[00:00:00] [music]
[00:00:04] Welcome back to another episode of Tech
[00:00:06] Unhinch where tech gets human powered by
[00:00:07] code district. I’m your host Rabia Jabid
[00:00:09] and my guest today is Rob Teal, a
[00:00:11] technology leader whose career runs from
[00:00:13] US Air Force to sea suit of a global
[00:00:16] tech company to the top of state
[00:00:18] government. Most recently, he served as
[00:00:20] CTO for the state of Oklahoma where he
[00:00:22] led the effort to bring AI into public
[00:00:24] services for millions of citizens. He’s
[00:00:26] also the author of Lansac leadership and
[00:00:29] today he’s field CEO at Giga. Today we
[00:00:31] are talking about how the public sector
[00:00:34] can actually adopt AI in a way that
[00:00:36] works and what most people get wrong
[00:00:38] about it. Welcome to the show Rob.
[00:00:40] >> Thank you so much. I really appreciate
[00:00:41] it.
[00:00:42] >> What’s the single biggest reason AI
[00:00:44] stalls in government? The technology,
[00:00:46] the budget, the culture or the politics?
[00:00:49] >> I would say the biggest reason it stalls
[00:00:50] is data governance. One of the big
[00:00:52] challenges that we have in government is
[00:00:55] unlike business, we have scope creep
[00:00:58] when it comes to allowing data pile up
[00:01:00] over years and years and years, a
[00:01:02] megabyte to a gigabyte to a terabyte to
[00:01:05] a pabyte. And so when you start turning
[00:01:07] AI tools on, one of the big challenges
[00:01:10] is well, what data do we want to give it
[00:01:11] access to? And this seems like a pretty
[00:01:13] simple thing to stall for, but in
[00:01:15] government it is not easy, particularly
[00:01:17] because of all the regulatory compliance
[00:01:19] issues and challenges we have. And so
[00:01:21] that’s probably the biggest stall is the
[00:01:23] data governance component. Making sure
[00:01:25] that when we do turn on the tools that
[00:01:27] they’re restricted to only the data
[00:01:29] we’re supposed to give them. Now some of
[00:01:30] the smaller use cases are easy like a
[00:01:32] chatbot. I mean you could turn that on.
[00:01:34] That’s pretty simple. But when it comes
[00:01:35] into the more complex use cases really
[00:01:38] data governance is a big roadblock. And
[00:01:40] in other times if we have the fruit is
[00:01:42] lowhanging enough we’ll have a use case
[00:01:44] that’s so easy to implement. It doesn’t
[00:01:46] have a lot of governance challenges to
[00:01:48] it necessarily because of what it’s
[00:01:50] doing and we’re able to implement it
[00:01:52] quickly. But overall with government,
[00:01:54] the data governance is the major
[00:01:56] roadblock for implementing AI. You got
[00:01:58] to actually solve that problem before
[00:02:00] you can start thinking about how you
[00:02:01] want to roll AI across your your
[00:02:03] organizations and your government.
[00:02:04] >> Before you could even attempt AI, what
[00:02:07] did you have to fix first in your data
[00:02:09] or legacy systems in the exposure that
[00:02:11] you have had or in your experience with
[00:02:13] legacy systems? Well, legacy systems are
[00:02:16] interesting because a lot of them work
[00:02:18] great. And I may have a unique opinion
[00:02:20] on this, but if you have a system that’s
[00:02:22] doing math or really basic things, the
[00:02:25] old systems work really, really well.
[00:02:27] Problem is, they don’t integrate well.
[00:02:29] It’s before the era of APIs and they
[00:02:32] don’t share with others well. Just
[00:02:34] sharing wasn’t a thing necessarily. If
[00:02:36] you look at the history of networking
[00:02:37] and sharing, some of those old programs
[00:02:39] were more set up like workg groupoups
[00:02:41] than they were enterprise. And so that’s
[00:02:43] the biggest challenge is taking that
[00:02:45] legacy information, extracting it, and
[00:02:48] then publishing it. If you’re not
[00:02:49] replacing, like you can do a forklift
[00:02:52] replace of a legacy system and put an
[00:02:54] all new system in. But if you have 30
[00:02:55] legacy systems and you all want them to
[00:02:58] talk to each other or you want them to
[00:02:59] talk to some people and maybe not
[00:03:00] others, then you’ve got to come up with
[00:03:02] a stop gap measure to modernize or at
[00:03:05] least make things good for your
[00:03:07] constituents where you pull it out, you
[00:03:09] put it somewhere where it can be used
[00:03:11] and integrated with an API and then you
[00:03:13] publish that to everybody else. So
[00:03:14] that’s one of the workarounds. Uh it
[00:03:16] works really really well particularly
[00:03:18] because whenever you’re on track for
[00:03:20] replacing a legacy system, it’s never
[00:03:22] fast. It moves at the speed of
[00:03:23] government which is very slow generally
[00:03:26] speaking and you’re it’s like oh how
[00:03:28] long is it going to take this you know
[00:03:29] your kids will be graduating from high
[00:03:31] school and going to college you know but
[00:03:33] by the time you finish replacing this
[00:03:35] legacy system having them run in
[00:03:37] conjunction that shouldn’t prevent you
[00:03:38] from modernizing. You can do some really
[00:03:40] creative things with the basic export
[00:03:42] features and import also and you can
[00:03:45] have them running in conjunction with
[00:03:46] each other, the old and the new in a way
[00:03:49] that doesn’t have corruption problems or
[00:03:51] any kind of data integrity problems and
[00:03:54] really modernize while you’re replacing
[00:03:56] the system. But that’s the biggest
[00:03:57] challenge for sure. One of the
[00:03:59] advantages of it also with not having
[00:04:01] the integration if you do have sensitive
[00:04:02] data you when you do the export I mean
[00:04:04] you know exactly what data you’re
[00:04:05] pulling out and when you turn on the API
[00:04:07] the API doesn’t have access to the
[00:04:09] legacy system it just has access to the
[00:04:11] table you told it to. So there’s some
[00:04:13] cool things you can do and you can also
[00:04:14] encrypt it even I mean so it’s basically
[00:04:17] airgapped if you want it to be it’s
[00:04:18] logically airgapped and digitally
[00:04:20] airgapped uh if you do it that way. So
[00:04:22] that’s one way you can do cool things
[00:04:24] that are preventing you from
[00:04:25] implementing AI. You hear a lot of
[00:04:27] excuses. We can’t do it. The legacy
[00:04:29] system’s never going to integrate. Well,
[00:04:31] can it export? I bet you it can. Can you
[00:04:33] do an import? I bet you it can. And you
[00:04:35] just got to get creative.
[00:04:36] >> You know, talking about public trust,
[00:04:38] um, it’s everything in government. So,
[00:04:40] how do you adopt AI without eroding
[00:04:42] that?
[00:04:42] >> That is a really great question. So,
[00:04:45] public trust with particularly with PII
[00:04:47] with encryption, there’s lots of ways to
[00:04:49] secure people’s information. People have
[00:04:51] a number of concerns. If you turn these
[00:04:53] tools on, are they going to actually be
[00:04:55] invasive to my privacy? Because people
[00:04:58] that are using a lot of these AI tools,
[00:05:00] they have no idea how they work behind
[00:05:02] the covers. They just know that if I use
[00:05:04] AI, it’s going to be like big brother
[00:05:06] and it’s going to come and spy on me.
