[00:00:00] [music]
[00:00:04] Welcome back to another episode of Tech
[00:00:06] Unhinged where tech gets human. I’m your
[00:00:07] host Rabia Javeed and my guest today is
[00:00:10] Eric Gross, a technology leader with over
[00:00:12] 35 years of experience. Throughout his
[00:00:14] career, he has driven engineering
[00:00:16] transformations and software systems
[00:00:18] with major organizations including the
[00:00:20] US Navy, IBM, K Force, Pluralsight, and
[00:00:23] ROMtech. Authoring seven books including
[00:00:25] the uncommon engineer and co-founding
[00:00:27] the tech academy. Today he’s the founder
[00:00:29] of Mad Wizard Marketing and a technology
[00:00:31] career adviser who has helped thousands
[00:00:33] of engineers transition into
[00:00:35] architecture and executive leadership
[00:00:37] roles. Welcome to the show, Eric.
[00:00:39] >> Thank you so much. I’m very, very glad
[00:00:40] to be here.
[00:00:41] >> Well, Eric, we’ll start off with the ice
[00:00:42] breakers. You started your career
[00:00:44] maintaining nuclear reactors for the US
[00:00:47] Navy where unexpected failures aren’t
[00:00:49] really an option. Did that Navy guy
[00:00:51] monitoring control systems ever think
[00:00:54] he’d end up as a tech career coach? And
[00:00:56] how does that zero tolerance mindset
[00:00:58] shape you? How you look at AI risk
[00:01:00] today?
[00:01:01] >> I love the question. I’ll take up two
[00:01:02] parts of it. Did I see in my future
[00:01:04] anything like career coaching? 100% not.
[00:01:06] I’ll be frank. I wasn’t friendly person
[00:01:07] in the Navy. Besides the fact that I was
[00:01:09] a young idiot. We’re all not that bright
[00:01:11] when we’re young, even though we think
[00:01:12] we’re amazing. It’s just a different
[00:01:14] culture and like mentoring and sitting
[00:01:17] down and finding out what’s at the root
[00:01:19] of a blocker that someone might have,
[00:01:22] especially if it’s an internal belief
[00:01:23] that they have and helping them to
[00:01:25] recognize the fact that they’re actually
[00:01:26] really really awesome and they shouldn’t
[00:01:28] listen to those internal voices that say
[00:01:30] otherwise. That’s not part of the Navy
[00:01:31] culture. The Navy culture is sit down
[00:01:33] and shut up and do your job, right? and
[00:01:36] you would never complain like with
[00:01:38] sincerity and say, “I I feel like I’ve
[00:01:40] got something going on inside and I just
[00:01:42] kind of kind of need to work it out.”
[00:01:43] No, you don’t do that in the Navy,
[00:01:45] right? So, no, I never saw that in my in
[00:01:46] my future, right? However, the
[00:01:48] engineering culture, the discipline
[00:01:50] required to be able to train people
[00:01:52] effectively so that they can be trusted
[00:01:54] to operate a 185 megawatt nuclear
[00:01:56] reactor that the entire ship is relying
[00:01:59] upon for sailors to stay alive. Those
[00:02:01] lessons stayed with me for decades.
[00:02:03] they’re still there every single time I
[00:02:04] operate a machine or a computer or
[00:02:06] design a system like it’s not the same
[00:02:08] stakes but the stakes are real right so
[00:02:10] no I didn’t see myself moving into uh
[00:02:12] career coaching or anything like it at
[00:02:13] the time and probably I wouldn’t have
[00:02:15] been a lot of help to a lot of people at
[00:02:16] that young age I’ve gained a little bit
[00:02:18] of wisdom over the years
[00:02:19] >> well um Eric now if we ground uh you
[00:02:22] know the topic a little and define
[00:02:23] certain terms that we are going to be
[00:02:25] discussing today at length systems
[00:02:27] thinking and AI are two massive
[00:02:29] buzzwords at the moment from your days
[00:02:31] monitoring nuclear reactors to running
[00:02:33] enterprise ID. How do you define a
[00:02:35] system when AI agents are constantly
[00:02:38] modifying code workflows and data in
[00:02:40] real time?
