[00:00:00] [music]
[00:00:04] Joining me today is Chris Savage, CTO of
[00:00:06] ALIANDO of Microsoft Front Runner
[00:00:08] partner delivering cloud data and AI
[00:00:10] transformation for enterprises in more
[00:00:12] than 30 countries. Chris has spent over
[00:00:14] 20 years in IT. He started out building
[00:00:16] data centers and went on to lead global
[00:00:18] engineering teams and has guided
[00:00:20] companies through the move from on-prem
[00:00:22] to cloud and now into AI. Our topic of
[00:00:24] discussion today is the right way to get
[00:00:27] AI into your business. Chris, welcome to
[00:00:29] TechUnhedged. Really good to have you on
[00:00:30] the show.
[00:00:31] >> Thank you, Ashar. Looking forward to our
[00:00:32] discussion today.
[00:00:33] >> So, let me start off with a very direct
[00:00:36] question. How are you using AI inside
[00:00:39] Eleandro right now and you know, what
[00:00:41] does a normal work day look like for
[00:00:42] you?
[00:00:42] >> Well, being a frontier partner, we’ve
[00:00:45] got lots of agents doing lots of things
[00:00:47] for us and and that’s no different for
[00:00:49] me. So, we started on our AI journey
[00:00:52] back in 2023 right after I joined the
[00:00:55] company and my day starts off with a
[00:00:57] summarized inbox and uh activity. So,
[00:01:01] I’ve got an agent that reviews
[00:01:02] everything that’s going on in in my
[00:01:04] inbox and my day and says, here are the
[00:01:06] things that we think we should start on
[00:01:08] and then here are the the top emails
[00:01:10] that need to be responded to and all of
[00:01:12] your tasks. It’s really makes my job a
[00:01:15] lot more efficient. Uh I don’t have to
[00:01:17] go wandering through inboxes. I’ve also
[00:01:19] got an agent that pulls because I’m over
[00:01:21] our managed services team, I’ve got an
[00:01:23] agent that pulls all of our current
[00:01:25] ticket stats and delivers those and
[00:01:28] highlights anything that might be
[00:01:30] delayed or impacting clients in a bigger
[00:01:33] way that that we’re not aware of. So,
[00:01:35] I’ve got that agent. Um our HR teams
[00:01:38] have um an automation agent. We just
[00:01:40] built that employee onboarding, employee
[00:01:43] offboarding, kind of managing
[00:01:44] permissions and accounts and things like
[00:01:46] that. We’ve got engineering agents in as
[00:01:49] part of our delivery model now. So,
[00:01:50] things like uh scripting out or helping
[00:01:53] build um Terraform for migrations and
[00:01:56] and other um delivery aspects of our
[00:01:59] business and they’re quite useful. You
[00:02:02] know, our teams are are now consuming
[00:02:04] millions and millions of tokens every
[00:02:06] month and it’s
[00:02:07] we’re we’re putting it to good use.
[00:02:09] >> Yeah, that’s good. That’s good. Well,
[00:02:10] you’re comfortable enough sharing
[00:02:12] your data and everything emails and
[00:02:14] giving it access to agents?
[00:02:16] >> Yeah, and we’ve got a really good
[00:02:18] agentic governance model that we set up
[00:02:21] for both our internal use and for
[00:02:24] clients when we build out client AI
[00:02:26] solutions. We call it
[00:02:28] we treat an agent like a an employee
[00:02:31] where it’s you get limited amount of
[00:02:33] read-only access, not a lot
[00:02:36] you know, actions or activities it’s
[00:02:38] that the agent can do independently to
[00:02:40] where we make sure that any actions
[00:02:43] especially substantial actions are
[00:02:45] controlled and limited by permissions
[00:02:48] given to different agents. That’s how we
[00:02:50] kind of mitigate any of those runaway
[00:02:52] agent problems that a lot of companies I
[00:02:55] think when they jump into it they don’t
[00:02:57] think about oh, I should be limiting
[00:02:59] what it should do. So, we’ve already
[00:03:01] built that into our platform.
[00:03:03] >> Right. Okay, that’s good. So, I’m
[00:03:04] imagining it’s somebody logs into that
[00:03:06] platform, they can see what’s coming in
[00:03:08] and then they can approve or you know,
[00:03:10] not approve or make any edits or changes
[00:03:12] and then hand it over, right? So, it’s
[00:03:14] not And also choose to say that okay,
[00:03:16] these tasks or these things are are okay
[00:03:19] to go ahead. So, you know, you don’t
[00:03:20] need to ask me or something of those
[00:03:21] lines.
[00:03:22] >> Right, exactly. Yeah,
[00:03:24] there are little things that it can do
[00:03:25] like our AI agent for HR. So, we’ve got
[00:03:30] it set up to where it can go in and
[00:03:32] create new user accounts and we’ve got
[00:03:34] very defined permissions and activities
[00:03:36] that it does and we know that when we
[00:03:38] hit the button to submit all it’s going
[00:03:40] to do is create that new user, apply the
[00:03:42] licensing and things like that. So, it’s
[00:03:45] controlled.
[00:03:46] >> That’s good. So, it’s often you know,
[00:03:48] that people used to figure out
[00:03:49] repeatable and frequency of tasks. So,
[00:03:52] task that are repeatable in a you know,
[00:03:55] in a what do you say? In a very defined
[00:03:58] manner. So, you can actually predict
[00:03:59] that this is what needs to be done and
[00:04:01] the frequency is high is where it makes
[00:04:03] the most sense to you know, kind of do
[00:04:05] that. So, onboarding is one of those use
[00:04:07] cases at your organizations.