[00:05:07] So, you have to do some education your
[00:05:10] constituents and users that this is not
[00:05:12] what it’s doing. It’s very restricted.
[00:05:13] Here’s the guardrails we have. So,
[00:05:15] you’ve got to do a lot of that. The
[00:05:16] guardrail work is important. And then
[00:05:18] also, you’ve got to do some buyin
[00:05:20] things. Look at how much easier it is
[00:05:22] for you to use AI because it’s making
[00:05:24] connections that previously you would
[00:05:26] have to go to four or five different
[00:05:28] websites to solve for and it’s solving
[00:05:30] for you right here in one place very
[00:05:32] quickly. And so it’s a combination of
[00:05:34] selling it and getting buy in and then
[00:05:36] demonstrating the value by increasing
[00:05:38] efficiency and convenience and accuracy.
[00:05:41] So if you’re if like the oldfashioned
[00:05:42] way of making a procon list like here’s
[00:05:45] all the good things, here’s all the bad
[00:05:46] things. If all the good things are, I’m
[00:05:48] saving you all kinds of time. Your data
[00:05:50] is always going to be accurate. It’s
[00:05:51] never going to be expired. You don’t
[00:05:53] have to go to six websites. You can go
[00:05:55] to one. The bad or the con is it might
[00:05:57] spy on me. It’s easy to actually tell
[00:05:59] people it’s really not spying on you.
[00:06:01] [laughter] And then also, it’s not going
[00:06:02] to take your personal data and publish
[00:06:04] it if you build it right. And so, you
[00:06:06] just have to the privacy concerns you
[00:06:08] have to actually go through methodically
[00:06:10] and make sure you’ve got great buy in
[00:06:12] before you start. I would start with
[00:06:13] that first because we know in technology
[00:06:16] we can do all the other things.
[00:06:17] Convincing people that we we have your
[00:06:19] information secured and it’s private
[00:06:21] that’s another matter altogether. It’s
[00:06:23] easier said than done.
[00:06:24] >> No that was a great answer Robin before
[00:06:26] I dive into you know the next theme for
[00:06:28] the questions that we have under this
[00:06:29] topic. I would like you to talk a bit
[00:06:31] about your journey from being the US Air
[00:06:33] Force officer to you know becoming a
[00:06:36] seasuit in tech side of the world. Were
[00:06:39] you always a techie guy or how did these
[00:06:41] two roads meet?
[00:06:42] >> Started out being very technical. One
[00:06:44] thing that’s unique about the military
[00:06:46] in particular is that they have give you
[00:06:48] leadership training. Part of in order
[00:06:50] for you to advance in the military, they
[00:06:52] send you to formal leadership school. So
[00:06:54] imagine your company or a company
[00:06:56] sending their team for weeks like two
[00:06:58] months because we want you to come back
[00:07:00] and know how to lead people. Well, this
[00:07:01] happens at every layer in the military.
[00:07:04] It happens on the enlisted side. So
[00:07:06] before you can become a certain rank,
[00:07:08] you have to complete leadership
[00:07:09] training. And it’s the same thing for
[00:07:11] officers. Before they’ll let you com
[00:07:13] become the next rank, you have to
[00:07:14] complete leadership training. And so
[00:07:16] it’s very unique in that matter because
[00:07:18] most corporations do not do it this way.
[00:07:20] And so it’s natural that you’re going to
[00:07:22] lead. It’s it’s guaranteed if you’re
[00:07:24] going to be make rank, you’re going to
[00:07:25] lead other people. It’s not always like
[00:07:27] this in corporate culture. You can be
[00:07:29] just a technician or an engineer and not
[00:07:31] really be a leader of anybody as far as
[00:07:33] officially. you can lead them informally
[00:07:35] because you’re good at what you do, but
[00:07:37] the expectation always is you’re going
[00:07:39] to lead. Even though I’ve been technical
[00:07:40] through the years and I was technical
[00:07:42] and still am, I’ve had lots of
[00:07:44] leadership training and so it just
[00:07:45] naturally rolled in to we want you to
[00:07:48] lead the people. And so it’s a
[00:07:49] combination of being technically
[00:07:51] competent, keeping your skills up,
[00:07:53] knowing um all the roadblocks and the
[00:07:55] gotchas, knowing what success looks
[00:07:57] like, and then simultaneously being able
[00:07:59] to lead and inspire others. And that’s
[00:08:01] kind of how it happened for me. I have
[00:08:03] no problem working with very technical
[00:08:05] people. I have no problem working with
[00:08:06] sales or marketing, whoever it is. And
[00:08:08] that was really the journey for me on
[00:08:11] transitioning from just being
[00:08:13] technically in the weeds all the time,
[00:08:14] the weeds or the trenches, whatever you
[00:08:16] want to call it, into I’m in the weeds
[00:08:18] sometimes and in the trenches sometimes,
[00:08:20] but I’m also out of the trenches and
[00:08:21] weeds directing traffic.
[00:08:23] >> Procurement and legacy debt slow
[00:08:25] everyone down. If you could change one
[00:08:26] rule about it tomorrow, what would it
[00:08:29] be? If I was to change one rule about
[00:08:31] legacy debt and procurement, if I start
[00:08:33] talking about procurement, I will go I
[00:08:35] could talk about it for hours. You’re
[00:08:37] talking about government or just in
[00:08:38] general?
[00:08:39] >> Governor, public sector.
[00:08:40] >> Public sector. The way that the public
[00:08:42] sector procures is not great. It’s what
[00:08:45] it does is that it puts them at an
[00:08:47] economic disadvantage because of how
[00:08:50] long it takes them to procure. You got
[00:08:52] beautiful new startups that are
[00:08:54] innovating that they save the government
[00:08:55] for last because they don’t have the
[00:08:57] working capital or the sales pipeline to
[00:09:00] be able to support how long it takes for
[00:09:02] government to procure. Now, people in
[00:09:04] government will say, well, we’re faster
[00:09:06] than others. They’re not faster than
[00:09:07] others. They may I don’t know who
[00:09:09] they’re faster than, but that’s the main
[00:09:11] disadvantage that we have. So, it ends
[00:09:13] up costing us more for the products and
[00:09:15] we end up always being behind on
[00:09:17] innovation because the people that are
[00:09:19] new save us for last. and rightfully so
[00:09:21] because if I have a deal that can close
[00:09:23] in three months for my new business or I
[00:09:25] have a government deal that takes 18
[00:09:27] months, I’ve got to get the deal that
[00:09:29] closes in three months. It’s just the
[00:09:30] way business works. So, it really puts
[00:09:32] us at an economic disadvantage. And
[00:09:34] there’s also some things we can do
[00:09:36] legislatively that helps with that. You
[00:09:38] know, when I was in I’ve been in the
[00:09:39] private sector also, not just in the
[00:09:41] public sector. And what we did in the
[00:09:43] private sector is they would want me to
[00:09:45] get three quotes. I mean, this sounds so
[00:09:47] elementary and simple. So, you go get
[00:09:49] three quotes. you show that you did your
[00:09:50] due diligence and your analysis and they
[00:09:52] say, “Hey, this looks good approved.”
