

and I guess it's because it's too early.

But Cursor made an announcement.

Well,

one of the Cursor developers said

they're looking to include

full flow development into

browsers as a thing that

they're going to build as a first part.

To be fair, though, I mean, any IDE,

that you are whether whether

you're using ai or not

getting feedback on this

thing contextually whether

it's going into the ai or

not it's just useful in it

so I guess it kind of makes

sense just as a thing um do

you want to hear my

surprise little new little

surprise I think I think

you might like this um

So the Three Gorges Dam in China,

which they're using

obviously for energy production,

I think it does maybe,

I don't know if it's three

or six percent of the energy in China,

basically provides that level.

But it's altering the shape

of the Earth and its

rotation and apparently is causing,

this is what NASA says,

is causing days to be lengthened by,

I don't know, like,

zero point zero three six

milliseconds per day

because of just the change of

the shape.

They did then go on saying

even a car on a road could

impact by some fractional, you know,

very insignificant fraction,

but still a fraction.

They were then saying about

essentially the human

interference through

I guess,

buildings and construction and all

the other things that we're

kind of doing is, you know,

does alter the planet to some degree.

And it's something that they

need to be aware of environmentally.

And, you know,

maybe there needs to be more

assessments done on what

the long-term impact might

be to these things.

I thought you might quite like that.

That was good.

Interesting.

I would love to know the

actual figures and the

paper that NASA obviously wrote.

If you can provide all those

in the podcast notes or to me,

that would be great.

What are you saying?

You're saying this is fake news?

I don't know what's fake news these days.

And we need fact-checking.

Wow, wow, wow, wow, wow.

Not specifically with you, obviously.

I don't even know what you're saying.

Look, if you see some news... John, John,

you know that scene in

Superman when he goes round

the world the wrong way really,

really fast and goes back in time?

Yeah.

That's not a thing.

That's not what the Three

Gorges Dam is doing right now.

Do you mean that's not a thing?

If I was to go round the world really fast,

I'd go back in time.

I saw it in film.

In Superman.

Yeah.

Also,

if I get in a car and I've got a

friend called Doc...

and I get in a car and you

know it goes super fast

through the space-time

continuum I also can go

back in time and that's why

you you you individually

can in a car and that is

the definition of

relativity fine I will find

the source I mean I know

that I know that it was a

sustainability uh I think

sustainability times is

where I was reading it from but um

I will find the NASA source

to see what it said.

The fact that they mentioned

cars and other things,

I'm not really sure how, like,

it obviously said this.

It is obviously NASA,

they're quoting NASA.

So it must have come from them, I presume.

That bit can't be fake.

As to the context of whether it was,

you know, how close, I don't know.

I'll share that with you.

And then others probably,

maybe we can post it

somewhere so people know.

Absolutely.

Cool.

Well, that is it for this episode.

A little bit over the standard time,

but hopefully it was useful for people.

We'll be tuning in obviously next week,

every week,

every Wednesday and publishing

whatever's going on.

Indeed.

A bit more into cloud specifics next week.

Well,

it depends on what happens in the world,

really, doesn't it?

Well, like I said,

I was at the Amazon Summit

and everything was AI.

So it's very hard.

We're not off topic.

Eighty percent of everything

would seem to be AI related.

But, you know,

there are other things going on.

There was some big

investments I noticed being

made by VC companies in the

platform engineering space

around Kubernetes and

infrastructure and AI.

I think thirty eight mil.

given or maybe maybe

something I think they had

thirty eight mil and then

had a hundred and eight mil

on the second round or

something but anyway so we

can talk about that like

investments looks like the

bcs are now getting back

into the game it looks like

um and now starting to

invest again because I

think there's a bit of a

slowdown and all of that so

it looks like it's picking

up but maybe next week we

could talk about that but

who knows who knows what's

going to happen so I don't

know we don't know all right

an AI uprising and Skynet

and the Terminator.

And the Terminator, yeah.

Just as real as the Superman thing.

And I'm going to now start

sending you loads of

messages on WhatsApp and

see what the AI agent

replies back with now.

Now I know it's not you.

So I'll just speak to you all next week.

See you later.

Bye.