[00:04:09] >> Oh, yeah. Absolutely.
[00:04:10] >> Okay.
[00:04:11] It’s It’s one of those easy things. It
[00:04:12] used to be a ticket, 10 different people
[00:04:14] had to do 10 different things and now
[00:04:16] we’ve got HR at the front end. Agents
[00:04:19] take care of the deployments and and
[00:04:21] assigning permissions and doing all the
[00:04:23] fun things that my IT staff used to do.
[00:04:26] And now it’s really they get the output
[00:04:29] that the agent says, “Here’s what I did.
[00:04:31] Here’s a creation.” They validate
[00:04:33] everything. So, there is a validation
[00:04:35] point that our team has to do, but it’s
[00:04:37] at the very end and it’s just a double
[00:04:38] check of Yep, everything got done the
[00:04:41] way it needed to. So,
[00:04:42] >> And there were a couple of little
[00:04:43] hiccups and we figured out that you
[00:04:45] know, some of the permissions we’d we’d
[00:04:47] restricted it too much. So, we had to
[00:04:49] expand it just a tiny bit to give it to
[00:04:51] the right.
[00:04:51] >> That’s always good, right? Start with as
[00:04:53] tight as possible and then just once you
[00:04:55] figure out that hey, you know, what
[00:04:56] changes you need to make then you make
[00:04:58] those small adjustments. So, building on
[00:05:00] the this, there’s a lot of AI hype as
[00:05:02] well, right? So, you know, people and AI
[00:05:04] and everybody just started off saying
[00:05:06] replacement, jobs lost, etc. Job
[00:05:09] eclipse, apparently that’s the word
[00:05:11] that’s going around. How do you filter
[00:05:13] out between the AI hype and between the
[00:05:17] spots where you know, AI can really
[00:05:19] create some good enterprise value.
[00:05:21] >> Well, we start off with when we do we we
[00:05:23] do a lot of advisory work on AI and and
[00:05:25] then go into actual implementation. And
[00:05:28] what we always start with is let’s look
[00:05:30] at some key business processes and then
[00:05:33] we have break that down into individual
[00:05:37] tasks. And then we look at okay, so this
[00:05:40] business process is supposed to drive
[00:05:42] this either cost savings or you you
[00:05:45] revenue increase. So, as long as we’re
[00:05:47] looking at things in with the lens of a
[00:05:50] financial aspect and a financial goal of
[00:05:53] what we’re going to implement with AI, I
[00:05:55] think that’s where the hype leaves and
[00:05:58] you start creating that real value. You
[00:05:59] need to make sure that you’ve got the
[00:06:01] right governance in place as as you move
[00:06:03] forward with any kind of AI
[00:06:05] implementation, but to me it’s what’s
[00:06:08] the business driver? What’s What’s the
[00:06:10] one little tiny thing that we can do
[00:06:13] that’s, you know, costing a human 3
[00:06:15] hours a day that an agent can go do in
[00:06:18] in just a couple of hours or less,
[00:06:20] right? Agents work very very fast and
[00:06:22] sometimes that speed is what gets you in
[00:06:24] trouble.
[00:06:25] >> Yeah, [laughter] that’s becomes always
[00:06:27] that is, right? So, I say to my team
[00:06:29] that you have a spoiled brat who is at
[00:06:31] your command in terms of AI, but that
[00:06:33] kid is like maybe 20x smarter than you.
[00:06:36] You just need to learn how to rein it
[00:06:37] in, right? So, it’s like
[00:06:40] >> [laughter]
[00:06:41] >> And yeah,
[00:06:43] that’s the way it is. So, one of the
[00:06:44] things I mean, you mentioned that you
[00:06:46] guys do a lot of advisory work. So, I’m
[00:06:48] guessing you’d start with, you know, AI
[00:06:50] workshops and and kind of going deeper
[00:06:52] down into organizations and
[00:06:54] understanding I mean, I’ve always said
[00:06:56] it’s always starting with the problem.
[00:06:58] It’s not the technology, right? So, it’s
[00:07:00] really understanding what kind of issues
[00:07:03] are you facing and you know, kind of
[00:07:04] talking of that sort out. So, how do you
[00:07:06] guys, you know, manage that? Do that?
[00:07:08] What are the right questions at times
[00:07:09] that, you know, helps leaders also
[00:07:11] identify? Because unless they tell you,
[00:07:14] I mean, it’s hard to, you know, figure
[00:07:16] out as well. So, what are the those kind
[00:07:18] of questions that at times you guys ask
[00:07:20] that, you know, help people help leaders
[00:07:21] also pick the right problems for AI to
[00:07:23] solve.
[00:07:24] >> Yeah, so it starts with the work. The
[00:07:26] work that people do. We sit down and we
[00:07:28] start looking at mapping where time,
[00:07:31] money, and errors might happen within a
[00:07:34] given process or given task. Because
[00:07:36] what we want to focus on is, again going
[00:07:38] back to to the previous question, how do
[00:07:40] we add that enterprise value? So, we
[00:07:42] start with the value stream. We map out
[00:07:45] here’s the task that that the client
[00:07:47] has. Here is the individual, you know,
[00:07:50] elements of those tasks. We walk them
[00:07:52] end to end through those processes.