[00:09:54] But the government, it’s a completely
[00:09:55] different model for depending on the
[00:09:57] value of what you’re trying to do. You
[00:09:59] have to put out a request for proposal.
[00:10:01] Oh my gosh. So, it’s very, very long and
[00:10:03] drawn out. And that in fact, I like I
[00:10:05] said, puts the government at a
[00:10:06] disadvantage. And that’s the one thing I
[00:10:08] would change is I would make that
[00:10:10] process shorter. I would allow for a lot
[00:10:13] more flexibility from the buyers while
[00:10:15] simultaneously raising the integrity of
[00:10:18] the system to weed out corruption. So,
[00:10:21] the big reason they have the RFPs is
[00:10:23] because a lot of people like, “Hey,
[00:10:24] we’re we’re in government. We’re going
[00:10:26] to we can expedite your deal and there’s
[00:10:28] corruption that’s involved.” You read
[00:10:29] about this in the news all the time. So,
[00:10:31] the government has all these extra
[00:10:33] layers in there to protect against
[00:10:34] corruption, but in fact, it hurts them
[00:10:36] in the long run because it costs them
[00:10:38] more. And there’s more modern ways to
[00:10:40] actually prevent corruption than doing
[00:10:42] it the oldfashioned way. And everyone in
[00:10:45] government is still doing it the
[00:10:46] oldfashioned way. If if you’ve ever
[00:10:48] played video games and you’re losing in
[00:10:50] the video game and things aren’t going
[00:10:52] well, you could hit the reset button and
[00:10:54] just hit reset. I’m starting the game
[00:10:55] over. I didn’t go far enough, whether
[00:10:57] it’s Mario, whatever game you’re
[00:10:58] playing. This is what they need to do
[00:11:00] with procurement and government. They
[00:11:01] just need to hit the reset button, get
[00:11:03] some people around the table to actually
[00:11:05] redesign the whole thing and make it
[00:11:06] where it’s a winning model for
[00:11:08] government where it has a positive
[00:11:10] impact on us. We’re not losing money. We
[00:11:12] can close faster. We can be where we’re
[00:11:15] corruption free. We follow all the
[00:11:16] ethical rules and we can innovate like
[00:11:19] the private sector or be really close
[00:11:21] behind them. We’re years behind. we’re
[00:11:22] literally almost like a full two years
[00:11:25] behind on the private sector when it
[00:11:27] comes to government when really we can
[00:11:29] be much much closer and which it helps
[00:11:31] the constituents. So that’s my take on
[00:11:33] government procurement. I follow I I’ve
[00:11:35] always followed the rules when it comes
[00:11:36] to these things. But it can be extremely
[00:11:38] frustrating as a leader when you’re
[00:11:40] trying to move things forward and do
[00:11:42] your your plan as a CTO like you know
[00:11:44] what whatever you’re trying to implement
[00:11:46] and you want you know they say what are
[00:11:47] you going to do your first 90 days it
[00:11:49] ends up there’s like no way you’re going
[00:11:50] to do it in 90 days in government. It
[00:11:52] would unless you are the governor and
[00:11:54] you can write executive orders and even
[00:11:56] then they’re constrained on what they
[00:11:58] can do based on what the current law
[00:11:59] says. It’s very challenging to be to
[00:12:01] move and shake in government.
[00:12:02] contractors running the old systems make
[00:12:04] money keeping them alive. So, some
[00:12:06] resist upgrades intentionally. Did you
[00:12:09] see that? And how did you get past it?
[00:12:12] >> What you do is you do it in a balanced
[00:12:15] way. Like I said, I’m not going to just
[00:12:16] turn off the legacy system and then flip
[00:12:19] the switch and have a migration for a
[00:12:21] really old system magically happen where
[00:12:23] it’s goes from old to new overnight. It
[00:12:25] takes time. And so if the contractor is
[00:12:28] interested in staying in business, you
[00:12:30] have that conversation and you say,
[00:12:31] “Look, thank you for maintaining this
[00:12:32] legacy system over the years. It’s been
[00:12:35] working great for us. Here’s the reasons
[00:12:36] we need to modernize. So we’d like to
[00:12:38] have you along with us on the journey
[00:12:40] cuz no one knows the system better than
[00:12:42] you, but we’re no longer going to be on
[00:12:43] the old system. We’re going to the new
[00:12:45] system. We’d like to have you with us.”
[00:12:46] and they either have to jump on board
[00:12:48] and be on the train or they stay with
[00:12:50] you until you migrate and then you have
[00:12:52] to wave goodbye which is very sad uh for
[00:12:54] people that have done that for a long
[00:12:56] time. But at the same time that you
[00:12:57] can’t hold up progress. The systems will
[00:12:59] be less secure. Your constituents won’t
[00:13:01] be as well served. The data won’t be
[00:13:02] governed as well. The list just keeps
[00:13:04] going on and on with the advantages of
[00:13:06] modernization. Now, not everything needs
[00:13:08] to be modernized. Some things just work
[00:13:10] and don’t need to be whisbang fancy. For
[00:13:12] example, you a calculator, whether it’s
[00:13:14] a calculator you hold, an old one like
[00:13:16] you actually physically hold from 1950,
[00:13:19] works the same as the calculator on your
[00:13:21] computer. So, some systems don’t need to
[00:13:23] be modernized. They’re just doing their
[00:13:25] job. They’re called legacy. They may be
[00:13:27] on an old system or an old version, but
[00:13:30] the fact is they’re working just great.
[00:13:32] So, don’t break it if it’s not broken. A
[00:13:34] lot of people immediately lean towards
[00:13:36] modernization when it’s not always
[00:13:38] necessary. You have to do the analysis.
[00:13:40] Rob, don’t you think that modernization
[00:13:42] is absolutely necessary when one talks
[00:13:45] about, you know, process automation, one
[00:13:47] talks about deploying AI into their
[00:13:50] systems and that is the layer that you
[00:13:51] absolutely cannot just neglect if you
[00:13:53] know you are moving towards that side.
[00:13:55] >> It’s true, but it depends. It depends on
[00:13:57] the situation. Not everything is
[00:13:59] digital, right? Some things are analog.
[00:14:01] Particularly in government, we forget
[00:14:02] this, right? It’s not just like
[00:14:04] e-commerce. You have people physically
[00:14:05] going to places for services. There’s
[00:14:07] vehicles, there’s there’s trucks,
[00:14:09] there’s fleets, there’s fire
[00:14:11] departments, there’s emergency services,
[00:14:13] there’s the list goes on and on. People
[00:14:15] that take care of the roads. And so some
[00:14:17] of these things aren’t digital in
[00:14:18] nature. You’ve got maybe you have an old
[00:14:20] machine that’s you they use to calculate
[00:14:22] tire pressure. You know, there’s all
[00:14:24] these analog things that uh that are out
[00:14:26] there. You just have to go and look at
[00:14:27] it and analyze it and figure out does
[00:14:30] this really need to be replaced? Most
[00:14:31] enterprises that are in the digital
[00:14:33] world, you would want to modernize
[00:14:36] everything. Rabia, so you’re right about
[00:14:37] that. Modernization is necessary. I’m
[00:14:39] not trying to downplay it. Sometimes we
[00:14:41] rush in government to modernize when we
[00:14:43] have a bunch of really great systems
[00:14:45] that are working just fine. If
[00:14:46] particularly if they’re doing sort of an
[00:14:48] analog job, then uh you have to do that
[00:14:50] on a case-byase basis.
