Understanding Practical AI and the Future of Automation – With Joseph George

Channel: Alex Kantrowitz

Published at: 2025-04-07

YouTube video id: 8NMpKU7sL4Y

Source: https://www.youtube.com/watch?v=8NMpKU7sL4Y

There's been so much talk about what
generative AI can do and little about
what it's actually doing today. So,
let's look at how it's being applied
practically. In a conversation with Goto
general manager and SVP of its IT
solutions group, Joseph George, we're
doing a deep discussion looking at the
technologies applicability in an
interview presented by GoTo. And I think
you're all going to really enjoy the
discussion and the examples that you're
going to hear from Joseph today. Joseph,
great to see you. Welcome to the show.
Thank you, Alex. It's a pleasure to talk
to you this morning. So, GoTo is the
company that has underneath it log me
in, which is the service that I'm sure a
lot of us have used where you having
you're having a technical issue with
your computer, somebody will log in and
then help you fix it. And it's
fascinating to me that you're starting
to use artificial intelligence in this
process, not just in a surface level
way, but a deep way. So, can you tell me
a little bit about how artificial
intelligence helps you with that
process?
Absolutely, Alex. Yeah. So, absolutely
correct. Go to provides a set of
products and services uh across the
communication space, collaboration as
well as solving the needs of IT
customers and and for the IT products,
we're branding those Log Me In. As you
mentioned, LogM is a brand that's been
known in this industry for a long time.
Uh and you'll increasingly hear us use
Log Me In for branding our products and
services in the space. And absolutely,
they're being transformed by AI. uh it's
amazing is we live in an age where
you're looking at what types of problems
you solve for for customers and it's not
only about solving those problems more
efficiently in a better manner but it's
about solving a whole range of new types
of problems that we weren't even
thinking about before. So I'm excited
about being in the space at this time
and also running the business at at log
me in for a broad range of products as
well. Okay. And so can you give us like
some examples? You're talking about how
you're going to use AI to solve current
problems and longer term problems. So,
how's that actually working in practice?
Absolutely, Alex. If you think about
Goto and Log In, we also serve a range
of customers uh for small medium
businesses and managed service
providers. Uh we offer a unified
endpoint management platform. So, it's a
platform that think of it as delivering
it in a box, right? Those are small
companies. they want to be able to get a
solution that provides a broad range of
their IT needs for managing their
endpoints. Uh and also for enterprise
customers uh we have uh a product suite
uh called uh rescue. Uh and rescue has
been in this indust industry for a long
time. Uh rescue has always been about
connecting the expert to that end
system. So, if you or anyone listening
to this podcast, if they have an issue
with their laptop, you know, you end up
opening up a ticket and then somebody
comes in from your IT department, right?
So, you've got a technician that
actually gets access to your laptop and
and they actually fix the problems by
logging into your lap laptop and they're
fixing those issues. Uh, and it's always
been about access. So connecting the
expert to the system, how does the the
expert connect in seamlessly, securely,
and impact the the end user as as little
as possible. AI changes that because
it's no longer just about access. So
think of the steps that the technician
goes through, right? When they're going
in fixing a problem on your laptop and
when they're done, typically they've got
to go document that or they've got to
create a knowledgebased article. Those
are the ownerous steps that they often
end up skipping because they don't have
time. They're going from one call to the
next or they don't do it well. And this
is where AI comes to the rescue, right?
AI can now start to capture what happens
in that session, automatically create a
session summary and then create a
knowledge article and and actually not
only create it, but also help maintain
it as well. And that's just one example,
Alex. We'll talk about many other
examples. It's amazing what what's
happening in the space. Yeah,
definitely. And I can imagine for most
people there's the doing the job and
then there's a documenting the job. And
you know we've always been talking about
like what's the AI use case going to be
and there are these like broad buzzwords
that AI will help people do more
efficiently and do more of the human
work and less of the wrote work. And I
think what you're pointing to here is
like a pretty good example of you have
an IT technician who's going to come in
and fix a computer in the background.