[00:07:54] Where does a human do something and what
[00:07:56] does that involve? Then we start
[00:07:58] quantifying exactly what each of those
[00:08:02] sort of decision points might cost or
[00:08:05] the overall task is going to cost. Once
[00:08:07] we have that down, then we can start
[00:08:10] classifying it and because sometimes a
[00:08:12] human really needs to be the one doing
[00:08:14] the work, right? It doesn’t
[00:08:16] doesn’t make sense. So, you know, we
[00:08:18] look at each individual task and say,
[00:08:20] “Okay, can does a human really need to
[00:08:22] do that?” Or can we have an agent go and
[00:08:25] either make that decision based on some
[00:08:27] criteria or here’s what the agent can
[00:08:29] generate. Then we calculate out, “Okay,
[00:08:32] so an agent’s going to be able to do
[00:08:34] this, this, and this and it’s going to
[00:08:35] take x amount of hours for us to go
[00:08:37] develop that.” So, we want to provide an
[00:08:40] ROI calculation as we think through what
[00:08:43] the solution is going to be for the
[00:08:44] client. And then, you know, out of
[00:08:46] everything, we always start small and
[00:08:48] then get big. Pick one specific agent
[00:08:52] task because we’re going to kind of
[00:08:53] demonstrate to the value of of what
[00:08:55] we’re doing to the client. So, we we do
[00:08:57] proof of value work where we start with
[00:09:00] a small small implementation based on
[00:09:02] the overall mapping of those tasks or
[00:09:04] the business process. And once we have
[00:09:07] that specific kind of starting point
[00:09:09] narrowed in, we go develop it, we let it
[00:09:11] run, we let the client use it, get to
[00:09:13] get used to working with it, too,
[00:09:15] because there’s a as you implement these
[00:09:17] things, there’s a a level of change
[00:09:19] management that you have to go through.
[00:09:21] So,
[00:09:22] we work with the client. And so, you
[00:09:24] know, we let them kind of do that and
[00:09:26] then we come back and say, “Okay, how
[00:09:27] did how’s it working out?” And then that
[00:09:29] allow then we can move into, you know,
[00:09:31] greater phases or different
[00:09:34] taking on different tasks now at that
[00:09:35] point.
[00:09:36] >> Okay, So, that’s more on understanding
[00:09:38] the processes, identifying the use
[00:09:41] cases, figuring out the ROI. The ROI is
[00:09:44] looked upon from two factors. One, hour
[00:09:47] saving or giving people time back, or B,
[00:09:50] directly increasing the revenue. So,
[00:09:51] that’s that’s basically, you know, the
[00:09:53] top and the bottom line. Other than
[00:09:54] that, so once let’s say that’s figured
[00:09:56] out, what kind of what does AI readiness
[00:09:58] look for for a business, right? From a
[00:10:00] perspective of you talked about change
[00:10:02] management, but then there’s data as
[00:10:03] well, and then there’s access to data as
[00:10:05] well. How good is the data? How bad is
[00:10:07] the data? So, what does that sort of
[00:10:09] kind of look like?
[00:10:10] >> So, we have to sort of take a look at
[00:10:13] what information is an agent going to
[00:10:15] need, and where does it reside? Full
[00:10:17] sort of mapping of knowledge within the
[00:10:20] company. So, we’ve got databases. Where
[00:10:22] is Where what databases does the agent
[00:10:25] need access to? Or what body of
[00:10:29] knowledge, what document? Data
[00:10:31] classification is going to be super
[00:10:33] important for organizations as they
[00:10:35] leverage unstructured data, documents,
[00:10:37] presentations, all of these things.
[00:10:39] Because that’s where lots of agents
[00:10:42] expose information to end users, where
[00:10:47] we didn’t think that was available to
[00:10:49] anyone.
[00:10:50] >> Right. Right.
[00:10:51] >> The agent had access because the user
[00:10:53] had access, and the user has access. No
[00:10:55] one knew the user had access. So, we’ve
[00:10:57] run into that a couple [laughter] of
[00:10:58] times where it’s like, “Oh, everyone’s
[00:11:00] salary was in the spreadsheet, and
[00:11:02] someone in HR just put it up and shared
[00:11:04] it, not thinking anyone would actually
[00:11:06] see it.” Those are the kinds of things.
[00:11:08] >> Interesting.
[00:11:09] >> Get through a data classification, a
[00:11:11] permissions hygiene activities. And then
[00:11:14] we look at data quality. So, yes, we’ve
[00:11:16] got this great database, but no one’s
[00:11:19] really touched it, and you know, like we
[00:11:21] we’ve had an ERP an agent that was
[00:11:23] connecting to an ERP system, and the ERP
[00:11:26] system had data that had never been
[00:11:28] cleaned up. And so, it was surfacing
[00:11:31] weird things, and the client couldn’t
[00:11:32] understand why and so we started talking
[00:11:34] to them. If you say this is highlighting
[00:11:36] things that are really old, how long has
[00:11:38] it been since you kind of cleaned up
[00:11:40] your data in the CRP system? They said,
[00:11:42] “Oh, we’ve never touched the data. It’s
[00:11:44] just we add more, we add more, we add
[00:11:45] more, we add more.” And it’s like, “Oh,
[00:11:46] well, that’s why we’ve got a problem
[00:11:48] because it’s what we’re asking it to do
[00:11:50] is go find data and all of the data that
[00:11:53] you’ve got here is not very clean. So,
[00:11:55] you got to we got to clean it up.” So,
[00:11:56] there was exercises in data hygiene that
[00:11:59] we had to make sure that you’ve got the
[00:12:00] right identity management system in
[00:12:03] place. You actually need that control.