[00:02:41] >> I mean a system is a system. It’s you
[00:02:43] know a composite of coordinated
[00:02:45] interrelated functions that have an
[00:02:48] overall goal. There’s a purpose behind
[00:02:49] the system, right? So the definition of
[00:02:51] what a system is hasn’t changed. I think
[00:02:53] what’s happened is we have brought into
[00:02:56] the fold a huge element of change in the
[00:02:59] structure of that system which cannot
[00:03:01] itself think. Full stop. Computers can’t
[00:03:03] think. Right? So as I’ve seen people
[00:03:05] progress in their careers and this has
[00:03:06] been true true for me and any of the top
[00:03:08] professionals we’ve worked with. What
[00:03:09] they accumulate over the years is like
[00:03:11] two different buckets. One of them would
[00:03:13] be the pure technical skills. Like
[00:03:15] there’s some tools, some language, some
[00:03:17] framework that they get really skilled
[00:03:18] in and that’s great. and they’re like a
[00:03:20] fine technician like compared to what
[00:03:22] I’m a master craftsman with using those
[00:03:24] tools. But the other thing they
[00:03:25] accumulate you don’t really list out on
[00:03:27] a resume in the same way and that is the
[00:03:30] ability to demystify confusion and to
[00:03:34] think with the first principle
[00:03:35] underneath something. The rock bottom
[00:03:37] fundamentals with it you wouldn’t
[00:03:38] necessarily put on a resume I can
[00:03:40] distill complexity down to first
[00:03:42] principles parts. Okay, that doesn’t I
[00:03:44] don’t think applicant tracking software
[00:03:46] really responds to that well, right? But
[00:03:48] it is a vital thing for anyone who’s
[00:03:50] just even if they’re just going to be
[00:03:51] and I didn’t mean this disparagingly top
[00:03:52] level individual contributor. They’re
[00:03:54] going to build up that reservoir of
[00:03:55] knowledge about systems and more
[00:03:57] importantly principles behind why a
[00:03:59] system designed a certain way. AI will
[00:04:01] not do that. It doesn’t really think in
[00:04:02] those terms. It just operates off of
[00:04:04] patterns and predict the next token.
[00:04:06] Right? The definition of the system
[00:04:07] hasn’t changed. What’s changed is we’re
[00:04:09] passing over control to both the design
[00:04:11] of those systems and their maintenance
[00:04:13] once they’re built to an agent. I use
[00:04:15] agent in the figurative and literal
[00:04:16] sense which will never actually have
[00:04:18] that deep understanding and so unless
[00:04:20] used properly what we’ve done is brought
[00:04:22] in a huge element of risk for the
[00:04:24] long-term health of those systems.
[00:04:25] That’s my point of view.
[00:04:26] >> How is offloading complex tasks to AI
[00:04:29] changing our own mental models and our
[00:04:31] capacity to think holistically.
[00:04:33] >> I use AI almost every day for my own
[00:04:35] development. And so, as someone who’s
[00:04:36] been doing this for decades, there was
[00:04:38] an aspect of what was being done for me
[00:04:40] using the AI tools that I didn’t
[00:04:42] recognize was going to be really harmful
[00:04:44] to someone new in their career. And
[00:04:46] here’s what that is. Let’s just get down
[00:04:48] to like the brass tax of like when
[00:04:49] you’re building with AI. A lot of the
[00:04:51] time you give the task an agent or
[00:04:53] you’ve got put it into plan mode and
[00:04:55] they’ve worked out these big, you know,
[00:04:56] major milestones along the whole journey
[00:04:58] of building out whatever it is you’re
[00:04:59] building and the AI will essentially
[00:05:02] build a whole chunk of the app. Frankly,
[00:05:03] with like Fable, you could tell it build
[00:05:05] the entire app. Don’t ask me for
[00:05:07] permission on anything and it will chunk
[00:05:09] away for a few hours and have a finished
[00:05:10] app. And there’s varying levels of
[00:05:12] quality to what it builds. And of
[00:05:13] course, we all know that you’re not
[00:05:14] going to trust it 100%, but most of us
[00:05:16] don’t take the time to read every single
[00:05:18] line. But that’s not even the point. For
[00:05:19] the person new in their career, here’s
[00:05:21] what’s missing from that experience.
[00:05:23] I’ve advocated for a long time when I
[00:05:25] working with engineers at any point in
[00:05:26] their career that you eat an elephant
[00:05:28] one bite at a time. And what I mean by
[00:05:30] that is whether you’re debugging a
[00:05:32] system or you’re designing and building
[00:05:34] a system, you build one tiny chunk of it
[00:05:36] and when you’re about to hit run or F5
[00:05:38] or whatever it is, you’ve already got in
[00:05:40] your mind a hypothesis of what exactly
[00:05:42] is going to occur. You’re running a
[00:05:43] little test and then you hit run or F5
[00:05:45] or whatever and you’re confirming your
[00:05:46] thinking and you’re doing that in tiny
[00:05:48] chunks at a time. Not only does that,
[00:05:50] you know, provide a really quick
[00:05:51] feedback loop, right? But it confirms
[00:05:53] your mental hypothesis that your
[00:05:55] approach to that tiny thing inside the
[00:05:58] grander scheme. Okay, I’m on the right
[00:05:59] track. And the moment you hit a snag and
[00:06:01] it doesn’t work the way it expects, you
[00:06:03] have to rethink. But you’re rethinking
[00:06:05] in a tightly bound space, that exact
[00:06:07] hypothesis is wrong. What’s wrong with
[00:06:09] my assumptions? When the LLM spits out,
[00:06:12] you know, 5,000 lines of code spread
[00:06:14] across seven different functions and
[00:06:15] there’s a config file and it says, okay,
[00:06:17] you know, milestone one is done, and you
[00:06:19] go in and it isn’t 100% what you want,
[00:06:22] and you’ve got to try to figure out what
[00:06:23] it’s thinking. You’ve got to suddenly
[00:06:24] grasp a lot of content mentally that you
[00:06:28] didn’t build. And sure, an experienced
[00:06:29] engineer can look through that created
[00:06:31] code and really quickly go, “Okay, good.