Now the AI is watching, taking notes and
helping them basically put together the
documentation uh that they would be
required to put together afterwards.
Stuff they don't really like doing. I I
would imagine because you get into, you
know, IT services, you really you care
about fixing stuff. You don't care about
documenting it. And so this is one
example, I think, of how you can use
Genai sitting in the background. You go
in, you remote in, you do what you need
to, and then all of a sudden all that
documentation is waiting for you. Am I
am I c am I getting it right? Yeah,
exactly. I mean, it's the toil, right?
The toil that you had to deal with, it's
starting to take away that toil from
your day-to-day. And and so think about
the technician, right? The technician is
now actually spending time exactly like
you said, solving real problems. The
toil is taken away. Uh it's not only
documenting those steps, it's creating
knowledge articles. And there's a couple
of additional applications from that.
The first
is the next time you get a call for a
similar sort of problem and a less
experienced technician has to solve it,
guess what? They have a head start
because now those steps have been
documented. They can now follow those
steps. And even better, we can now
actually automate and create scripts to
automate those steps. So rather than
have somebody go through manually click
through a whole bunch of steps with
Genai, you can take a summary and create
essentially an execution plan and start
to run those and and so the world of
having a human technician now actually
supervising what I'd call virtual
technicians that can execute commands on
their behalf and run in parallel on a on
a number of different machines at the
same time. That world sounds a little
bit like science fiction, but guess
what? It's true. It's already here,
right? We have our products as part of
our uh unified endpoint management
product. Uh that product set is is
branded uh log me in resolve. We have
those capabilities available for
customers today. Uh and we're continuing
to work on and improve that. And and and
it's amazing the the not only the
productivity it drives, but also the
efficiency uh and also in terms of
customer satisfaction. It's not just
satisfaction on the end of the
technician. They're more effective. they
can learn more quickly, but also the end
employee here whose laptop is being
fixed. Uh you're getting a better
experience at the end of of this as
well. So Joseph, I think what you're
saying is pretty profound because I'm
thinking about there's so many
professions that you need to like
basically document things at the end of
your work. Um my father for in for
instance he he just retired but he was a
doctor and he you know what he would
spend his nights literally submitting
paperwork and I always said like it
would be great if there was an AI that
could kind of be his co-pilot and stand
there with him or or sit there with him
whatever it is and then at the end of
the day when he's done with his patients
it takes the notes it summarizes and
then he just goes and checks. And so it
it seems like we're going to we're
definitely seeing that in medicine.
We're seeing it in IT. We're seeing it
all over the economy. But I think what
you shared was a step even further than
that, which is that once you have this
AI sitting in the background, uh, and
we're going to talk, by the way, cuz I'm
definitely interested in hearing how you
built this. Uh, but just let's focus on
the use case for a moment. as it sits in
the background and it views case after
case of similar problems. It now can
take effectively the combined knowledge
of the entire organization and then
share it in a a digestible way with even
a junior technician to help them make
sense of problems they hadn't seen
before. Yeah, 100%. I mean, and there's
there's multiple aspects to it as well,
right? The the first is as a technician
is fixing a problem, getting assistance,
actually having guidance, right? So the
AI can actually start to give you guided
advice and and and tell you what about
this, did you miss this step or how
about this thing that that you should
have done? Right? So now you you have
like you said a co-pilot or an assistant
actively helping you through the
process. So you're able to debug and fix
the problem more quickly and
efficiently, but also the fact that
you're able to create the summary. And
then even more interesting is the
summary is valuable. That's the easy
part from an AI perspective, right?
That's that's what you're getting from
the LLM. It's it's it it basically
understands uh based on the steps that
are being taken how do you actually
create a summary from that. But what's
fascinating is is taking it from a set
of steps to actually uh an automation
task that can be run and executed. uh if
you think about the before world before
AI you actually had to go ahead and use
maybe a low code no code tool to define
a workflow right how do you convert
those steps into a workflow that can be
executed and now instead the LLM is
automatically generating that execution
plan so it can be run and even better as
you run that script uh the LLM is also
determining if if you encounter behavior
that you didn't expect it's resilient
enough to be able to work around that
and so you're getting much more
reliability and automation in practice
uh that's actually helping the customer
and also helping the technician as well.