[00:14:52] >> So Bob, you know, one can build by or
[00:14:54] partner. When does it make sense to
[00:14:56] build your own AI capability versus
[00:14:59] handing it to a hyperscaler?
[00:15:01] >> That is such a good question. I will say
[00:15:04] right now I envision that people
[00:15:06] building their own AI capability will be
[00:15:09] more forthcoming. The best examples I’ve
[00:15:11] seen is when it’s kind of a hybrid. You
[00:15:12] can develop your own with a partner. So
[00:15:14] the partner comes in and kind of aids
[00:15:16] you with your development of the AI. I
[00:15:19] would say the best bet government is to
[00:15:21] get with a great partner that has to do
[00:15:23] with AI governance. So being able to
[00:15:26] govern all of your agents, being able to
[00:15:28] have observability of your agents, being
[00:15:31] able to control them on every layer of
[00:15:34] the OSI model. So you can imagine all
[00:15:36] the layers of it. You have to be able to
[00:15:38] control all. You saw what just happened
[00:15:40] with hugging face open AI being able to
[00:15:43] do a lateral movement across their whole
[00:15:45] network just as an AI tool and it
[00:15:47] impacted everything. So you got to make
[00:15:49] sure that’s all right before you start.
[00:15:51] If you are not an expert in this
[00:15:53] subject, uh I think it’s a mistake to
[00:15:55] try to do it yourself without a partner.
[00:15:57] That’s my take on that. And not only
[00:15:59] that is you want to have a partner,
[00:16:00] you’re not getting the value if you’re
[00:16:02] homegrowing these. You can have all the
[00:16:04] AI tools that can help you homegrow
[00:16:06] them. And obviously there’s always
[00:16:08] exceptions like if you’re building a
[00:16:09] chatbot. Problem is, if the chatbot’s
[00:16:11] not secure, it can be a back door. I’m
[00:16:12] in favor of the partner model if not
[00:16:15] just straight away having a partner do
[00:16:17] everything for you in the beginning
[00:16:18] until your adoption increases you get
[00:16:20] more subject matter experts and then you
[00:16:22] can start moving to a more hybrid type
[00:16:24] of situation where you may develop some
[00:16:26] of your own agents
[00:16:27] >> before we talk about the reach and
[00:16:29] equity maybe you can share a bit about
[00:16:31] the AI tools that you’re using and
[00:16:33] you’re very comfortable with and also
[00:16:35] the ones that you’re um deploying at
[00:16:38] your leadership level within your teams.
[00:16:40] >> Great question. I use almost every every
[00:16:42] AI tool that’s out there. It’s uh I’ve
[00:16:45] been in it for a while and so it reminds
[00:16:48] me of when you had all these operating
[00:16:50] systems. You want to try everything. So
[00:16:52] it’s sort of like I’m at the ice cream
[00:16:54] shop and I have all the flavors of ice
[00:16:56] cream to choose from. So I’ve played
[00:16:58] with most of them. Gemini, Copilot,
[00:17:01] Quad, Chat, those are probably the main
[00:17:03] ones that I’ve messed with and I’m using
[00:17:05] all of them for in different ways for
[00:17:07] different uses just to try it and see
[00:17:09] how I like it. The ones that we’re
[00:17:11] implementing are really based on what
[00:17:13] they do best. So, Copilot may be better
[00:17:16] for enterprise. If you have a huge
[00:17:17] Microsoft environment and you already
[00:17:20] really invested with them as a partner,
[00:17:22] then it’s probably a good bet. If you
[00:17:24] need more, if you’re a big Google shop,
[00:17:26] then you can get in and use Gemini in a
[00:17:28] really useful way. Now, even IBM has got
[00:17:30] Bob coming out. So, I’m I’m engaged with
[00:17:33] IBM and Watson on some AI projects. The
[00:17:36] AI that I’m implementing now is
[00:17:38] basically at the user level for them to
[00:17:40] use it as a tool securely. I have other
[00:17:42] projects that I’m working on that that
[00:17:44] are very very great but I can’t disclose
[00:17:46] them because they’re it’s sensitive in
[00:17:48] nature but they’re using AI to its
[00:17:50] maximum capacity multiple um agents
[00:17:53] Gemini Watson you know Claude all
[00:17:56] wrapped up and using it in a very very
[00:17:58] awesome way to help me predict things in
[00:18:00] the future as far as business
[00:18:01] forecasting really deep learning
[00:18:04] analytics. I would love to talk more
[00:18:05] about that, but when it’s out, I think
[00:18:07] I’ll be able to talk about it. But right
[00:18:08] now, I’m right in the middle of building
[00:18:10] it. So, I’m in a very complex AI project
[00:18:12] right now. The ones I’m implementing
[00:18:14] mostly are getting on everyone’s, you
[00:18:16] know, people using it on their desktops,
[00:18:19] the ones that I’ve seen implemented at
[00:18:20] least, and just adoption overall.
[00:18:22] >> Yeah, great answer. Having said that,
[00:18:24] Rob, you know, most AI tools work best
[00:18:26] in English for people comfortable with
[00:18:28] tech in state like Oklahoma with Spanish
[00:18:30] speaking families, rural areas, and
[00:18:32] tribal nations. How do you keep AI from
[00:18:34] leaving those people behind?
[00:18:36] >> It’s one of the greatest things about
[00:18:37] AI. Literally, that’s one thing. If if
[00:18:40] the AI was a gift to society, breaking
[00:18:43] down language barriers is the gift.
[00:18:45] There’s nothing more beautiful than
[00:18:47] being able to communicate with someone
[00:18:48] that doesn’t necessarily know your
[00:18:50] language. And now I remember in my early
[00:18:53] days when I traveled places, I kept an
[00:18:55] actual physical dictionary, Spanish
[00:18:58] dictionary, in my pocket. And if there’s
[00:19:00] a word I didn’t remember, like a
[00:19:01] vocabulary word, I’d have to look it up
[00:19:03] in a Spanish dictionary, like a little
[00:19:05] pocket one that big. And now I can put
[00:19:07] on headphones and flip to the language
[00:19:09] and it flows through using AI and
[00:19:11] translates it in in my ear. It’s it’s
[00:19:13] almost I mean, we know how it’s working.
[00:19:15] So it’s not a miracle, but it feels
[00:19:17] miraculous for guys like me who had to
[00:19:19] use the book and look it up in the book.
