Subscribe
Share
Share
Embed
Authentic, Authoritative, Unapologetic ServiceNow commentary by Cory "CJ" Wesley and Robert "The Duke" Fedoruk
Duke: All right.
Well, we almost blew all the
punch lines in the green room.
So Corey, why don't you tell them
what we're actually, what we for
sure are now going to talk about.
CJ: All right, dude, today we've
got a very special guest with us,
friend of the show, Jace Benson.
Jace, man, it's always
good to have you here.
Welcome to the show.
Jace: Hey, thanks for inviting me back.
I appreciate that.
Duke: Jace is like the
friend of everyone, right?
CJ: Yes.
Duke: nobody's not friends with Jace.
CJ: Dude, like to a person, right?
Acknowledge.
Whenever I saw Jace, they're like,
man, is that Jace guy awesome?
Jace: Oh, man.
I'm just another guy I put on, you
1 leg at a time like everybody else.
So we can all be as good
or whoever we want to be.
Right.
CJ: Yeah, that's so magnanimous, but
still no, man, you are pretty awesome.
And so thanks for taking the
time to chat with us today.
Jace: no, I love it.
I love an opportunity to
talk , with you fine folks.
You guys always have great
insights and inspire me sometimes.
I'm like, well, I really should
question how this works or question why
we're doing this or what we're doing.
What value does this add?
Duke: Speaking about what we're
doing and value adding, , Jace
just released AI in a box.
Tell us about that.
Jace: Yeah, so I really like owning
the things that I want, my blogs.
My content, I like having it as
mine and I think that corporations
don't necessarily want to have their
data controlled by other people.
Entities, And so I set off to make a
self hosted AI solution for service now.
And that's what AI in a box is.
That's like it's core
underpinning value that, and it
doesn't cost an arm and a leg,
CJ: so I'll, I'll leave the capitalist
angle, for a bit later and figure
out why it doesn't cost arm and a
leg when AI is like the number one
through 20, trending items on Google.
But yeah, , tell us a little bit
about what you mean in a box.
That's the part of it that I feel like
it's really catchy marketing, , but I
feel like there's something there too.
Jace: Well, it was called
something else in a knowledge.
Somebody said, Oh, so
it's like AI in a box.
And I said, Oh, I'm going to
use that because really it's
a self contained unit that.
Does the artificial degenerative AI stuff.
It's not the same as ServiceNavs
because it's obviously running on
your own hardware and it's doing
its own, they call it inferences.
It's only inferences of the text you
give it, but it's self contained.
It doesn't make calls
outside to like open AI.
you download the models,
you download the software.
And then it becomes a server that you
can hit like any other integration you
have internally within your network,
CJ: Nice,
Duke: that.
I love the, in the box for that, right?
Cause this is like a box you take
home and it's got your AI in it,
Jace: right?
And you can,
Duke: to you at that point.
It's like
Jace: and, you know, it's
not sending stuff out, right?
You know, it's not, getting
processed out of country.
I've heard some, shared infrastructure
might be running cross country borders.
I don't know what the official
stance on that stuff, but , , it
gets weird, but if you have control
over it, you control where it is.
When it's on.
Who can access it what it does?
CJ: Yeah, and I could tell you having
gone through, , the CTA training
and such, there are definitely some
regulations about, like, where you
can store data depending on what
country you're in and things of that
nature and the type of data, right?
And so, I don't think that that
changes because you're throwing it
into a model and spreading across
whatever happens inside the A.
I.
black box, right?
Like it's still data, And it still needs
to correspond to those regulations.
So being able to control it, being
able to control where it is and
also maybe even being able to
deploy it across your organization.
But in ways that in discreetly, align
with those different regulations, I
think it's also pretty cool, right?
Because you could deploy.
Several of these things, right?
You can have, say, if there are
regulations where, I think Great
Britain has one of these where the data
that's produced there needs to stay
there or something like that, right?
Like you can deploy your own AI
in the box for Great Britain,
but then you can have another
ones , for the rest of the EU, right?
Like that's something that you could do.
Jace: I don't have it set up.
So you can have it distributed.
I have a couple of them at once right now.
I'm still working on just a.
Singular AI in a box per org but
that totally could be possible.
Yeah.
CJ: Look at me jumping ahead on you, man.
So put that on one on your, uh,
Duke: Uh, Bye bye.