[00:06:33] No, no, that’s wrong.” If we do it that
[00:06:35] way, it’s going to be brittle 6 months
[00:06:37] from now. And here’s why. A new engineer
[00:06:39] isn’t going to have that ability. And
[00:06:40] so, what we’re losing, we’re losing some
[00:06:42] of that ability to work through those
[00:06:44] problems yourself, right? And there’s
[00:06:46] been a lot in the media about that and
[00:06:47] it’s valid, but we’re also losing the
[00:06:49] way applications are really built. No
[00:06:51] one builds the whole thing and then
[00:06:53] presses run. No one does that. No one
[00:06:55] builds a giant module and then runs that
[00:06:57] whole module. We make a small
[00:06:58] hypothesis, we build, we test, we
[00:07:00] expand, we build, we test, and that’s
[00:07:02] gone with the age of the AI.
[00:07:03] >> The risks are high, but yeah, there
[00:07:06] there are positives, but then yeah,
[00:07:07] there are negative.
[00:07:08] >> Oh, there’s lots of positives. Don’t get
[00:07:09] me wrong, there’s lots of positives. I
[00:07:11] love this technology.
[00:07:12] >> Yeah.
[00:07:13] >> But if you’re new in your career, Yeah.
[00:07:15] it isn’t serving you well in a couple
[00:07:17] specific areas. So Eric, where are most
[00:07:19] CIOS and CTO’s getting it wrong today?
[00:07:22] You know, I’m sure that you’ve worked
[00:07:23] with many. You talk to many still. Are
[00:07:26] they treating AI as a quick tool to plug
[00:07:28] into old broken systems rather than
[00:07:30] rethinking the system itself?
[00:07:33] >> There’s a couple different patterns that
[00:07:34] I’ve seen because I’ve been working with
[00:07:35] lots of senior leaders on these AI
[00:07:38] discussions, whether buy versus build
[00:07:40] and who the vendor is and what parts of
[00:07:43] our existing, you know, business do we
[00:07:45] leverage AI in. And there’s a few
[00:07:46] different sort of camps or areas where
[00:07:48] you can quote unquote get it wrong,
[00:07:50] right? One of them actually comes down
[00:07:52] in my mind to an engineering term that a
[00:07:54] lot of senior leaders just don’t know.
[00:07:56] And I I understand it. I don’t expect
[00:07:58] them to. They’re not engineers, which is
[00:07:59] a deterministic function compared to a
[00:08:01] non-deterministic function. A
[00:08:02] deterministic function, if you give it
[00:08:04] the same input, it will always give you
[00:08:06] the exact same output over and over
[00:08:07] again till the end of the time. 2 plus 2
[00:08:08] will always give you four. A
[00:08:10] non-deterministic function can give you
[00:08:11] a different output for the same input.