[00:19:20] So any opportunity we can get to use AI
[00:19:24] to break down the language barrier, we
[00:19:26] should be doing it. We should be doing
[00:19:27] it in government. We should be doing it
[00:19:28] in our daily lives. Uh because the more
[00:19:30] you get to know people and the more that
[00:19:32] you are able to communicate in their
[00:19:34] language, the more enriched your life
[00:19:36] will be. And this gets into my book
[00:19:38] lunch sack leadership, right? I talk a
[00:19:40] lot about that in my book is uh getting
[00:19:42] to know people, getting to know their
[00:19:44] culture and being able to speak their
[00:19:45] language is really wonderful. It’s the
[00:19:48] essence of life. If you’re not doing it,
[00:19:49] you’re missing out. If you want to have
[00:19:51] an enriched life, go talk to people that
[00:19:53] don’t speak your language. Try to learn
[00:19:55] it. Eat the food. Learn the songs. learn
[00:19:57] the culture and let it enrich you like
[00:19:58] it’s never have before and government
[00:20:00] it’s easy one for public sector right
[00:20:02] because people now can read it read it
[00:20:04] in their own language but I would say
[00:20:05] it’s much bigger than that cuz I think
[00:20:07] it’s going to unlock the whole world it
[00:20:09] already has I mean I don’t think it it’s
[00:20:10] already has
[00:20:11] >> given that you talked about your book
[00:20:12] you know what’s the story behind naming
[00:20:14] it lunch sack leadership
[00:20:16] >> I actually went to a round table and all
[00:20:19] day long at the conference it was all
[00:20:21] text slides it was death by powerpoint
[00:20:23] and it was highly technical now some of
[00:20:24] them were really interesting so No
[00:20:26] discredit to the people up there. I had
[00:20:28] to host a round table and one of the
[00:20:30] topics that were ne almost never talked
[00:20:32] about was leadership. Like, okay, we
[00:20:34] understand you’re technical and you’re
[00:20:35] doing thing good things with tech, but
[00:20:37] how many people do you have reporting to
[00:20:38] you? 30. What do they think about that?
[00:20:40] And how are you leading them? How are
[00:20:41] you inspiring them? So, I uh I had a
[00:20:44] crazy idea that I would ask people how
[00:20:46] they made a peanut butter and jelly
[00:20:48] sandwich at my round table. So, I went
[00:20:49] around and asked everybody, very
[00:20:51] complex, actually. There’s hundreds upon
[00:20:54] hundreds of possibilities. And so I used
[00:20:56] it as a metaphor for business outcomes.
[00:20:58] That the peanut butter and jelly
[00:21:00] sandwich is a business outcome. But
[00:21:02] every person, no matter who they are,
[00:21:03] makes it slightly different. And so as a
[00:21:05] leader, you’re all working together as a
[00:21:08] team to accomplish the business outcome.
[00:21:09] But your job as a leader is to find out
[00:21:11] what makes them tick. Find out how they
[00:21:13] like to be led, how they like their
[00:21:15] sandwich. In the end, you get the same
[00:21:16] outcome. But the way you lead them is
[00:21:18] what how what what inspires them is what
[00:21:20] creates team cohesion. And it shows that
[00:21:22] you’re kind and considerate. So that’s
[00:21:24] really how the and when I did this I got
[00:21:26] a great response and it was all CIOS and
[00:21:28] CTO’s they said that was so great Rob we
[00:21:30] loved it and we thought that was the
[00:21:32] best round table we’ve been I said you
[00:21:34] got to be kidding me I just made it up
[00:21:35] it was sort of a creative moment an
[00:21:37] epiphany if you if you would say and so
[00:21:39] I did it again at another one and I got
[00:21:41] the same response I said I’m going to
[00:21:43] write about this and that’s that was how
[00:21:44] I got got on the book lunch sack
[00:21:46] leadership cuz you put the sandwich in
[00:21:48] your lunch sack some people say sack
[00:21:50] lunch some people say lunch sack which
[00:21:52] is a point of controversy for some of my
[00:21:54] executive friends.
[00:21:55] >> Talking about agencies now, they trust
[00:21:58] new generative AI more than the older AI
[00:22:00] that they’ve relied on for years for
[00:22:02] things like fraud detection. Why trust
[00:22:04] the flashier tool over the proven one?
[00:22:07] >> Very good question because fraud can get
[00:22:10] happen in lots of ways. You want to be
[00:22:12] able to apply different layers of threat
[00:22:15] intelligence to your models so they can
[00:22:18] detect it and and most of this is based
[00:22:20] on operational experience. And this is
[00:22:22] the challenge. Sometimes the AIs don’t
[00:22:25] have the operational experience. They’re
[00:22:27] not looking at it from a bird’s eye view
[00:22:29] holistically. They’re just going out and
[00:22:31] you tell it go do these three things and
[00:22:33] it just goes and does those three
[00:22:34] things. So being able to teach those
[00:22:36] tools context and being able to actually
[00:22:39] enrich what it’s doing. So it’s not just
[00:22:41] going through with an algorithm that
[00:22:43] check 1 2 3. Maybe it could be that and
[00:22:45] we know this is the way it works now
[00:22:47] with all these ID.ME tools and you could
[00:22:50] just name the identity management tool.
[00:22:52] So Cure is a great one. There’s dozens
[00:22:54] of them. I’ll probably get in trouble
[00:22:55] for not mentioning all of them at once,
[00:22:57] but there’s so many good identity
[00:22:59] management tools when it comes to fraud.
[00:23:00] And they say, “Hey, if if one if they
[00:23:02] have one and three, it could be fraud.
[00:23:04] If they have two, five, and seven,
[00:23:05] probably okay.” So it takes a while to
[00:23:07] train those. They’ve already trained
[00:23:09] them. we should definitely be using
[00:23:10] them, but at the same time, there’s
[00:23:12] going to be tools that try to thwart
[00:23:14] them because they if a if you have a an
[00:23:16] AI that’s actually with malicious intent
[00:23:18] and you have a threat actor, they can
[00:23:20] make the AI do the things that that get
[00:23:22] the one the one and three, right? So,
[00:23:25] you’ve got to really train it. You have
[00:23:26] to have experience on what threat looks
[00:23:28] like. you know, the threat actor,
[00:23:30] whether it’s digital, a digital threat
[00:23:32] actor, like it’s an AI tool someone
[00:23:34] deploys to try to pretend it’s somebody
[00:23:36] else, it I guarantee you it would get
[00:23:37] good at it, especially with deep fake
[00:23:39] and some of the tools they have. They
[00:23:41] even have it where you take take a
[00:23:42] picture of yourself and you know, you
[00:23:44] can show that you can fake out those
[00:23:46] systems all the way. When it comes to
[00:23:47] fraud and identity management, we have
[00:23:49] to really keep a sharp eye on it. And I
[00:23:51] would say that it’s going to get back to
[00:23:53] the old ways where it’s more physical
[00:23:56] and less digital. that may not make a
[00:23:58] lot of sense to people. Some of the old
[00:23:59] tradecraftraft ways of identifying who’s
[00:24:02] who may actually become more practical
[00:24:04] than the digital ways in the future. And
[00:24:06] that’s a long-term prediction. That’s my
[00:24:08] take on the fraud alert stuff.
[00:24:09] >> We see that everyone’s rushing to LLMs
[00:24:11] in a government context. Adaptability,
[00:24:14] FYI requests, accountability. What
[00:24:17] breaks um that vendors don’t warn you
[00:24:19] about?