Wait, so are you saying that not only is
the AI watching in the background and
helping to give advice to let's say less
seasoned technicians on problems they
may not have encountered before or maybe
just a few times and helping them work
through it. As the system watches more
technicians remote in and solve problems
for people, it can now in an automated
fashion start solving those problems
itself. Yeah. As as you see the first
time you've seen the problem and you
understand here's the signature, you've
captured the steps to fix it and then
you see another ticket show up. Uh you
you immediately understand here's this
ticket. It's very similar. I already
have the recipe. I have the steps to be
able to fix it. And now you can actually
have a virtual technician kick that off
and start to fix it for somebody else.
So you might have three or four other
customers call in, right, with exactly
the same problem. And instead of the
human technician having to respond to
each one, you can have the virtual
technician start to fix those problems.
Now ultimately obviously you want to
make sure that the human being is in the
loop uh and they're able to uh supervise
and determine how much the virtual
technician runs. So that is the
environment we're in right now where
it's a coexistence of human agent
working with these virtual agents. But I
can see the virtual agents taking on
more autonomy over time. Especially if
these are proven types of problems uh
and you understand and you've got a
success rate of solving them, then you
could reach a case where there's even
more automation happening there as well.
So, am I picturing it the right way that
there's maybe like a human agent sort of
like as a manager of these virtual
agents that's seeing requests come in
and saying, "Okay, we have the fix this
problem agent, so we're going to go send
that out and have it try a bunch of
diagnostics and resets and fixes and see
if it can solve it and then go and check
back with the client." It's even better
than that, right? It's it's basically
where the tool can tell you these three
problems, these three tickets are
actually associated with these desktops.
So even even the simple task of saying
the ticket is associated with this user
who has these desktops, that's usually a
manual step, right? Somebody has to go
through and figure out which laptop are
they calling about. So all those types
of steps where you're actually getting
to here's where the problem is, that's
automated and now it's actually able to
tell you here these are this is the
problem that we've seen. the virtual
technician can go ahead and execute it.
Go ahead, press press execute and it'll
run in the background. You can supervise
it and when it's done, it tells you if
it succeeded or not. And you can see
what what's happening in parallel across
a set of laptops at the same time. So
you see the the whole advantage. Number
one, it's far more efficient for the
technician. They don't feel stressed
where they have to respond to all these
different uh problems and switch from
one call to the next. A lot of the toil
is being taken away. Uh and also for the
end customers, it's much more efficient
as well. They're not sitting in line
waiting for the technician to free up as
well. So that's the beauty of what we're
seeing. It's basically where the
employee or customer experience is much
better as well. Uh and the technician is
now spending time more creatively.
They're not spending time doing manual
steps and running through it every
single every single hour of the day.
That's pretty remarkable. All right. As
a journalist, I have to ask you, this
stuff works. Absolutely. Absolutely.
This stuff works and and it'll only get
better, right? what we're doing, what
we're seeing uh especially as we look at
uh more specialization as as you get to
much more domain knowledge around
specific types of problems, we'll see
that improving as well. But generic
types of problems absolutely works works
really well as well. Uh and and and it
actually if you think about the whole
endpoint space, it's taking us from uh a
mode where we operate very much
reactively. This is always about how
does the user figure out there's a
problem and call in and you get help.