Jace: it makes a lot of sense, right?
If you have a company that has, entities
in multiple countries, If you need
to keep certain data ran against in
certain places, like you'd want to
somehow devise a way to say, Oh, this
incident is for a United Kingdom.
Resident it needs to be therefore be
processed on United Kingdom hardware,
that's not possible today, but it
could be like, you could install
it and then you could like, set up
the calls to use that mid server.
CJ: Tell me what you get, I just
bought AI in the box, right?
I got it set up.
I assumed that this was like a Docker
container or something like that.
And, you know, and I
got it up and running.
Tell me how I get that connected to
service now and tell me what I get.
Duke: a
Jace: All right, so right now
it's, I'm not going to name the
price because the price is going to
CJ: No, no, no, no.
You should never name pricing.
Yeah,
Jace: Right.
but right now it's a very low static
price Price is just going to go up as
a fixed amount, but when you purchase
it, you get an update set and you get
a zip file and install instructions
to install the zip file as a server
in your choosing where you are.
you take the zip, Send it over to a
server, you SSH into the server, you
extract the zip, and you run an install
file, and it installs it after it asks
you a few questions, where it says give
me a username and password that will
be used to connect to this machine.
You'll want to note that down so
you can set it up in your instance.
And then it'll start on its own
and then you can leave the server.
You go into the update site after
you've imported into your instance.
There's a settings page.
You build a credential where you
put in that username and password.
the mid server.
in the IP address and port to access
this machine you just set up and it
defaults to the tiny dolphin model,
but you can change that if you want to
use a different model or bring your own
model in for the, share of AI stuff.
Once that's all set up right now, it just
has a summarize button to show that it
does work, but because it's set up using
like the event system under the hood.
You can call it from any back
end APIs you can think of.
So I whipped together a subflow,
so you can just call a subflow
and say, Hey, summarize this task
and it'll summarize that task.
And you can choose to take the output
of that, write it to the work notes or
send an email or whatever you want to do.
Duke: can we dive deeper into
that because it's just , yes,
I got this new, AI in a box.
It's totally mine.
It's affordable, but the really important
thing is what does it do for me?
Jace: thinking about what value
does generative AI in general
add to ServiceNow or any tools?
And it's not necessarily that,
oh, it can just summarize text
or do these things really.
What are we after when we're looking
at this generative AI stuff, right?
We're looking for either
feedback or direction or write.
if we're summarizing something, hopefully
we're using that summary to save us
time, but not reading the whole record.
But really, maybe we should summarize
and reassign like that should be like
the action that happens like only
really summarize it when it's going
to get reassigned to the new group.
So, the new group can just read the
summary instead of the whole spiel,
CJ: Yeah.
Duke: Maybe I'm getting ahead and
putting more on your to do list
like Corey, but can it summarize
more than 1 object at a time?
Jace: Yeah.
So.
Duke: incident, can I say, give it a
list of incidents and have it summarize
basically what's been going on this week?
Jace: So right now, the only,
, functionality that it has is just, I'm
calling it simple summarization, which
is you say, give it a specific task.
You give it a template that kind of
looks like a mail script, and then you
can use a scripted notation inside
of it to give it related records.
So you can't say like summarize
all these incidents, but you
could write a background script.
To like generate an event against
all those records and have
it summarize all 10 of them.
But I won't summarize them
as a collection right now.
If you understand what I'm saying.
Duke: Yeah, it's not 1
summary is 10 summaries.
Jace: It's 10 summaries today.
But if, I need to add like
a general inference, option.
So you could say like, Hey,
take these things and use
this prompt and deal with it.
Right.
So you could say, Hey, here's 10
summaries, summarize these 10 incidents.
In that case, you could do
it, but I like that idea.
I need to add that.
That's a very basic core, like
idea that I realized I don't have.
CJ: So aside from Robert and I adding a
bunch of stuff to your like to do list,
what I really like about
this concept, is the trust.
With AI, there's a lot of,
conversation around trust.
like we started off talking about,
like, where's the data going
to be, who has control of it?
Right?
Like what am I shipping
my data off somewhere?
That's not gonna be used to train a model.
That's gonna be, you know, be, federated
out to a bunch of different people.
And then, well, my data starts showing up
in their answers, when they ask questions
of the eye with this situation, right?
Like, I think the number 1
selling point of this is that you
have the trust that your data,
stays inside of your protection.