[00:24:19] >> I would say security problems. So you
[00:24:22] can if you don’t have the right guard
[00:24:23] rails on your we talked about this
[00:24:24] earlier with data governance. You can
[00:24:26] turn on your AI to start learning and
[00:24:28] all of a sudden it’s learning things
[00:24:29] that you don’t want other people to see
[00:24:31] cuz you never governed it right. So I
[00:24:33] think that would break some things. You
[00:24:35] also have the possibility of turning it
[00:24:37] on and if you don’t that your your
[00:24:38] firewall rules and your some of your EDR
[00:24:41] and XDR and your your threat detection
[00:24:43] tools may look at it as bad. And vice
[00:24:46] versa, if you have something that
[00:24:47] emulates your AI, your LLM, it could be
[00:24:50] all cleared because you didn’t put any
[00:24:51] uh rules on it. So now it it once it’s
[00:24:53] behind it’s entered into your system and
[00:24:55] made the initial breach, it can move
[00:24:57] around as freely as it wants, just like
[00:24:58] we saw with Hugging Face. So I would say
[00:25:01] that that’s there’s more than one answer
[00:25:02] to your question. There’s probably five
[00:25:04] answers and I would say they have to do
[00:25:06] with every single layer of your IT
[00:25:08] infrastructure and your software
[00:25:09] infrastructure and your cloud
[00:25:10] infrastructure to your firewalls.
[00:25:12] There’s nowhere that an LLM is not
[00:25:14] going. All the modern enterprise tools
[00:25:16] now that are being released have MCP
[00:25:18] wrappers around everything. So you can
[00:25:20] instead of it being having to go in
[00:25:21] there and use language, programming
[00:25:23] language, you can just use regular
[00:25:25] language and query it and it does
[00:25:27] whatever you tell it to do. It’s in
[00:25:29] everything now and it can break
[00:25:30] everything. That’s what I’ll say. But it
[00:25:32] could also solve a lot of problems, but
[00:25:34] it definitely has the potential to break
[00:25:35] everything if you deployed in your in
[00:25:37] your systems without thinking about it.
[00:25:38] Well, Rob, was there a moment when you
[00:25:40] deliberately chosen to not automate
[00:25:43] something even when the efficiency case
[00:25:45] was obvious?
[00:25:46] >> No, I’m all for automation. The fact is
[00:25:48] when normally those roadblocks are they
[00:25:51] can be proven that the road the
[00:25:52] roadblock can be removed. Are we sure
[00:25:54] that it’s secure? You just have to do
[00:25:55] extra validation. Are we sure that it’s
[00:25:57] doing what we want it to do? Are we sure
[00:25:59] that it’s going to be secure? The data
[00:26:01] in it is secure so when it’s actually
[00:26:03] transporting it and communicating back
[00:26:05] and forth that that’s encrypted. Are we
[00:26:07] sure that it’s secure? That it’s not
[00:26:09] vulnerable to an outside attack. That’s
[00:26:11] a different type of security. There’s
[00:26:12] data security and then there’s system
[00:26:14] security. So once you validate those,
[00:26:16] you should be able to remove all
[00:26:17] obstacles to automation. Sometimes you
[00:26:20] have regulatory compliance and other
[00:26:22] factors that will block you and then you
[00:26:24] have to wait for the rule to change or
[00:26:26] you’ve got to go some kind of a
[00:26:27] workaround. But there should be nothing
[00:26:29] that prevents you from automating if you
[00:26:31] have your systems built right. So you
[00:26:33] should be able to automate. And
[00:26:34] remember, automation is not just always
[00:26:36] AI. I mean, there’s lots of ways to
[00:26:38] automate more than one. So, and and and
[00:26:40] if you’re smart, you do you can do a
[00:26:41] combination like a hybrid where you’re
[00:26:43] using different automation tools and
[00:26:45] workflow processes, good oldfashioned
[00:26:47] code and AI al together. And when it
[00:26:50] comes to automation, I would say if
[00:26:51] you’re saying no to it, there’s it’s
[00:26:53] probably a problem in your organization
[00:26:55] where they just don’t want to change.
[00:26:56] >> Yeah. Know, very interesting approach.
[00:26:58] So, talking a bit about the process and
[00:27:00] the people say AI is a mirror. The
[00:27:03] moment you use it, it shows you every
[00:27:04] broken process that you’ve got out
[00:27:06] there. How fast do you have to fix the
[00:27:08] process before you can really automate
[00:27:10] something?
[00:27:11] >> The speed is the is of the essence. It
[00:27:13] depends on the use case. If it’s all
[00:27:16] digital, you could probably fix it
[00:27:17] pretty quickly. If it’s not all digital
[00:27:20] because not all processes are digital.
[00:27:22] If I’m in a manufacturing line in a
[00:27:24] business, I’ve got people moving all
[00:27:25] around the floor. If I’m running
[00:27:26] emergency services, there’s all kinds of
[00:27:28] other analog things that are non-digital
[00:27:30] that are involved in the process. And
[00:27:32] our world in fact is mostly analog. So
[00:27:35] processes don’t just live in the digital
[00:27:37] world. If you’re dealing with just a
[00:27:38] digital process like a transaction, it
[00:27:40] should be something that you can solve
[00:27:42] quickly for and remove the roadblocks
[00:27:44] barring some technical restriction. Like
[00:27:46] I was saying before, if I try taking a
[00:27:48] legacy system and saying, “Hey, let’s
[00:27:50] write an API for this.” It’s so old that
[00:27:52] it won’t communicate with an API. You’ve
[00:27:54] got to get creative and move the
[00:27:56] information outside of that and uh put
[00:27:58] it someplace where it can be consumed in
[00:28:00] a modern place. Depending on what you’re
[00:28:02] trying to do, you can move the process
[00:28:04] quickly. The more digital it is and the
[00:28:06] more analog the process when just
[00:28:08] because it’s analog doesn’t mean it’s
[00:28:09] bad. You can’t digitally you can’t
[00:28:11] digitally walk your dog. You can’t you
[00:28:14] can you can play a video game and
[00:28:15] digitally hit a baseball, but it’s
[00:28:17] nothing like being out there at bat and
[00:28:19] swinging at the ball. So, that’s my take
[00:28:21] on that. It depends on what process
[00:28:22] we’re looking at. And there’s a lot of
[00:28:24] processes around that still involve
[00:28:26] humans. And so if the humans are
[00:28:27] involved, then you’ve got to do some
[00:28:29] process re-engineering. It’s more than
[00:28:31] digital. That’s also something you have
[00:28:32] to consider. I know that’s probably not
[00:28:34] the hot hottest topic, but if then
[00:28:36] there’s so many industries like this,
[00:28:38] Rivia, it is in a lot of industries that
[00:28:40] aren’t just people working behind
[00:28:41] computers.
[00:28:42] >> Yeah, I know you’re right. And you know,
[00:28:44] while we were talking about the dog,
[00:28:45] unless it’s a digital dog, because
[00:28:46] recently I was looking at [laughter] a
[00:28:48] video and China did come up with a
[00:28:50] crawling bot that could climb up
[00:28:52] mountains and even take up the loads of
[00:28:54] the mountains and it was it was a bit
[00:28:56] insane to see something like that, but
[00:28:57] yeah, it’s it’s very much true.