Everything we talked about Alex was
about improving that process. But think
of it as one step further where now you
know I described we've got a unified
endpoint management platform uh that's
branded uh log me in resolve and there
we're actually monitoring your
endpoints. We we've actually got
telemetry so we can see if something's
going wrong. uh we're also able to push
out patches and fix things. Uh and
imagine we've got these support calls
also happening as well. So in that
environment, think of where we could
start to monitor, see that there's a
problem ahead of time and start to
trigger these virtual technicians even
before the user calls in. Uh that's
obviously work in progress, right? We're
not there yet where everything's
happening automatically, but we've got
all the pieces and we're putting those
together and that's the world we're
getting to firmly where you don't even
have to call in and talk about the
problem. The system detects that there's
a problem, tries to fix it ahead of
time, and then ideally informs you
afterwards. There was a problem, we took
care of it. Just imagine the savings and
the and the satisfaction that you have
from a user perspective. So the road map
is to go to basically reactive. somebody
has a problem and right now like maybe
there's a human technician that's going
out and remoting in and fixing what's
going on to a place where I think you're
this is what's happening now where AI is
going can basically handle this
virtually to one step even further than
that which is if you anticipate there's
going to be a problem on a computer the
bot can come in and fix it before
someone even calls support. That's
absolute absolutely where we're headed,
right? And and that is because you've
got you've got the telemetry that you're
collecting from those endpoints. You can
start to figure out there is a problem.
Some something's happen, right? You're
logging in and every time you log in.
Sometimes even before you notice it, it
can start to look at the data and and
see that your performance is actually
slowing down. There's an issue here. And
by the way, we've seen this problem
before. We know what the fix is. We can
go in and fix it. Right? That's the
world that we're going to. And it'll be
a much more productive world. Uh
obviously the role of the technician
changes at that point right you're not
trying to reactively respond to things
and running through runbooks and manual
steps at that point a lot of the toil
the drudgery is taken out of your
day-to-day and now you're actually able
to focus on more value creating
activities as well right so it
absolutely changes the role but it's
going to be a better world for not only
technician but also the the end employee
as well it's so interesting because what
you're describing is something that I
think we can see in so many industries I
mentioned medicine already. Uh we're
also on the channel either live already
or about to be live. We have a
conversation with uh Tom Egimire, the
CEO of Zenesk. And it seems like this is
where customer service uh in general is
headed as well. Um where else do you
think this type of technology might be
able to be applied? Uh honestly if I
think about any environment where you've
got an expert connecting in remotely to
a system uh and experts going through
their knowledge and and and the
experience they have uh is there a way
for the tools to become experts right to
to actually learn from the experts and
to start to provide that expertise right
so any environment where you have that
setup this applies right and and and and
and think about the savings instead of
somebody actually physically going out
to uh a device that's that's out there
and and actually trying to figure out
what's happening. We fixed the problem
by having the ability to remotely log
in. And and now if you're able to take
it one step further where it's not just
a connectivity tool for experts, but
it's actually providing expertise,
helping uh newer employees learn from
experts and and start to automate steps
and create automation scripts
automatically. uh that starts to become
a much more efficient and productive
environment for for all concerned.
I want to know how this is going to
change uh the workforce in the future
because I hear these stories and I kind
of I want to think of maybe the bright
situation where uh let's say you're a
technician, you're now really just
working on the most thorny issues and
some of the basic stuff the AI is going
to take for you. Uh, but I do think that
this could really have a profound change
uh on the way that we work. Um, if I if
I'm just getting started, this is a
question we get often. If I'm just
getting started, usually what happens is
the company pays me a low salary to
start, you know, and learn on the job,
uh, mess up a bunch and, you know, sort
of get my bearings, do some of that
lower value work that still needs to be
done and then get up to speed. But the
more I hear about these AI systems, the
more I wonder like whether that job is
going to be something that we're going
to still have in the coming economy. So
I just want to get your thoughts on
since you're building it, how the
workforce is going to change with this
technology deployed. Yeah, I mean that's
a really good good point and we always
worry, right? Is the AI going to replace
the the human being? Is our IT work is
no longer needed because of AI and and
ultimately I look at this as a tool,
right? So AI isn't replacing human
beings. It's essentially humans and IT
uh technicians that use AI will replace
those that don't. Right? I'm I'm sort of
borrowing a line from Eric Brenelson at
the uh the from Stanford who who
actually used that example and he used
in the context of lawyers but I think it
just applies equally in the IT setting.