[00:08:14] And if you understand those two things
[00:08:15] fundamental level, then you’ll never
[00:08:17] make the decision to use AI where you
[00:08:20] need a deterministic output because it
[00:08:22] will not give it to you. That’s not
[00:08:24] their function. And so when I’ve had
[00:08:25] discussions with leaders who are trying
[00:08:27] to push AI into an area of the company
[00:08:29] that absolutely requires that
[00:08:31] deterministic output where for the same
[00:08:33] input you’re always going to get the
[00:08:35] exact same output. That’s one of those
[00:08:36] fundamental confusions that sometimes
[00:08:38] needs to be sorted out. The other one is
[00:08:40] when you’re trying to on the buy versus
[00:08:42] build. I see this a lot that because not
[00:08:44] that many people really understand the
[00:08:46] fundamental nature of generative AI,
[00:08:48] it’s really easy for a competent
[00:08:50] salesperson to sell a CIO a bill of
[00:08:53] goods to show that like their AI tool is
[00:08:56] going to come in and just like be the
[00:08:58] hottest thing since sliced bread and
[00:09:00] unlike before where let’s say a CIO was
[00:09:02] doing engineering up until about 5 years
[00:09:04] ago and has now moved to that leadership
[00:09:06] role, they could evaluate a lot of these
[00:09:07] tools really well, but if you haven’t
[00:09:09] been down in the trenches working with
[00:09:11] these new generative AI I tools, you
[00:09:12] don’t have that same ability to discern
[00:09:14] BS and truth in these sales pitches. And
[00:09:17] so, you’ll see AI tacked on in areas
[00:09:19] that either one, it’s super expensive
[00:09:21] and their in-house team could do it like
[00:09:23] that, or two, AI doesn’t even belong
[00:09:25] there. So, those are the two buckets,
[00:09:27] but there’s one more, and it’s actually
[00:09:28] got nothing to do with AI. It has to do
[00:09:30] with the data quality that the AI is
[00:09:32] going to operate off of. Many
[00:09:33] organizations I’ve worked with aren’t
[00:09:35] even ready for using a lot of the
[00:09:37] generative AI tools if they utilize the
[00:09:39] actual core data that’s been built up
[00:09:41] over years or decades of their company
[00:09:43] because that data is not yet ready to be
[00:09:45] consumed by AI. You see this a lot of
[00:09:47] time with like people records whether
[00:09:48] it’s a you know a customer or a patient
[00:09:50] or whatever spread across five different
[00:09:52] databases under five different SAS
[00:09:54] products. What’s the source of truth for
[00:09:55] a patient? the formats are different
[00:09:57] between all of them as an example and
[00:09:59] there’s false or incorrect data in there
[00:10:01] and you’re going to throw an AI in there
[00:10:02] that is non-deterministic and trust what
[00:10:04] comes out of it. You got to look at your
[00:10:06] data first. So those are the three sort
[00:10:07] of areas I see where we’re getting it
[00:10:08] wrong and I’m not a doom and gloom guy.
[00:10:10] I’ve seen incredible applications of AI
[00:10:12] but they have to be fit for the purpose
[00:10:13] for what that tool is what generative AI
[00:10:15] is.
[00:10:15] >> So in your recent work Eric running data
[00:10:18] systems and AI gateways what is the
[00:10:20] biggest architectural trap when
[00:10:22] integrating AI? How do you prevent local
[00:10:25] AI optimizations from creating invisible
[00:10:28] enterprisewide technical debt down the
[00:10:30] line?
[00:10:30] >> I actually think that the biggest hurdle
[00:10:32] isn’t a technical one. It’s a baseline
[00:10:35] education one. I’ve seen rollouts of AI
[00:10:38] tools, whether they’re off-the-shelf
[00:10:39] stuff from Frontier models or popular
[00:10:41] chat bots or their bespoke built from
[00:10:43] in-house AI tools where the strong
[00:10:46] knowledge of the fundamentals of AI,
[00:10:47] generative AI in particular, are just
[00:10:49] inside the engineering team and those
[00:10:51] robots do not go well. This is a
[00:10:54] paradigm shift not just in the
[00:10:55] technology area but in the business area
[00:10:58] across the world, right? And if you’re
[00:11:00] not investing in AI literacy across
[00:11:03] every single person working at the
[00:11:05] company, you’re not going to have a
[00:11:08] really good roll out of whatever tool it
[00:11:10] is you’re you’re putting in place. I
[00:11:12] guarantee you that the vast majority of
[00:11:14] the non-technical workers who are going
[00:11:16] to be impacted by or directly use these
[00:11:18] AI tools that are made, they don’t know
[00:11:20] what Eric’s yammering on about
[00:11:22] deterministic and non-deterministic.
[00:11:23] Like tell that nerd to shut up, right?
[00:11:25] And yet most of the time they’re just
[00:11:27] given a license to a really powerful AI,
[00:11:29] you know, chatbot tool and said, “Be
[00:11:31] more productive. We’re all using AI.
[00:11:33] We’re an AI first company.” But they
[00:11:34] don’t really know what that means. more
[00:11:35] more precisely they haven’t been given a
[00:11:37] a simple basic orientation on what is
[00:11:40] this tool actually doing how does it
[00:11:42] work without getting nerdy and technical
[00:11:43] you don’t need to you can explain it in
[00:11:44] about 5 minutes but more importantly
[00:11:46] what kind of work are you already doing
[00:11:48] that is and isn’t the right stuff to
[00:11:50] feed into this tool and that little
[00:11:52] teeny differentiation which again it
[00:11:54] takes about 5 minutes to really
[00:11:55] illustrate that to someone from that
[00:11:56] moment forward they’re like oh well this
[00:11:58] stuff I’m not going to use a AI for it
[00:12:00] at all but this stuff this takes me
[00:12:02] about 2 days to the beginning of every
[00:12:03] month to work out this report And it’s
[00:12:05] the same format every time. I can just
[00:12:08] generate the format in about a minute
[00:12:10] and a half and plug in a couple specific
[00:12:11] numbers and I just got back a day and a
[00:12:14] half every single month. I like that.