[00:12:05] An agent should be treated like an
[00:12:06] employee, so you’ve got to make sure
[00:12:08] that as you craft what an agent is
[00:12:11] doing, it’s got the right permissions at
[00:12:13] the right levels and you’ve got the
[00:12:14] right the right control plan for that.
[00:12:16] And then you need to identify who’s
[00:12:18] owning this agent. To your point
[00:12:20] earlier, you know, it’s like a little
[00:12:22] kid. I always call agents interns.
[00:12:24] An agent is like just a fresh college
[00:12:27] graduate intern. They don’t know a whole
[00:12:28] lot. They know enough of what you’re
[00:12:30] telling him to go do and that’s all they
[00:12:31] know. But, someone needs to own the
[00:12:34] agent. Someone needs to to own that
[00:12:36] employee, the virtual employee. And so
[00:12:38] we we talk about that. And at the
[00:12:40] highest level, you’ve got to have
[00:12:42] commitment from the top level of the
[00:12:44] organization that we’re going to run
[00:12:46] agents. We’re going to run AI in a
[00:12:49] manner that benefits our employees and
[00:12:52] our company. We don’t have a whole lot
[00:12:54] of disruption and angst about people
[00:12:56] losing jobs or anything else. At least
[00:12:59] to to clients and really emphasize the
[00:13:01] fact that what we should be doing with
[00:13:02] AI is improving productivity, not
[00:13:05] eliminating productivity.
[00:13:07] >> Do you see, I know we talked about other
[00:13:10] saved revenue increased, but do you see
[00:13:13] quality improvement in the work quality
[00:13:15] that gets out? Is that a factor to
[00:13:17] consider as well? That the quality of
[00:13:19] work automatically starts improving.
[00:13:21] >> Yeah, yeah, I would say in some cases
[00:13:23] it’s
[00:13:23] >> In some cases where there’s subjective
[00:13:25] work, not definite. I mean, if you’re
[00:13:26] onboarding something, then you know that
[00:13:28] hey the email address needs to be
[00:13:29] created XYZ needs to be done. That’s not
[00:13:31] somewhere you you can have quality but
[00:13:32] let’s say assume there are cases where
[00:13:35] recruitment team is sorting through
[00:13:37] multiple CVs or you know coming in and
[00:13:40] 1000 CVs came in because it’s a human
[00:13:41] working the she or he was able to do
[00:13:43] let’s say you know find 50 good CVs and
[00:13:46] all of sudden you the agent finds 100
[00:13:48] good CVs because you know it can read it
[00:13:50] at a consistent pace and everything. So
[00:13:52] is that something that has kind of come
[00:13:54] up?
[00:13:54] >> Yeah, for sure, right? So
[00:13:57] and and your CV is a good example
[00:13:58] because we we’ve actually helped agents
[00:14:00] that analyze different kinds of
[00:14:02] documents and and obviously speed which
[00:14:05] can do is is incredible but
[00:14:07] >> consistency it can do with, right? So
[00:14:09] human gets tired but the agent never
[00:14:11] gets really tired.
[00:14:12] >> Well, and you know you give the agent
[00:14:15] here is what we’re looking for. Here’s
[00:14:17] your defined set of criteria then to
[00:14:20] your point it’s going to be very
[00:14:21] consistent in the results that it
[00:14:22] delivers back to you.
[00:14:24] >> Right. Right.
[00:14:25] >> So you know there’s definitely
[00:14:27] improvement in consistency across the
[00:14:29] board when you start getting those those
[00:14:31] kinds of data feedback that that the
[00:14:33] agent is delivering back to you.
[00:14:35] >> Since we’re talking about this I know I
[00:14:37] kind of
[00:14:38] wanted to talk about later but I think
[00:14:39] this is the right time. What I’ve also
[00:14:41] noticed is so most of these use cases
[00:14:44] often are just you know workflow
[00:14:47] automation kind of use cases do this do
[00:14:49] that and etc. And we were we’ve been
[00:14:50] doing those since I don’t know when,
[00:14:52] right? So a lot of people feel that for
[00:14:54] every scenario an agent is required when
[00:14:57] it’s not when it’s just a sequence of
[00:14:58] steps that need to be followed. So I’m
[00:15:01] guessing when you guys are you know
[00:15:02] talking to owners or or owners of those
[00:15:05] agents or anything it has to be laid out
[00:15:07] that hey this is where the decision
[00:15:08] making has to happen the rest is just
[00:15:10] you know good old workflow automation.
[00:15:12] >> And yeah and and that identifying that
[00:15:14] too at the very beginning is important,
[00:15:16] right? Because
[00:15:17] again it’s like okay so we can we mapped
[00:15:20] our process we understand who’s doing
[00:15:22] what and how it’s working, but you know,
[00:15:24] it it’s always a a tricky sort of thing
[00:15:27] when you’re talking to a client that,
[00:15:29] you know, they want to do everything in
[00:15:30] AI. It’s like, wait,
[00:15:32] there’s still humans that need to do
[00:15:34] some things and walking them back a
[00:15:36] little bit from just AI everywhere, AI
[00:15:39] all the time, AI everything because it
[00:15:41] doesn’t
[00:15:42] it just doesn’t make business sense to
[00:15:44] have it like
[00:15:44] >> Did you feel the need to redesign a
[00:15:47] certain workflows around the agents so
[00:15:50] that it’s just not, you know, the AI on
[00:15:52] top or a bolt-on approach versus being a
[00:15:55] proper process-oriented approach?