So what I see is roles will evolve right
where uh you had mundane repetitive
tasks those types of things will start
to be automated and done much more
efficiently by by the by the AI right uh
and and and obviously as that frees up
capacity I think part of the challenge
is because within an IT environment
teams are always reactive they're
resource constrained they're always
dealing with uh putting out fires and
and they're not able to dedicate time
How do you evolve? How do you optimize?
How do you actually stop from spending
your time on break fix to actually
providing value for the organization
right as an IT team? Can you actually
help the business differentiate and it
no longer becomes uh one where you're
constantly fixing problems proactively
uh reactively, but you're instead
proactively going in addressing those
issues and those employees can now focus
on much more creative types of tasks. Uh
I I think that becomes important for us,
right? We we we can't train people to
just execute tasks. We have to make sure
that even the way we learn and even the
way we approach problems, it has to be
about creativity. It has to be about
thinking and solving and connecting the
lines that the AI is not ready to do
right now. Yeah. I think there's like
such a thing as over reliance on AI. I
just saw an example on Twitter where
somebody talked about how like they had
cursor they worked with cursor uh on a
fourmonth coding project and started to
trust too much to the AI and then it
sort of like wiped their progress. But I
think what you're talking about keeping
the human in the loop being uh realistic
about what this AI could do versus
turning everything over to it. I think
that's probably going to be the way uh
that businesses end up putting this into
practice in a way that benefits
everyone. Yeah, absolutely. There's got
to be checks and balances in the whole
process, right? If you think about it,
you've got to make sure that whatever
you're executing is is going to work,
right? Not only reliably, but
consistently as well. And so you have to
make sure at some level if if there's
certain set of AI steps, is the human
being validating that, making sure that
it's it's working correctly. And
similarly, if you've got human beings
doing tasks, the AI could be uh
providing guidance and and telling you
about things you've mix you've missed.
So, it's really a a close collaboration
between human beings and AI agents
working together. Uh the role of AI
agents as that space matures, you'll see
them doing more specific tasks uh rather
than generic ones. So, that that will
evolve over time. Uh but again, is it
going to fix everything and and do
everything automatically? uh you know we
we've seen these these trends in the
technology. There's obviously certain
sets of problems that it can solve
really well, but there's sets of
problems especially when you're thinking
about uh predictive types of
capabilities, uh causal types of
analysis, reasoning, there's all these
traits that human beings beings beings
bring to the uh equation here. Uh and so
AI will evolve, but there's plenty of
space for human beings to be able to
contribute uh for the foreseeable
future. That's be optimistic about the
whole scenario. Yeah, I want to know how
you built this stuff. So you're talking
about very advanced uh artificial
intelligence that's not just sitting
along and figuring out what people are
doing but synthesizing and it seems like
learning from different situations. So
what's your tech stack? Yeah, I mean
it's it's it's smart engineers working
on this, right? So, uh, generative AI is
is the is the main building block for
this and and we're using standard
foundation models. Uh, and and
essentially a lot of the work is is how
do we make sure we have the right prompt
engineering in place? Uh, we've got the
right uh the right uh prompts that we're
sending and and and then we're also
training this on data and and making
sure that we're uh adjusting our our IP
so we can actually solve these problems
reliably. uh and and that's where we are
right now. Uh obviously as we look into
the future we see a lot more options
opportunities for evolving that
especially with domain specific
learning. Uh and then over time as well
I think we'll get to a world where
federated learning is also an option. Uh
but obviously as you think about
federated learning where you're learning
from one set of experiences and and
transcribing that to another
environment. uh you've got to also make
sure you're dealing with privacy and
security and access types of concerns as
well. So I think we're quite a ways away
from there. Uh but that's where the the
future is heading as well. Does the
model have to be taught uh or walked
through every situation by a person to
be able to handle it on its own? No,
it's Okay, you're shaking your head.