[00:12:16] Right. So it really is that literacy
[00:12:18] piece applied across the board.
[00:12:19] >> When you were at Rome Tech, you know,
[00:12:21] you did happen to save half a million
[00:12:23] dollars a year while building a
[00:12:24] production rack platform. If you can,
[00:12:26] you know, talk a bit about it, how did
[00:12:28] you happen to save that for that
[00:12:30] particular enterprise?
[00:12:31] >> It’s interesting. Most there’s two broad
[00:12:34] buckets that we found where there was
[00:12:36] savings. One of them was honestly just
[00:12:37] housekeeping. An audit of actual usage
[00:12:40] of these SAS licenses that we had. This
[00:12:42] is probably true for most enterprises
[00:12:44] out there, right? You do an audit on it
[00:12:45] and all we did was just audit the access
[00:12:47] logs and find out who’s actually logged
[00:12:49] into this thing and used it for the last
[00:12:51] 30 days. And we tried sending out
[00:12:53] messaging saying, “Hey, if you’re not
[00:12:54] really using this, let us know.” Look,
[00:12:56] people’s inboxes are crammed full. We
[00:12:58] got very little response back on that.
[00:13:00] But when we did a soft delete on that
[00:13:03] account, if you hadn’t logged in in 30
[00:13:05] days, it’s wild. 4% of the people whose
[00:13:08] accounts we soft deleted emailed a
[00:13:10] couple weeks later and said, “Hey, I
[00:13:11] can’t get in right now.” And we went,
[00:13:13] “Oh, sorry.” and reactivated their
[00:13:14] account. 96% of the people that we
[00:13:16] removed their account, who hadn’t used
[00:13:17] it in 30 days or more, never said boo.
[00:13:20] So, we’re paying for this license and
[00:13:21] they’re clearly not using it as
[00:13:23] evidenced by the audit logs. And they
[00:13:24] weren’t going to use it as evidenced by
[00:13:26] the fact that a couple months later,
[00:13:27] they still hadn’t said anything. Right?
[00:13:29] So that that’s one. It’s just you know
[00:13:30] housekeeping. But the other one is most
[00:13:33] SAS product have an overlap in
[00:13:35] functionality for a business. People
[00:13:37] have tried to cram so many features into
[00:13:39] their SAS products that you’ll find that
[00:13:41] you know this platform over here which
[00:13:43] is really was not intended for HR
[00:13:44] specifically does have an HR- like
[00:13:46] function. And another program over here
[00:13:48] which again not intended 100% for HR or
[00:13:51] maybe even really right in that HR
[00:13:53] wheelhouse. It’s got those same
[00:13:54] function. And so you find that you’re
[00:13:56] doing the same thing in two different
[00:13:57] places, right? And so finding a way to
[00:13:59] attract just the thing you’re using a
[00:14:01] SAS product for, which is usually a
[00:14:03] small subset of the huge number of
[00:14:05] features that they talk about when they
[00:14:06] sell the SAS product to you, right? If
[00:14:07] you’re you’re only using two or three
[00:14:09] things out of that, especially in the
[00:14:10] age of generative AI, helping software
[00:14:13] developers, what we found was we could
[00:14:15] very quickly create our own in-house SAS
[00:14:18] product that just replaced the two or
[00:14:20] three things that we were paying a
[00:14:21] tremendous amount of money for in this
[00:14:23] other SAS product that had so many
[00:14:25] features we weren’t touching. And so you
[00:14:26] pull those two or three things out,
[00:14:28] they’re in an in-house solution, and you
[00:14:29] get rid of that subscription. And now
[00:14:30] you’ve done two things. There’s the
[00:14:31] economic savings, which is great. But
[00:14:33] because that is custom software you
[00:14:35] built in house, now you can expand it
[00:14:36] over time tailored exactly to your
[00:14:39] company. You’re making your own SAS. So
[00:14:41] those are the two basic strategies. And
[00:14:43] by the way, I think at last count it was
[00:14:45] 1.6 million. So very effective. Now,
[00:14:47] there are companies that 1.6 million is
[00:14:49] a drop in the bucket. That’s not the
[00:14:50] situation here. So it’s very welcome
[00:14:52] savings. And you know the second part of
[00:14:54] the question Eric where you know how
[00:14:55] should a CIO be able to evaluate whether
[00:14:58] adding you know an AI to an existing
[00:15:01] stack would make their system more
[00:15:03] resilient. Is there like a checklist
[00:15:05] that you have on you that you adore to
[00:15:08] before diving into that and share with
[00:15:10] our listeners?