I think the number two thing for me
once I trust that, you know, okay, I
can use this AI for whatever I want
it, to Robert's point, it's like, all
right, so it's like, what next now,
what can I have it do for me now?
So Jace, one of the things that, I hear
a lot from my clients, Is that, when we go
in and looking into like open AI, right.
Like we kind of know what that does.
Tell me what like AI in a box does.
And then how we could inject some of
our data into those AI models, to
make all of this work a little bit
better for us versus the generic in
the sky solutions that exist currently.
Jace: When you're asking a large
language model to do something.
It's just guessing the next best thing.
It's like the most expensive
autocomplete system you can buy.
Right.
CJ: All right.
Jace: right.
That's what, that's what AI is.
But when you give it like 3 paragraphs
of suggested best thing, best next
things to do based on the current text
that's given, it's going to give you
like the best next answer, best next
thing because it has it right there in
its context above the data you're doing.
CJ: Gotcha.
Right.
Jace: a user is asking for a answer to.
How should I, I'm trying to think
of like a good example here.
Yeah.
Say, like, you wanted to
add, , suggested resolution notes.
Right?
And you don't have it yet
because the instance not solved,
or maybe you put it resolved.
But you don't know if it
should list a list of steps.
Or just like, thank you, for letting
us know this issue is resolved.
You may want to give it a
couple examples of hey, here's
an incident that was resolved.
And here was a resolution
note of user called and said,
this isn't an issue anymore.
And maybe you'll give it another one that
said, have the user reboot their computer.
And then they rebooted it and they had
to install a patch, For the same thing.
And if you give it that kind of
context, it may use that stuff.
Those things become more likely as the
next best token to be included in your
resolution notes, especially if you asked
it to, hey, take the last few resolution
notes from the previous incidents and
give them as an example for the current
short description and work notes given.
Um, So then it's you're having
an inference against content
that makes sense for it.
But building those prompts like that is
complicated and hard to talk through.
CJ: I follow you.
Duke: is that the work that you were in
a box is taken out of the system doing
Jace: I want to leave some of it
in the system to like build those
prompts and then send it over to the
self hosted large language model and
vector databases to, get those answers.
So there's
Duke: I see.
So, the configuration of it
is still inside service now.
Jace: the configurations
inside ServiceNow.
I presented this a few months
ago to one of the partners,
and their feedback was clear.
Jace, this is way too complicated.
And so I spent like two months before
knowledge, like simplifying it down.
And it's really simple right now, but
I'm re enabling things that I had that
were really complicated to set up.
CJ: Yeah.
Duke: been working on this?
Jace: Well, do you remember ScribeMonster?
Duke: Yeah.
So like, you're talking
like a year, right?
Jace: When I stopped working on that,
about three months later, somebody said
something on Discord and I thought,
you know what, I'm going to try.
And I started down this
path of a self hosted.
Largely like a self hosted
solution for service now.
Duke: Did you know AI before that?
Or
Jace: I knew what I learned.
Duke: your, your masterwork in it?
Jace: I knew what I learned from
building that scribe monster project,
which I recently open source.
I learned a ton from building that
And I'm using a lot of the things
I learned from that going forward.
Duke: It's a ton of energy, man.
It's a lot of energy.
, CJ: did you intend on scribe monster
to be like a learning project to
eventually get to AI in the box
or did it just work out that way?
Jace: I built scribe monster
to learn AI, that is true.
I started that project and I
shipped in like a week, right.
It had a bunch of
updates to it eventually.
And then I just stopped updating
them because companies became
like adverse to running anything
on anybody else's hardware.
Duke: Can we just talk about the
appreciation for how much of his extra
energy Jace has , spent on this whole
thing, , not only just scribe monster,
but then getting to the AI in the box.
And where do you find the time and
energy when you've got a full time job?
We got to get Jace a sample
of MagicMind if we can.
MagicMind is a channel sponsor of ours.
It is the world's first
productivity drink.
It's got matcha.
It's got nootropics.
It's got all kinds of good stuff.
Sometimes I'll have it with my coffee in
the morning, in the afternoons around this
time where I usually have my three o'clock
crash, take my magic mind, boom, I'm back
into flow state for the rest of the day.
So Jace, thanks so much for, taking
all that extra mental energy and
building these things for us.
And we'll see if we can get
you a sample of magic mind.