[00:15:57] >> always look at your processes and go,
[00:15:59] wait, this process doesn’t even make
[00:16:01] sense. Is this something, you know, we
[00:16:03] run through in Aliado yearly exercises
[00:16:05] on, you know, do we continue doing this?
[00:16:07] Do we stop doing this? Or do we make
[00:16:09] changes to doing whatever it is, you
[00:16:11] know, whatever process it is. So,
[00:16:13] there’s always that sort of process
[00:16:14] improvement exercise a business should
[00:16:17] go through before even getting to an
[00:16:19] agent because, you know, there’s a lot
[00:16:21] of things that, this is the way we’ve
[00:16:23] done it all the time kind of factors
[00:16:25] within business that doesn’t necessarily
[00:16:28] get solved by throwing a bunch of AI on
[00:16:30] top of things that are just inefficient
[00:16:32] anyway.
[00:16:33] >> If a company is, you know, kind of
[00:16:35] looking to implement AI wants to you
[00:16:38] know, put it into a process and it’s
[00:16:40] want to cut the time into half, let’s
[00:16:43] say, you know, you you’re thinking that
[00:16:44] it’s going to take weeks or something
[00:16:46] months, where can it actually start off
[00:16:48] from? What’s the maybe the easiest way
[00:16:50] to do it? Or
[00:16:51] >> For me, the the starting point for us,
[00:16:53] and I’ll I’ll kind of use Aliado as
[00:16:55] client zero, was just using kind of
[00:16:58] standard off-the-shelf chatbots, right?
[00:17:00] So, Microsoft Copilot, Claude, ChatGPT.
[00:17:04] Getting used to understanding and just
[00:17:07] getting information retrieval, I think
[00:17:09] is a good way to kind of get started
[00:17:11] with AI. That’s how we started.
[00:17:13] Microsoft Copilot was all of our first
[00:17:16] interaction with some of the chat. Some
[00:17:18] of us did chat GPT early on. And so, I
[00:17:21] think getting started that way makes it
[00:17:24] a little safer. You get to kind of see,
[00:17:26] “Oh, wow, this is really cool. Look at
[00:17:27] all the things it can kind of do. Now,
[00:17:29] can I have it do this thing cuz I hate
[00:17:31] doing it every day? And how do I get
[00:17:32] this now moved into a sort of an agent
[00:17:35] that can just kind of do these things
[00:17:37] over and over like the the agent I use
[00:17:40] every day to start my day, which is
[00:17:42] summarize my activities for the day and
[00:17:44] tell me what to prepare for.” So, that
[00:17:47] then gets into, you know, taking it from
[00:17:49] the chatbot to, “Okay, you’re going to
[00:17:51] help me organize my life.” Then,
[00:17:53] building into Now, now we’re going a
[00:17:55] little deeper into, “Oh, I’ll Now, I can
[00:17:58] have an agent just do this task for me
[00:18:00] because, you know, it’s it’s something
[00:18:02] I’ve got to do every day, but I don’t
[00:18:04] really need to be the one pressing all
[00:18:06] the buttons and moving the mouse and
[00:18:07] clicking the mouse.
[00:18:09] >> So, the big question I think a lot of
[00:18:11] organizations have nowadays is on prem
[00:18:14] versus cloud regarding
[00:18:16] the usage and everything. What are your
[00:18:17] thoughts on that?
[00:18:18] >> I actually talked to a client yesterday
[00:18:20] who was
[00:18:21] >> [laughter]
[00:18:21] >> Well, who built out a very nice AI data
[00:18:25] center. They’ve invested a lot of money
[00:18:27] in it. And they’re running out of
[00:18:28] capacity. Imagine that. You’ve got a lot
[00:18:31] of agents doing lots of things. They’re
[00:18:33] running out of capacity. So, they came
[00:18:34] to us and said, “Listen, we need kind of
[00:18:36] a AI capacity failover, if you will,
[00:18:40] inside of a cloud provider.” And so, you
[00:18:44] know, I walked them through, “Well,
[00:18:46] here’s here’s a couple different options
[00:18:47] for you. I think where you should be
[00:18:49] going is you need to build a front end
[00:18:52] to your on prem agents that looks at,
[00:18:56] “Okay, how busy are things?” And, “Oh,
[00:18:59] this server’s too busy. This server’s
[00:19:01] too busy. Okay, we’re going to punt this
[00:19:02] to the cloud.” And I think you can you
[00:19:04] can start I think there’ll be, you know,
[00:19:06] more of that on prem where you got sort
[00:19:09] of a an AI director, right? A a traffic
[00:19:12] you know, no different than a load
[00:19:14] balancer from way back when, right?
[00:19:16] >> That’s a That’s a good analogy,
[00:19:18] actually, yeah. Or any for air traffic
[00:19:20] controller or load balancer. Load
[00:19:21] balancer is more suitable in our world.
[00:19:23] But, yeah. I mean, determining whether
[00:19:25] where to route this request, absolutely.