[00:15:11] >> Yeah I love that question. The
[00:15:12] resiliency piece is the key for me at
[00:15:14] least to me having some kind of a point
[00:15:16] of view about this question because
[00:15:17] without a lot of thought ahead of time
[00:15:19] introducing AI into the system almost
[00:15:21] always in decreases that resilience. It
[00:15:24] does so because one, you can never fully
[00:15:26] trust the output of an LLM of a model,
[00:15:28] right? And two, often the function that
[00:15:31] they are replacing or adding into a
[00:15:33] system is one that isn’t really well
[00:15:36] understood. So, especially if it
[00:15:37] involves human factors. There’s if you
[00:15:39] have a human being who’s done the same
[00:15:41] sort of function inside a company for
[00:15:42] any length of time, they’ve built up a
[00:15:43] tremendous amount of knowhow, tribal
[00:15:45] knowledge, the weird edge case that they
[00:15:47] handled this one time with a customer
[00:15:49] and they’re like, “Oh, that’s super
[00:15:51] important.” and they just kind of tell
[00:15:52] everybody, but it never makes its way
[00:15:53] down to a document anywhere. That sort
[00:15:55] of thing, right? And so if you’re trying
[00:15:57] to get an AI, which is in a general
[00:15:59] sense, AI is just getting computers to
[00:16:01] behave like humans. If you’re trying to
[00:16:02] get an AI to do that work that a human
[00:16:04] did and you don’t do a really good data
[00:16:06] capture of what does that really mean on
[00:16:08] the front end, what you’re actually
[00:16:10] introducing into the system when you
[00:16:11] replace that function with AI is less
[00:16:13] effectiveness, less resilience, more
[00:16:16] upset for the people that have to
[00:16:17] consume the work of the AI. On the other
[00:16:20] hand, if you do that data capture really
[00:16:21] well and you can give the AI very
[00:16:23] explicit instructions and copious
[00:16:25] examples of really good work, well, now
[00:16:28] it can produce near human quality work
[00:16:30] much faster and at lower cost. So that’s
[00:16:32] one aspect of it. The other aspect of
[00:16:34] like how CIOS and such can get into like
[00:16:36] how do you make that decision? This goes
[00:16:38] back to the data quality thing that I
[00:16:39] was talking about. If I’m a CIO and I’m
[00:16:41] being asked to like look at either an
[00:16:43] in-house, you know, off-the-shelf
[00:16:45] product that’s going to in any way touch
[00:16:46] my actual business data, my first
[00:16:49] question is where’s my data quality
[00:16:50] team? What does that data look like?
[00:16:52] When we stick an AI on top of this
[00:16:53] garbage in, garbage out, what are we
[00:16:55] dealing with? I would want to know that
[00:16:56] first and then talk to me about the AI.
[00:16:58] Even if the board is screaming that we
[00:17:00] have to be an AI first company, that CIO
[00:17:02] in concert with if he or she has a chief
[00:17:04] data officer, that’s when they need to
[00:17:06] put their attention before they go all
[00:17:07] in on AI. You’ve driven, you know,
[00:17:09] engineering transformations at massive
[00:17:11] organizations like IBM and Pluralsight.
[00:17:14] What is the biggest difference in how a
[00:17:16] Fortune 500 company approaches AI system
[00:17:19] architecture versus a lean enterprise?
[00:17:21] And where do the big players usually
[00:17:23] trip up?
[00:17:24] >> So, there’s a couple different aspects
[00:17:25] of this. If you’re a small player, then
[00:17:28] unless you are really special and nimble
[00:17:30] and pretty creatively thinking, you’re
[00:17:32] probably using some combination of the
[00:17:34] frontier models from anthropic or open
[00:17:36] AI uh or or or Google maybe having a
[00:17:39] little bit of like verbal guidance to
[00:17:41] your engineers to try to use the not the
[00:17:44] cutting edge model and you’re conserving
[00:17:45] tokens and that sort of thing. So that
[00:17:47] isn’t even speaking to architecture,
[00:17:48] it’s just speaking about how we use AI.