If you're interested in magic mind, just
go ahead into the description below.
You'll see that there's a link.
Use the promo code CJ and the Duke 20,
give you 20 percent off your first order.
Jace: I'll definitely
have to give it a try.
That sounds like something I could use.
CJ: Piggybacking off of that, I think
it is a really good point that you
have used a significant amount of
your mental, capacity, to enhance the
ServiceNow ecosystem as a whole, it's
not lost on me how much you give back.
To the community, And as I know, it's
not lost on a lot of folks out there
because they're the folks that you're
giving back to, everywhere I go on
LinkedIn, I see, you know, folks, you
know, mentioning your name about just
something really cool that you've built
that they've gotten value out of or.
A conversation that you've
had with someone or, some help
that you've given someone.
So I just wanted to, pick it back into
what, what the Duke was just saying,
express that appreciation on behalf of
the community, . For all the work that
you do, you know, that's number one.
and then number two, I do actually want
an answer, , where do you find, where
do you find all the time and effort, man?
Jace: when I get into a phase of building,
I drop a bunch of other hobbies that I do.
I don't play video games, I don't do other
things, I want to see a thing finished.
I want to get it done.
So I spend the time after I could
put the kids to bed until I go to bed
working on this kind of stuff, which
gives me a couple hours every night,
Some nights I don't get anything done.
Other nights I do.
I wish I got up earlier so I was a more
morning routine person, but I'm just not.
I'm not wired that way yet.
But that's where I find the time.
It's before bed, generally.
And then I think about when I'm not
working, because it excites me because
these are the passion projects that
I, the things that I care about.
Duke: if somebody was interested in AI in
a box, tell us roughly how it deploys,
how long that takes and the first things
that they would do with it, to make
sure that that money was well spent.
Jace: Very simple.
I've been re simplifying that over
and over again, running through it
with different folks, , who I've had
trying this thing now, you, you pay
a fixed flat fee, and then you're
given that zip file and XML, you
download those two files, you push the.
Zip file to a server with like SCP
or our sink or magic wormhole or
however, you want to bring that file
to that server, you deploy it, you
install it, like I said before, and
then you connect it with the instance.
Once you install the update set
and run through the install page.
And right now, the immediate value is, you
have a sub flow that can summarize a task.
So you could say when an incident
gets reassigned, you could have a
work note written that says, Hey,
summarize this for that new group.
That'd probably be the
most value add today.
I have it presenting on the form when you
load a form, if there's a summary that
exists, that's newer than the last update.
And that's useful for presentations,
,
Duke: one thing I'm really curious about
is when you say it summarizes a task.
. what all is it summarizing about a
task and does it do as good a job
on custom task types as out of box?
Jace: Right.
So I made it.
So the update set includes a
table, which consists of a.
The table field, and then it consists of
a record template where you say, this is
the template I'm going to use to present
the data to the large linked model.
And then you're given a few other
fields to build out the few shot.
You can put in like a record.
It'll build out what that
data would look like.
And then you can modify that data.
So if you wanted to like, I don't know,
scrub something or add some other thing
to the example data, you could do that
for the two records that it gets included.
And you can use any custom fields that
you'd want within that, just like you
could a mail script and a notification,
Duke: One use case I see you can
go right away is, in projects,
there's the project status report.
Right.
CJ: Oh my God.
Duke: And also projects and
demand just have slews of.
big text fields
CJ: Yep.
Duke: demands, especially I mean, for
crying out loud at vivid charts, we
built something that we didn't like
build a summarizer, but we built a report
to present as a slideshow of demands.
imagine going to the demand and just
seeing the summary rather than the essay
somebody's written in every single field.
Jace: right?
Or that on a change.
I mean, I cannot believe how
much data is in a change,
Duke: Oh, yeah.
Or just, or just like, or the projects
CJ: hold on, hold on, dude,
dude, like change, right?
think about the people who approved
changes, typically management, often
upper management in some places, right?
Like they don't have time
to read all of that crap.
That's on the change record.
Right.
But you need all the crap
that's on a change record.
that approval record should
have a summary, right?
Duke: bingo, man.
Yeah.
Jace: I know the folks on listening
to this can't see, but can I
share a visual with you guys
and you guys can describe it?
Duke: you why.
Well, I will, I will get
you on the YouTube channel
and we will get you a video.