[00:19:27] That’s good.
[00:19:29] That’s a good example.
[00:19:30] >> I have my thoughts in the sense that
[00:19:32] since it’s so rapidly evolving, this
[00:19:35] technology is growing so crazily at a a
[00:19:38] speed that, you know, you wouldn’t even
[00:19:40] imagine a decade back, right? So,
[00:19:42] usually software getting upgraded at
[00:19:44] this pace used to take 5, 10, 15, you
[00:19:47] know, you could talk in decades. 10
[00:19:49] years, then another 10 years, and then
[00:19:50] so on. Right now, it’s it’s like 6
[00:19:52] months. Which is crazy, right? So, you
[00:19:55] invest so much money in hardware, in I
[00:19:57] mean, in in setting up those data
[00:19:58] centers, and all of a sudden, 1 year
[00:20:00] down the road, you have much cheaper
[00:20:01] hardware because hardware is
[00:20:02] accelerating at the same pace. This
[00:20:04] cycle of AI fixing itself, improving
[00:20:06] itself is, you know, works at a pace
[00:20:09] which is not known to us before. So, I
[00:20:11] mean, I have clients they started with
[00:20:13] on-prem. They had these doubts. I don’t
[00:20:14] want to put my data into Anthropic, or I
[00:20:17] don’t want to put my data into Open AI.
[00:20:18] Although, I have a I’ve always thought
[00:20:20] that you already have your data in AWS,
[00:20:23] Azure. How different it is from keeping
[00:20:25] your data there to keeping your data
[00:20:26] here. The only difference being that
[00:20:28] maybe these companies are new. They’re
[00:20:30] not as old as AWS, or the fact that it
[00:20:32] might use your data to train. But, you
[00:20:34] know, AWS uses your data to train as
[00:20:36] well, somewhere.
[00:20:37] >> [laughter]
[00:20:37] >> But, so, you know, and the other thing
[00:20:40] was that they made the center, and now
[00:20:43] they wanted to have the new H100, and
[00:20:45] then the other models, and then the
[00:20:46] other cards. But, it’s a never-ending
[00:20:48] game. I feel that, you know, unless and
[00:20:49] until the cost or the ROI is really
[00:20:51] justified, and your data is really,
[00:20:53] really, really so critical, I mean,
[00:20:55] think law firms or legal companies or
[00:20:57] banks or institutions like those, you
[00:20:59] might not really need to go to on-prem.
[00:21:01] >> No, I really don’t. I
[00:21:03] I personally, and and I’ve advised
[00:21:05] clients here. I’ve had clients ask me,
[00:21:07] “Well, should we just go ahead and build
[00:21:08] this out now?” Don’t don’t make that
[00:21:10] investment.
[00:21:11] >> Yes.
[00:21:11] >> You know, the the cloud is, I think,
[00:21:14] even is more secure than than some
[00:21:17] people’s data centers. I mean, I’ve been
[00:21:18] in this business long enough to know
[00:21:20] that uh yeah, here’s here’s the key card
[00:21:22] hanging right by the door because nobody
[00:21:24] wants to to remember to bring their key
[00:21:26] card. So, they just swipe and go into
[00:21:28] the data center, right? So, you know,
[00:21:30] just standard physical control sometimes
[00:21:33] lack in a a standard business data
[00:21:35] center. But, I think the the way all of
[00:21:38] the cloud providers, you know, GCP,
[00:21:41] Microsoft, AWS, all of what they’ve
[00:21:44] done, there’s a lot of security, there’s
[00:21:45] a lot of data security. All three
[00:21:47] providers have the same similar cloud
[00:21:50] responsibility model where you are
[00:21:52] responsible for securing access into
[00:21:54] this data center, right? Keep that that
[00:21:57] identity moat secured. So, taking out I
[00:22:00] think that that you can take out the
[00:22:02] data security argument on needing to
[00:22:05] build on prem. I mean, there’s certainly
[00:22:07] certain businesses, highly regulated
[00:22:09] industries, government entities that
[00:22:12] need that really high layer of security,
[00:22:15] right? They don’t want anything even
[00:22:17] connected to the internet, so they build
[00:22:18] out their own data centers and leave it
[00:22:20] like that. I Once you eliminate like the
[00:22:23] overarching security need, I I don’t
[00:22:25] have any
[00:22:26] >> There’s no other
[00:22:27] reason a strong strong enough reason to
[00:22:30] go that route.
[00:22:31] >> Yeah, because all of the cloud providers
[00:22:32] have great model access, all of them
[00:22:35] have the ability to scale quickly, too.
[00:22:38] You know, if you’ve got a big project
[00:22:40] coming on and you know you’re going to
[00:22:42] need to build, you know, 10 or 15 new
[00:22:44] agents because you’re you got new
[00:22:46] product release or something. Now, it’s
[00:22:48] click of a button to go get that
[00:22:49] configured and
[00:22:51] all of the constraints and all high cost
[00:22:54] of hardware, it’s much easier just to
[00:22:55] use the cloud.
[00:22:56] >> What metrics should a a CTO can put in
[00:22:59] front of the CFO or the board CEO to
[00:23:02] prove, you know, that an AI system is
[00:23:04] worth it what it cost. Because right now
[00:23:06] it’s more of a IT driven I feel it’s
[00:23:09] more of technology driven and IT driven
[00:23:11] initiative versus business owners, you
[00:23:14] know, driven initiative. So, and how do
[00:23:15] you keep those costs from spiraling once
[00:23:17] it’s in production?