[00:17:50] But when you’re building things that are
[00:17:52] themselves going to interact via API
[00:17:54] with those models, Europe’s often going
[00:17:56] to have to use that frontier model or
[00:17:57] one of the available ones from the big
[00:17:59] companies. Fortune 500 enterprise
[00:18:01] company has the ability because they can
[00:18:03] think with the right capital expenditure
[00:18:05] and they have the senior, you know,
[00:18:06] enterprise architects and AI architects
[00:18:08] who can think with this to go all right,
[00:18:09] we can see the writing on the wall that
[00:18:11] a year and a half from now our expenses
[00:18:13] for using the intelligence that comes
[00:18:15] out of these commercial companies are
[00:18:17] going to explode. Why don’t we just buy
[00:18:19] several server racks full of massive
[00:18:21] compute? Put a router right in front of
[00:18:24] that compute and install opensource
[00:18:26] models on those rack and then anytime
[00:18:28] any work is being done, whether it’s the
[00:18:30] developer themselves creating something
[00:18:32] or the developer created an AI enabled
[00:18:34] tool that needs to call a model, we have
[00:18:36] a simple filtering process that routes
[00:18:38] it to the right place. Your small shop
[00:18:40] is not going to be able to do that. They
[00:18:41] they can’t do that, right? So I know
[00:18:43] that isn’t even speaking to the
[00:18:44] architecture of the application itself,
[00:18:46] but frankly the use of a model to do
[00:18:49] core work inside the application is what
[00:18:52] it’s all about. Everything else around
[00:18:53] that is is a deterministic work. It’s
[00:18:55] the work we as as engineers have been
[00:18:57] doing for decades. And some small or
[00:18:59] large part of that AI application you
[00:19:01] build right at one spot is going to
[00:19:03] reach out to a model and say give me
[00:19:05] non-deterministic output. give me
[00:19:07] intelligence as far as a model considers
[00:19:10] intelligence and pull that back and then
[00:19:11] do deterministic things with it. So
[00:19:13] those are the things one of the big
[00:19:14] differences I see between that
[00:19:15] enterprise market and the smaller shop.
[00:19:17] >> So you know currently um you know at Mad
[00:19:20] Wisit marketing your whole mission is
[00:19:22] helping businesses make money and run
[00:19:24] faster using modern automation. What are
[00:19:27] the foundational AI tools and automation
[00:19:30] platforms in your core tech stack right
[00:19:32] now and why did you choose them? This
[00:19:34] has shifted so much over the last three
[00:19:36] years. Before is like the same kind of
[00:19:38] things that everybody uses for
[00:19:39] automation, whether it’s N8 or Zapier or
[00:19:43] Make or whatever. It’s interesting.
[00:19:44] Those companies are there to make a
[00:19:46] profit and I don’t begrudge them that at
[00:19:47] all. But I can’t tell you how many times
[00:19:49] over the last year I’ve looked at the
[00:19:50] specific automation I need and went, you
[00:19:52] know, I can build that in about 20
[00:19:54] minutes right now and I can drop it into
[00:19:55] a function out on Azure and it’ll cost
[00:19:58] me unbelievably small fraction of what
[00:20:00] happens over on say Zapier or or N when
[00:20:04] the usage spikes. So my main tool now is
[00:20:07] generative AI and especially serverless
[00:20:10] functions to handle a lot of because all
[00:20:12] they’re doing they’re just moving data
[00:20:13] for the most part. They’re moving or
[00:20:15] transforming data with AI. Now he can
[00:20:17] build a little tool that will move or
[00:20:18] transform data in a very short period of
[00:20:20] time and even with a click of a button
[00:20:22] and if you got your credentials it’ll
[00:20:23] deploy it out on Azure and you’re done
[00:20:25] and then you move on to the more
[00:20:26] meaningful work. So tech stack for me
[00:20:28] has changed a lot over the last year or
[00:20:29] two.
[00:20:29] >> You took an operating model and raised
[00:20:32] developer productivity 500% in around 5
[00:20:35] months. So [clears throat] how much of
[00:20:36] that was introducing new tools versus
[00:20:39] changing how engineers mentally approach
[00:20:41] problem solving?
[00:20:42] >> Very little of it was new tools. In fact
[00:20:44] I give you an example. Azure DevOps
[00:20:46] great tool for both your automations
[00:20:49] related to the software development life
[00:20:50] cycle you know your continuous
[00:20:51] integration deployment and delivery all
[00:20:53] that kind of stuff right but it’s also a
[00:20:55] great tool for managing work well we’d
[00:20:57] had it at the company for years and I
[00:20:59] think maybe 5% of the engineers were
[00:21:01] using it so a very fundamental thing we
[00:21:03] did was not adding a new tool but
[00:21:05] writing training on how to use it and
[00:21:07] making it policy that you have to
[00:21:09] empower people to know how to use the
[00:21:10] tool that’s already there and then make
[00:21:12] it required period end of story you
[00:21:14] don’t do work that isn’t electronically
[00:21:16] tracked inside the tool. That one little
[00:21:18] thing changes a lot. First of all, how
[00:21:20] much can we hold in our mind? Oh, I’ve
[00:21:22] got to do blank blank and blank today.