Jace: Well, here, I'll, I'll,
Duke: I don't know, I don't want to be
Jace: I'll, I'll throw this thing up and
we can, talk about it and cause then,
Duke: like an episode of some cooking
show where they're just eating in front of
you and you're like, damn, it's so good.
I wish you guys were here.
Jace: Can you guys, can you guys
see this AI summarizers table?
CJ: Yep.
I just put it on the stage.
Duke: Yeah,
Jace: Very, very simple table, right?
So let's just take a look at change
requests just because it has a little
complicated caveat, So in here you define
the prompt of what it's going to do.
You define the record template.
Duke: So the record template
is just, it looks like a bunch
of labels and field names.
Jace: yep, so it literally just
reads it out like current dot
when it does the evaluation,
It just reads out those values.
It gets the display
value, which is critical.
But here, I'm going to make a
change to this just so it triggers
a change to everything else.
so you can see it updated
this record a data and you can
see it builds out this with
Duke: It's just like a
sample record, right?
Jace: well, this is an actual
record in the instance.
Right?
So it read change 56.
it went ahead and got all
of the related approvals.
CJ: dude, dude, okay,
Duke: Yeah, I'm with you.
Jace: But it uses this as the
context when it makes that call
to summarize another change.
CJ: There are people at home, dude,
there are going to be people at home
listening to this like, what are
they talking about, but listen, man,
like the seasonings Jace, oh my god.
Duke: Okay.
So basically like, can it collect
data about sub records too?
Yes.
Jace: but I realized, oh, I probably
should include , The tasks has like,
well, I'm going to want to do this for
other tables, like approvals for changes.
And so I added this syntax.
but effectively anything that's in
this double bracket is evaluated.
And if you prefix it with
related, it'll do a query against
this table with this query.
And these square brackets are
different because it is Evaluating
Duke: management.
It's a spaceman stuff to me, but
Jace: but it, but it
reads the current CIS ID.
It says, Hey, give me all the approvals
Duke: right.
Jace: to like with this CIS ID
and then return these two fields.
Duke: So, Even if you're not a pro dev,
if somebody just explained, listen,
you just put these symbols around
it and it's going to work, like you
could totally get by it reminds me a
little bit about the squiggly braces,
dollar sign stuff with notifications.
Like I want to put the
number in the field.
That's what it reminds me of.
But again,
Jace: right there.
Duke: think about change.
I'm telling you, SPM we'll be kick
ass for this because everybody hates
writing their project status reports,
Jace: Yeah.
If you, if you'd like,
Duke: stats,
Jace: if you'd like, I'd like to
build an SPM summarizer for you and
project just so we can walk through it.
Not right now, but if
Duke: Yeah, absolutely.
Let's let's do it on video, but
Jace: yeah,
I would love to.
Duke: you can almost get it to just full
on automate the project status reports.
Jace: Right.
Okay.
CJ: that summary of , what's
going on in real time.
All just condensed into.
The relevant details, some people
don't need to know like what port
and switch that you're moving
the cord from, they don't care,
we're getting rid of that stuff.
We're bringing all the stuff that
matters, put it in front of people and
like a concise, this is what we're doing.
These are the people who have
already, weighed in on this, you
know, here's your decision maker pick.
, dude, you have no idea,
how much I am seeing
Duke: first
CJ: the analysis paralysis, of,
working with the ServiceNow platform.
so much data.
so many different interfaces.
The data is not nearly concise enough.
It's not presented in a way to
make it easy for folks who don't
have a lot of time to engage with
the system to make good decisions.
All of this stuff.
I'm seeing this thing that this AI
in the box, that you built answer
all of those questions for me.
I got my credit card out, dude.
Jace: Well, it's, money well spent, right?
The pricing model is
kind of different than a.
Regular pricing model we've
seen, cause I'm, I'm tired of the
sasses, just how they're priced.
It's so hard.
Did you guys hear about the
new Adobe, like Congress
Duke: insane.
That's
Jace: Adobe?
Duke: yes, I did.
And it's.
CJ: Catch me up.
Duke: from what, from my reading
of it, Adobe's new terms and
services is basically you give them
an unrestricted license to anything
you create, they can reproduce it
and use it in any way they want.
Jace: Oh, that's not
what I'm talking about.
Duke: Oh, really?
Jace: I mean, that, that's
important too, right?
But , the way that they
charge for it, right.