That's that's the beauty of the model,
right? It's it's it's learning. It's
it's actually we've we've got the model
to the point where uh you're actually
going through and executing certain
tasks uh and it's actually learning and
creating these summaries and creating
these scripts uh and and executing them
and and it's it's it's actually amazing
especially as we've seen updated
versions of of the the models the
underlying models themselves. Uh you can
see that not only the accuracy but also
the reliability of what you're getting
with these models is is is pretty
amazing as well. We debate often on this
channel about whether the product or the
model matters matters more. Um what has
the advancing models enabled you to do
in the product like as we get from let's
say GPT4 to 4.5 or I don't know are
using like claude like claude sonnet 3
to 3.7 what have these better models
enabled you to build? Yeah, I mean
ultimately the mo there's a couple of
things right with the models it's it's
the ability to solve uh problems
reliably. it's able to to solve new
classes of problems as well and it's
also performance as well right how you
can do it quick more quickly and there's
also the economics of of this matters as
well right ultimately you want to make
sure you can execute these tasks uh in a
manner that makes sense for the value
that you're providing so those are all
the pieces that help us but the product
is just as important as well right
because how the user interacts with the
system uh where uh models are leveraged
and you have the right sort of handoff
from human being to the model where
you've got the ability, you know, think
of it from a a user interface design
perspective. Uh for the technician to be
able to see exactly here's these virtual
technicians. Do I have control? Can I
see what's happening? How do I step in
there? Those are the product design
elements that are important as well. Uh
and those have to work in concert with
the underlying models that work as well.
Okay. Um, finally I want to ask you,
there's been this discussion in the AI
community that AI is going to hit a
wall. Maybe that means pre-training is
going to hit a wall. Um, but basically
people are saying that the models aren't
going to get much smarter and therefore
this is a bubble and you're in an
interesting position because you're both
seeing the ways that models are
advancing and you're putting this into
action from a product standpoint. What's
your view on the question of whether AI
is hitting a wall? Uh I'd be very
curious to hear whether you anticipate
that the party's almost over or whether
it's just beginning. No, from everything
we're seeing, we're still very early
stages of this whole uh transformation
that's happening, right? And it's
amazing, right? We were talking about
LLMs and we're leveraging the large
language models here, but now we're
increasingly talking about SLMs, right?
and domain specific agents and and and
if you think about it, Alex, even even
as recently as four or five months ago,
those weren't part of the the
vernacular, right? So, even in a short
period of time, we're introducing new
concepts here. And some of the things
that I talked about, it's it's it's
there. The generative aspect is amazing,
but if you're able to bring reasoning
into it, if you're able to bring
prediction, if you're able to connect
things and look at causal relationships,
we've got different types of AI
technologies that do some of those
elements. But imagine the power of tying
that together and where we can go. I I
think we're we're at very early stages
of of this overall transformation. We
learn along the way. uh they'll we we
also need to make sure there's there's
there's proper uh gates and and checks
and balances in the process as well uh
and and making sure that ultimately the
AI is is reliable that it's actually
consistent that it's working effectively
uh and and we're protecting IP as well
for customers right they've got customer
data as we deploy these solutions how do
you make sure that the data is protected
and that's not compromised so all these
things need to work together we'll go
through a process of evolution but I
really think just for the reasons that I
described, we're at very early stages of
this transformation process. Okay. So,
if people have been following along and
they want to hear more about about your
work or they want to potentially partner
with you, how do they get in touch?
Where do they go? LogMine.com, right? Go
to login.com. Uh, and you can, in fact,
some of the things that I talked to you
about, you can actually experience this
yourself, right? you can go in uh
there's there's there's trials you can
start there's videos you can start to
see that a lot of this is absolutely
there today solving value for our
customers uh and and welcome to uh reach
out to us through through logmain.com
great well Joseph thank you for coming
on thank you for sharing so much I
definitely learned a ton about how this
is going to work and for me again we
hear so much about what AI might do very
little about what it's actually doing
and for you to come on and share some
practical applications ations of the
technology that are in play today was
was really helpful and I'm sure the
audience will enjoy it as well. So,
thanks for coming on the show. Thank you
very much, Alex. It was a pleasure. All
right, everybody. Thank you to Joseph
and thanks for you all to you all for
watching. We'll be back on the feed with
another interview