[00:29:00] >> Yeah. And you know what I will say too,
[00:29:01] it’s a model for uh not creating things
[00:29:03] in a vacuum. I’m a big believer in
[00:29:05] having stakeholder buyin and before I go
[00:29:08] and do some big digital automation, I
[00:29:10] want to talk to the people first that
[00:29:12] are using it every day. How is this
[00:29:13] going to impact them? How are they
[00:29:15] currently working and how is the
[00:29:17] implementation that I’m proposing going
[00:29:18] to impact their daily work? And so if
[00:29:20] you don’t do that early on, you’re
[00:29:22] telling them it’s going to be more
[00:29:23] efficient and it should be on paper, but
[00:29:25] until you go talk to them, there’s
[00:29:26] always new innovations that come out
[00:29:29] from talking to the people that are
[00:29:30] using the product. Trying to design it
[00:29:32] in a vacuum is always a mistake. And
[00:29:34] this is the same thing whether it’s just
[00:29:35] digital only or whether you’re dealing
[00:29:37] with something that’s a hybrid of people
[00:29:39] like working in a mechanic shop or on a
[00:29:41] manufacturing floor. You’ve got to go
[00:29:43] talk to the people. You have to
[00:29:44] understand how they’re using it.
[00:29:46] Specifically in business because if you
[00:29:47] get it wrong, it’s going to impact your
[00:29:49] revenue. Sales goes down. Your workforce
[00:29:51] is disgruntled now. And it’s all because
[00:29:53] the IT guy came in and said, “Hey, let
[00:29:55] me implement this new system without
[00:29:57] talking to anybody.” It’s part for the
[00:29:59] course. Actually, in our work, we think
[00:30:00] that we don’t need to talk to people
[00:30:02] very often. It’s very sad. You could
[00:30:03] have a digital dog, but all you don’t
[00:30:05] get any of the advantages, the health
[00:30:07] advantages of having a dog. Like when
[00:30:09] you’re around a dog, it lowers your
[00:30:11] blood pressure and your heart rate.
[00:30:12] Petting a dog makes you be relaxed.
[00:30:14] Hopefully, it’ll be interesting, right?
[00:30:16] Because what we’ve been doing for
[00:30:17] thousands of years, they’re now they now
[00:30:19] call 4D, which is, oh, I can you can I’m
[00:30:22] in this digital area and I can now smell
[00:30:24] it and I have the water spraying on me.
[00:30:25] If you ever been on one of those rides,
[00:30:27] it’s 4D. But really, it’s just real
[00:30:29] life. And I think 4D is cool. So I’m not
[00:30:31] me bashing the 4D experience, but it’s
[00:30:33] so interesting. These engineers, they
[00:30:35] want to experience 4D. I was like, go
[00:30:37] outside and walk around, get yourself a
[00:30:38] dog, go for a hike, go for a swim.
[00:30:41] >> Yeah. What is the shift that you’ve
[00:30:42] observed in your students, Rob? You
[00:30:45] know, the ones that you’re teaching
[00:30:47] currently or the ones probably it’s
[00:30:49] probably just been 7 months that you’re
[00:30:50] there, but if you know, you would have
[00:30:52] to um comment on that. How are students
[00:30:55] different from you know the students who
[00:30:57] were there probably 5 years ago? I would
[00:30:59] say that that you you talked about it
[00:31:01] earlier. It’s um being more engaged
[00:31:04] digitally than they were previously. And
[00:31:06] so some of this is great. It’s a great
[00:31:08] advantage being able to text somebody
[00:31:10] instantly. This is technology that’s
[00:31:12] very old. I understand this, but being
[00:31:14] able to look things up using an
[00:31:16] intelligent LLM, not just googling it.
[00:31:19] So I would say that’s the biggest shift
[00:31:21] that I would see is that you have the
[00:31:23] traditional search engine. People no
[00:31:25] longer using the traditional search
[00:31:26] engine. They actually would just open up
[00:31:28] straight away their Gemini, their
[00:31:30] Claude, their co-pilot, you know, their
[00:31:32] their chat and ask it questions for
[00:31:34] searching and which gives them much more
[00:31:36] enriched and robust answers because then
[00:31:39] they can pivot instantly and get an
[00:31:41] intelligent flow, a seemingly
[00:31:43] intelligent flow of how topics might
[00:31:45] connect. Google doesn’t do that for you.
[00:31:47] They they try and it pops up there and
[00:31:49] Gemini says, “I can be your AI
[00:31:50] assistant.” So, you can do that, but
[00:31:52] people are just going straight to the
[00:31:53] LLM. So, and there’s a lot of value in
[00:31:55] that. I will say that when I’m teaching
[00:31:57] them, I’m teaching them how to do
[00:31:59] prompts properly and really start
[00:32:01] thinking about the Socratic method. I
[00:32:03] specifically taught my my peanut butter
[00:32:05] and jelly, I did an exercise where I had
[00:32:07] that the uh one of the professors come
[00:32:08] in and we made them a peanut butter and
[00:32:10] jelly sandwich, but we documented the
[00:32:12] whole process. So, one of the and I had
[00:32:14] tasks for each of the students to do
[00:32:17] with AI. One of them was looking up all
[00:32:19] the different varieties, possible
[00:32:21] varieties of bread and the types and I
[00:32:23] want you to research bread. And the next
[00:32:24] person is, I want you to research peanut
[00:32:26] butter and all nut butters. And then I
[00:32:29] want you to research all of the actual
[00:32:31] probabilities of how many different ways
[00:32:33] you can make a peanut butter and jelly
[00:32:35] sandwich. So I want you to use this AI
[00:32:37] tool and I want you to do analytics
[00:32:39] against that to show you all of the all
[00:32:42] the probabilities on how many possible
[00:32:44] options there are. And then I want you
[00:32:45] So I had them all using AI for an analog
[00:32:48] thing. And I’m actually up there at the
[00:32:50] table making a sandwich for the one of
[00:32:52] the professors. and uh and they
[00:32:54] absolutely love it. And then I had two
[00:32:55] students up documenting everything that
[00:32:58] the professor did while she was making
[00:32:59] the sandwich. So as she’s and they were
[00:33:01] up there on the whiteboard, I’m like,
[00:33:02] “Okay, did you notice what what that she
[00:33:04] did just then? She did it this way. You
[00:33:06] have to document that.” So I was
[00:33:07] teaching them how to use the tool in a
[00:33:09] creative way, the tools in a creative
[00:33:11] way, and I was also teaching them the
[00:33:12] human component of understanding how
[00:33:14] unique people are, being able to honor
[00:33:16] their preferences. Just like we talked
[00:33:18] about before you build something and
[00:33:20] roll it out, you should probably talk to
[00:33:22] the people. And that’s what I was
[00:33:23] teaching them in the class is you can
[00:33:25] sit back with your LLM and you can query
[00:33:28] it all day. But until you observe the
[00:33:30] professor making the sandwich, you’re
[00:33:32] never going to get it all. You’re going
[00:33:33] to miss it. So, uh, that’s that’s the
[00:33:35] kind of stuff that I like talking to
[00:33:36] them about is this intersection of
[00:33:38] technology and humanity um, and using it
[00:33:41] in a really meaningful way to have a
[00:33:44] positive impact on others. I call it the
[00:33:46] double bottom line which is you do
[00:33:47] something great in tech that makes the
[00:33:49] business a lot of money and it
[00:33:50] simultaneously helps people.
[00:33:52] >> You know one of the global studies said
[00:33:54] that the top barrier to government AI
[00:33:56] isn’t money or tech, it’s staff skills.
[00:33:58] What skill do you think is actually
[00:34:00] missing over there?