They charge every month for
it, but you buy it annually.
So you pay 700, 60 bucks a month.
But if you cancel mid contract,
you have to still pay half
of the remaining contract,
Duke: That's cheesy.
Jace: right?
But like, my whole point is these
software as a service solutions,
, they're fantastic to get paid from,
but they suck as a consumer as a
company needs to pay for these things.
So there's this company I
follow called 37 signals.
They're the guys behind Ruby on rails.
and they started this thing called once.
com and they're like, there's
something broken in the system and
it's how we sell some software.
Not everything needs to be a SAS.
Let's sell some things and let people
own them where they don't have to
keep paying a recurring fee every day.
And so they came up with Campfire
as a competition to like Slack and
Microsoft teams, As a chat thing.
Cause like, it doesn't necessarily
need to be up all the time.
It will be up all the time, but
it doesn't need to be persistently
everywhere all the time.
Like ServiceNow makes sense as
a SaaS, But does the generative
AI functionality make sense?
And so it's like a recurring fee that you
pay every month for on top of service.
Now, I don't know, I think you
should be able to own that stuff.
And so, because of that, I'm
selling this at a fixed price.
You pay once you get access to it
forever because it's your code at
that point, and I'm going to support
it until I get the version 2.
Duke: You might want to consider
Camtasia's model, which is super awesome.
you buy whatever version you buy
CJ: get this
Duke: and then it's not a monthly fee.
It's 200 bucks or something for Camtasia.
But then they keep on developing
it and putting out new versions and
then you buy the next version, right?
Just.
Jace: Right.
CJ: Right.
Second.
I am
Jace: set for version one and then when I
get the version two, if I want to add more
features, I'll sell that as version two.
CJ: And then you'll probably do
something like a support contract,
because since you're going to be selling
this enterprises where they can, opt
into like upgrades and, , certain
hours of support, they get you on
the phone and that sort of thing.
Right.
Jace: That's not on the website,
but that's planned to be there.
So you pay for a bucket of hours or you
pay per hour for support and then also
workshops to help people understand like
how this works and how ANN gen works.
CJ: Yeah.
So this is a labor, not, not just a
labor of love from the , the standpoint
of doing a technology, but also a
philosophical perspective on how we're
selling software and how you feel like
it should be done in a different way,
Jace: Yeah, not everything needs to
be so difficult to buy I struggle with
the value when everything is a SAS,
CJ: now, for sure.
Duke: It's a great capstone
for the whole show.
CJ: Absolutely.
Jace, man, this has been great.
incredibly informative.
I learned so much today and I'm
super, super stoked for this.
Duke: The click point for me was once
you figure out that the related records
can be part of the summary, there's
so many modules that are like that.
I mean, GRC, like, imagine again,
summarizing all your control tests,
Jace: right.
Duke: of them, thousands of them.
CJ: Oh my God, CMDB, CSDM, right?
Like summarizing all of those
relationships and giving me like a
quick overview of what happens if this
goes down, like, can you build, can
Jace: I mean,
CJ: summarize impact management?
Like what happens if
this switch goes down?
Duke: we go.
Another 40 minutes.
Jace: Well, so like, you have to
realize it's only going to know
to do what you've told it to do.
Because it's, it's like a spell checker.
Right?
So, right now, if you were to , include
like a CMDB summarizer to say like,
say you added UI action that said
give you like an impact analysis if
something went down and then you pick
a CI in its chain of records, you'd
want to give it examples of how that
actually would work with real things.
So it could infer, how the response
should look and what it's looking for.
like I said, it's just guessing,
CJ: I got you.
So the answer is yes, but I'll take it.
Jace: Right.
It might get you close, It
might give you a starting point.
Like, Oh yeah, this, actually would
have this problem or the problem
would happen here, but maybe it would
still be available on this port and
not that port, you know what I mean?
Like there may be redundancies.
It may not like suggest.
Cause it's simple.
Duke: The thing is, on this show,
we've, come up with, 3 or 4 just.
Fantastic good use cases for this and
that's with , a little less than 40
minutes of thinking so we are going
to get you the links to get in contact
with chase if you want to see AI in
a box and Man, thanks for watching
and we will see you on the next one.
Jace: Thanks again for having me.
I really appreciate this time
of yours, , Robert and CJ.
Duke: You're welcome
CJ: Thanks Jane's.