[00:23:18] >> Yeah. Well, I would actually my
[00:23:20] experience has been we’re we’re seeing
[00:23:21] more business Business owners. Business
[00:23:24] side of the business coming to us and
[00:23:26] saying, “I want this AI.” Because I’ve
[00:23:30] used chat GPT enough to know that I can
[00:23:32] get it to automate a few things, but now
[00:23:34] I want to go bigger with it. You know,
[00:23:36] I’ve actually seen more reluctance by
[00:23:38] internal IT teams on building out AI
[00:23:41] versus just the business
[00:23:43] >> That’s true, actually.
[00:23:44] >> Yeah, and saying, “This is a process. I
[00:23:47] know I can use an agent to do some
[00:23:49] things cuz I’ve already got the
[00:23:50] experience.” So, I think a little bit of
[00:23:53] shadow IT has influenced significantly
[00:23:56] how we adopt. When you look at what’s
[00:23:59] the ROI on a project, it really boils
[00:24:03] down to how much additional savings are
[00:24:05] we doing because I don’t need another
[00:24:08] human being to do these things. So, in
[00:24:10] our organization, we’re leveraging AI to
[00:24:13] help triage support tickets
[00:24:16] >> And answer
[00:24:17] >> support tickets. What that’s allowed me
[00:24:18] to do is we take on a lot more clients,
[00:24:21] but I don’t necessarily have to head
[00:24:23] count. So, you know, we’re doing twice
[00:24:25] the volume with the same number of
[00:24:26] people we did 2 years ago. That just
[00:24:29] having that kind of efficiency makes the
[00:24:32] ROI really visible.
[00:24:35] >> That’s the best case scenario, right?
[00:24:36] So,
[00:24:37] that’s like the ideal case scenario
[00:24:39] where you’re dealing
[00:24:41] That’s the easy one. 100 support
[00:24:43] tickets, two people, I’m dealing 2,000,
[00:24:45] two people. So, that’s like the perfect
[00:24:47] world scenario for anybody. Right?
[00:24:49] Because otherwise it it times becomes
[00:24:52] hard. I mean, you give a few hours back,
[00:24:54] absolutely. What are those hours now
[00:24:56] giving to the organization, right? So,
[00:24:58] let’s say if I make a few people a few
[00:25:00] department, you know, I give them a day
[00:25:02] an hour of the day back. Great. Awesome.
[00:25:05] Good for the person, but how does it,
[00:25:06] you know, cumulatively affect the
[00:25:08] organization? Are they adding in more
[00:25:10] value in terms of other things or stuff
[00:25:13] like that?
[00:25:13] >> Well, it’s always a matter, you know,
[00:25:15] it’s going to be a a bit of a balance,
[00:25:17] but it certainly you’ve got to tie
[00:25:19] productivity gains to dollar amounts. At
[00:25:21] the end of the day, a business shouldn’t
[00:25:23] be doing anything if it’s not going to
[00:25:24] drive any kind of return on on that
[00:25:26] investment. So, understanding how much
[00:25:29] all of your processes might cost you in
[00:25:31] a hard dollar figure.
[00:25:33] >> Right.
[00:25:33] >> we have this person, we pay them x
[00:25:35] amount of dollars per hour, we’re saving
[00:25:37] them 4 hours a week, and now those
[00:25:39] activities are turning into I’ve got
[00:25:42] sales people that now can reach out, do
[00:25:44] more sales, right? And so, we can tie,
[00:25:47] you know, what an agent is doing to
[00:25:49] facilitate sales to increase sales. And
[00:25:51] so, you can see some of those, you know,
[00:25:53] increases in hard dollar increases
[00:25:55] because they’re not spending a whole
[00:25:57] bunch of time, you know, drafting new
[00:25:58] contracts or doing whatever.
[00:26:00] >> They’re actually able to do more sales.
[00:26:01] Has it come up where the agents have
[00:26:03] spent more than the hours that we were
[00:26:04] saving of the people? Have you come
[00:26:06] across that so that or not yet?
[00:26:08] >> Not yet. Not yet.
[00:26:09] >> [clears throat]
[00:26:09] >> I’m sure that’s going to happen given
[00:26:11] just the AI kind of everywhere mentality
[00:26:14] on certain things.
[00:26:15] >> Right. Right.
[00:26:16] >> No, no, I was I was going to say I think
[00:26:17] next year what we’ll we’ll see a lot of
[00:26:20] is we built this agent, it’s costing us
[00:26:22] a fortune now. Help us fix that
[00:26:25] >> Correct.
[00:26:26] >> Help us fix that problem.
[00:26:27] >> And that could primarily become because
[00:26:30] people think throwing AI at every
[00:26:33] problem is the solution, right? So,
[00:26:34] versus really thinking this is the
[00:26:36] bigger workflow. I have to go from A B
[00:26:38] C, let’s say to F. Now, A to C can be
[00:26:41] just good old scripting, manual workflow
[00:26:44] automation, no agent involved. Maybe F
[00:26:46] is the point or, you know, E is the
[00:26:48] point where an agent really has to come
[00:26:50] in. So, when you go into these bespoke
[00:26:52] tools, they kind of put AI at every
[00:26:54] stage, which eventually increases cost.