[00:34:01] >> That’s a very good question. The skills
[00:34:02] would vary across the industry they’re
[00:34:04] in. But I would say in general
[00:34:06] government has two problems. They don’t
[00:34:08] generally sp pay for professional
[00:34:10] development and training. Now in some of
[00:34:12] the ones I’ve worked in they do they’ll
[00:34:14] provide like LinkedIn learning or
[00:34:16] they’ll provide uh ability for people to
[00:34:18] go and become certified but the problem
[00:34:20] with that is once they get certified
[00:34:22] they now become more valuable and
[00:34:24] government people generally are
[00:34:26] underpaid compared to the public sector.
[00:34:28] So it’s not necessarily a skill problem
[00:34:30] as much as it is as a retention problem.
[00:34:32] So you’ll have someone that gets skilled
[00:34:34] and learns a bunch of things and gets
[00:34:35] certified. They’re not probably going to
[00:34:37] stay in government because there’s a
[00:34:38] private sector job that’s offering them
[00:34:40] so much more money. Why would they stay?
[00:34:42] So the government overall has to get
[00:34:44] much more competitive on how they pay
[00:34:46] people. I mean it shouldn’t match the
[00:34:48] public sector, the private sector,
[00:34:50] excuse me, but it should be within 15%
[00:34:52] but 15% is a lot when you’re trying to
[00:34:54] get started. So it’s not a skills
[00:34:57] problem with the government. It’s a
[00:34:58] retention problem because all the
[00:34:59] skilled people leave. A private sector
[00:35:01] CTO can probably ship fast and break
[00:35:04] things faster, but you can’t really feel
[00:35:06] publicly on citizen services. How does
[00:35:09] that constraint change how you actually
[00:35:11] adopt AI?
[00:35:12] >> Well, you know what? It depends because
[00:35:14] sometimes we have the ability to to
[00:35:16] implement citizen services rapidly.
[00:35:19] Several of my colleagues here, one of
[00:35:20] them did automated our driver’s license
[00:35:23] and he they did that very rapidly where
[00:35:25] you could all do it, you could do it
[00:35:26] digitally. It was really quick and so it
[00:35:28] depends on the roadblocks. And then
[00:35:30] we’ve also had people do some really
[00:35:31] great things in transportation here with
[00:35:34] using the tools they already had. The
[00:35:35] good in government there are is
[00:35:37] lowhanging fruit and you can how you get
[00:35:39] the wins how you get the wins uh with
[00:35:41] that is getting those things
[00:35:43] accomplished early that have the m most
[00:35:45] impact to constituents like hey my
[00:35:47] driver’s license is now digital and I
[00:35:49] can do it really quick. That’s
[00:35:50] wonderful. And you can talk about that
[00:35:51] in the news. You can talk about it with
[00:35:53] the legislators and they all love it
[00:35:55] because it impacts them too. Another one
[00:35:57] we have one of my brilliant colleagues
[00:35:59] uh that I’ve worked with. His name is
[00:36:00] Tab Pierce. I’ve worked with him for
[00:36:02] years. He implemented something really
[00:36:04] great at the Department of
[00:36:05] Transportation that detects potholes um
[00:36:08] using AI. Really super fantastic uh use
[00:36:11] case. There’s others uh health and human
[00:36:13] services have done amazing things with
[00:36:15] how they interact with customers there
[00:36:17] using AI. So if you can grab the
[00:36:19] lowhanging fruit, it’s really really
[00:36:21] valuable. And the sooner you can do
[00:36:23] things that impact constituents, the
[00:36:25] more leverage you have in the future. I
[00:36:27] would say uh it’s not easy, but
[00:36:29] sometimes if you pick the right thing,
[00:36:31] go for the early win quick. Even though
[00:36:33] it may be something that long term you
[00:36:35] have a more complicated idea of what you
[00:36:37] want it to be, go for the easy ones. And
[00:36:39] there’s lots of easy ones out there.
[00:36:41] That’s what I would say.
[00:36:42] >> What is this one thing that you had to
[00:36:44] unlearn the hard way in your entire
[00:36:47] leadership tenure? I unlearned some
[00:36:49] things in leadership very early on which
[00:36:52] is it’s okay to admit when you don’t
[00:36:54] know something which is antithetical to
[00:36:56] being an engineer. Engineers hate doing
[00:36:58] this. Someone that’s technical hates
[00:37:00] saying they’re very hesitant to say I
[00:37:03] need to get help on this. So the early
[00:37:05] the sooner you can learn that cuz hey do
[00:37:07] you know about this? Do you know about
[00:37:08] that? It’s okay to say that you don’t
[00:37:10] know. Doesn’t mean that you can’t learn.
[00:37:12] It just means that you can’t be an
[00:37:14] expert in everything. So, always be
[00:37:16] leerary of someone that says that
[00:37:18] they’re an expert in everything. I don’t
[00:37:19] go to if I have a foot problem, I don’t
[00:37:22] go to the ear, nose, and throat doctor.
[00:37:24] I go to the podiatrist and they help me
[00:37:26] with my feet. If I have an ear, nose,
[00:37:28] and throat problem, I go to that doctor.
[00:37:30] Even doctors know this, right? But
[00:37:31] somehow in tech, we think that we should
[00:37:33] know everything. Same thing with
[00:37:34] lawyers, right? Lawyers have certain
[00:37:36] practices they specialize in. So, I like
[00:37:38] to say that I practice it. You have docs
[00:37:41] practice medicine. You have lawyers
[00:37:42] practice law. I practice it. And it
[00:37:44] doesn’t mean that I’m an expert in
[00:37:46] anything, everything. Excuse me. I can
[00:37:48] be an expert in some things and have a
[00:37:50] lot of knowledge on a lot of it. But
[00:37:51] it’s always okay to ask for help. Find
[00:37:54] someone you trust. Find a good leader
[00:37:55] and go to them and say, “This is an area
[00:37:57] that I really want to learn and be that
[00:37:59] person.” And then also don’t forget that
[00:38:01] you also have to do personal
[00:38:03] development. Don’t just invest invest in
[00:38:05] your tech skills. Go get some leadership
[00:38:07] training. Go take a personality test.
[00:38:09] Learn about the personalities and the
[00:38:10] preferences of your team members. It
[00:38:12] won’t just make you a better engineer.
[00:38:13] It’ll make you a leader. Even if you
[00:38:15] don’t want to lead people, everyone will
[00:38:16] be drawn to you because they want to go
[00:38:18] and ask you for help. They want to go to
[00:38:20] lunch with you. They want to be involved
[00:38:21] with your life because you took the time
[00:38:23] to become self-aware, get to know
[00:38:24] yourself both intellectually and
[00:38:26] technically and also personally and
[00:38:28] spiritually. So, it’s a win-win. That’s
[00:38:30] something that I had to in my early
[00:38:31] years. You know, I thought I had to know
[00:38:33] it all. And uh God help us. No one’s
[00:38:35] going to know it all.
[00:38:36] >> Well, Rob, thank you so much for such a
[00:38:38] sharp and practical conversation. Thank
[00:38:40] you for your time.
[00:38:41] >> Hey, you’re welcome. Thank you, Robia.
[00:38:42] Take care now.
[00:38:45] [music]