[00:26:57] I think with the a lot of teams when
[00:26:59] they hear AI, they worry about jobs and
[00:27:01] fear and my job is going away and
[00:27:03] everything. How do you leaders can, you
[00:27:06] know, potentially address that fear, get
[00:27:08] people to adopt the tools? Because
[00:27:09] that’s the biggest factor that I’ve seen
[00:27:11] where, you know, if it’s number nobody
[00:27:13] adopts the tool, then it’s not going to
[00:27:15] happen. And and we’ve seen that in
[00:27:16] digital transformation initiatives as
[00:27:18] well 30 years back, right? So, same kind
[00:27:20] of change management and understanding
[00:27:22] that change is the only constant and,
[00:27:24] you know, those kind of things. So, how
[00:27:25] do you How do you leaders do you think
[00:27:27] can address those things?
[00:27:28] >> You have to be honest about what’s
[00:27:30] changing. You know, given the right
[00:27:32] amount of runway and the right
[00:27:34] messaging, people can handle change.
[00:27:36] What they don’t like is implementations
[00:27:39] that don’t address what they’re thinking
[00:27:42] or how they’re feeling. And we as
[00:27:44] business leaders have to acknowledge
[00:27:46] that this is going to be a big change
[00:27:48] for some people. And we have to manage
[00:27:50] that change at the very beginning. Hey,
[00:27:53] we’re going to go through an AI
[00:27:54] transformation project. We want to make
[00:27:56] your job and your life better. As a
[00:27:58] leader, my philosophy as a leader is I’m
[00:28:01] a servant leader and my job is to make
[00:28:03] your job easier and better. I always
[00:28:05] tell anyone that that hires on at at
[00:28:08] Aliado that works under me, that’s one
[00:28:10] of the first things I tell them. So, you
[00:28:11] identify something that is going to make
[00:28:13] your job easier, well, bring it to me
[00:28:15] and let’s talk about how we can do that.
[00:28:17] So, I think if you start with, first of
[00:28:19] all, that that leadership premise at the
[00:28:21] very beginning of of your tenure at any
[00:28:23] company, but then as as you go through
[00:28:25] change management, people know that what
[00:28:28] you’re trying to do is really help them
[00:28:29] with their job. I think is is absolutely
[00:28:31] critical to making these things
[00:28:33] successful. If people don’t see the
[00:28:35] benefit for themselves, they’re going to
[00:28:37] ignore it, they’re going to reject it,
[00:28:39] they’re just not going to embrace what
[00:28:40] you’re wanting them to embrace, even if
[00:28:42] it’s sort of mandated. So, the way you
[00:28:45] have to overcome that is talk about it
[00:28:47] early, talk about it often, and talk
[00:28:49] about task and not job. You do this task
[00:28:51] every day. Do you like clicking all
[00:28:53] these buttons every day? No, you don’t.
[00:28:54] Let’s Let’s be honest. Let’s just have
[00:28:56] an agent do that, and then you can move
[00:28:58] on to other tasks that we we know you’re
[00:29:00] good at and we still value your
[00:29:02] contribution. And then showing where the
[00:29:04] saved time went, showing individual
[00:29:07] workers, “Hey, we’re Look at how much
[00:29:09] time you’ve saved now and what your your
[00:29:12] other productivity is going to look
[00:29:13] like.”
[00:29:14] >> it’s a
[00:29:14] >> Right. Those are Those to me are the
[00:29:16] best ways to manage some of that change
[00:29:18] management.
[00:29:19] >> only question. What’s one, you know, or
[00:29:21] maybe two lessons from your 20-plus
[00:29:23] experience in years in tech that you
[00:29:26] wish every CTO kind of understood before
[00:29:29] starting their AI journey?
[00:29:30] >> The patterns are always the same, right?
[00:29:32] >> Right.
[00:29:33] >> We’re going to throw All right, let’s
[00:29:35] let’s jump into
[00:29:37] big data centers. Okay, great. Now we’ve
[00:29:39] got this big data center, a lot of
[00:29:41] investment. Who’s managing it? Who’s
[00:29:43] controlling the cost? Who’s looking at
[00:29:45] the cost? Then we said, “All right,
[00:29:46] let’s get out of the data center
[00:29:47] business. Let’s move to the cloud.” So
[00:29:49] we move to the cloud. Now we’ve got
[00:29:51] skyrocketing cloud bills because no one
[00:29:53] really understood the cost and the
[00:29:55] reason to move to the cloud. You know,
[00:29:57] the cloud can be a great way to run a
[00:30:00] data center, but it’s can be the most
[00:30:01] expensive way if you’re not rethinking
[00:30:04] how you’re operating. And I think that’s
[00:30:06] the key here. It’s changing your
[00:30:08] mentality on what is possible and
[00:30:12] understanding and then owning what what
[00:30:14] that possibility is. What’s the
[00:30:16] possibility of the cloud faster
[00:30:18] applications? What’s the possibility of
[00:30:20] AI faster outcomes for business? And
[00:30:23] it’s really is.
[00:30:23] >> It’s been a real pleasure talking to you
[00:30:25] and having you on Tech Unhinged. I
[00:30:27] really appreciate your time, your
[00:30:28] candor, and the you know, openness with
[00:30:30] which you have discussed the questions.
[00:30:33] I really enjoyed the conversation. I
[00:30:34] hope you did, too.
[00:30:35] >> I did, Ashish. Thank you very much.
[00:30:39] >> [music]
[00:30:43] [music]