[00:21:24] We’re engineers. We’re bright. But
[00:21:26] that’s cognitive overhead to get all
[00:21:27] fancy and nerdy about it. I’d much
[00:21:29] rather just look at a list that’s
[00:21:30] already formulated and go, okay, I’m
[00:21:32] going to take the top one and I can
[00:21:33] forget about everything under that list
[00:21:34] and I’ll just work on it. But for the
[00:21:36] executives, it’s tremendously powerful
[00:21:38] for us to be able to prioritize work.
[00:21:40] And unless you’ve got some digital
[00:21:41] record of it, you’re also trying to hold
[00:21:42] it in your head. So no, not new tools,
[00:21:45] just educating people on the fundamental
[00:21:46] tools of the job and making them use it.
[00:21:49] And after that, it’s a refinement thing.
[00:21:50] How well fleshed out are your task
[00:21:52] descriptions? Do you have clear success
[00:21:54] criteria? Right? Are you passing off
[00:21:56] instructions to QA that can make it very
[00:21:58] clear to them what success means for
[00:22:00] this task? But that came after the
[00:22:01] fundamental was here’s the tool. We’re
[00:22:03] going to have you use it, but we’re
[00:22:05] going to train you first. We had the big
[00:22:06] training session. It’s an hour, hour and
[00:22:08] a half long. Here’s a simple example of
[00:22:09] the workflow. Any questions? Good. From
[00:22:11] now on, that’s the way you do it. And we
[00:22:13] approached that with each of the
[00:22:14] developer tools. So we didn’t have to
[00:22:15] introduce new things. We had to make the
[00:22:17] simple baseline. How does an engineering
[00:22:19] organization, you know, what are the
[00:22:20] tools they commonly use? Train people,
[00:22:22] make them use it. That itself was huge.
[00:22:23] >> So Eric, whether it’s heavy machinery on
[00:22:25] a farm or AI in a deaf shop, leadership
[00:22:28] confuses tool access with actual
[00:22:30] capability. Why is treating training
[00:22:33] like a content delivery video course the
[00:22:35] biggest flaw in tech today? And how do
[00:22:38] we shift from isolated self-learning to
[00:22:41] true apprenticeship?
[00:22:42] >> I saw this question last night by the
[00:22:44] way. I couldn’t believe that you put
[00:22:45] this in here because you put it in here.
[00:22:47] One, it’s a great question. I didn’t
[00:22:48] even tell you that I wrote a book about
[00:22:50] this last month publishing. The book is
[00:22:52] called We Trained Everyone in AI and
[00:22:55] Nothing Changed, which I think
[00:22:56] [laughter] unfortunately is an all too
[00:22:58] common story in the industry right now.
[00:22:59] It’s not it’s not across the board true.
[00:23:01] I’ve seen some great rollouts. Right.
[00:23:02] Okay. Here’s my opinion on the matter.
[00:23:04] having tried to do this with several
[00:23:05] companies and varying degrees of
[00:23:07] success. Some were great, some failed,
[00:23:09] and I’m trying to figure out why, right?
[00:23:10] There’s the AI literacy piece that we’ve
[00:23:12] already talked about, right? If you’re
[00:23:13] going to, you know, give someone a
[00:23:14] license, please at least give them the
[00:23:16] foundational knowledge of what the thing
[00:23:17] is. But there’s a very precise thing you
[00:23:19] can do with the individual and their
[00:23:21] manager. In fact, if you do it the right
[00:23:22] way, you don’t have to do it with
[00:23:24] everyone. You have the manager do it.
[00:23:25] And that’s after they understand again
[00:23:26] that five-minute discussion. What’s the
[00:23:28] fundamental nature of a generative AI
[00:23:29] tool without getting all the nerdy, you
[00:23:31] know, the nerdy pulling the nerdiness
[00:23:32] out of it? what are the kind of tasks it
[00:23:34] does great at and it’s really not suited
[00:23:35] for. You can describe that really
[00:23:36] quickly. You have the manager ask their
[00:23:38] person, what are the three most
[00:23:40] timeconuming or irritating things you
[00:23:43] work with on a daily or weekly or
[00:23:45] monthly basis that now that you
[00:23:46] understand a little bit about how this
[00:23:48] tool, you know, what it’s suited for?
[00:23:49] What are the three things that you’d
[00:23:50] really like to see it help you save time
[00:23:53] with or produce a better output faster
[00:23:55] with? Right? And then you do two things.
[00:23:57] One, you measure the time it takes them
[00:23:59] to do it now. And two, you happen to do
[00:24:02] it with AI. And then there’s a third
[00:24:04] bonus which can be really fun, but you
[00:24:05] can save that for the day of the
[00:24:06] training. And that third bonus is bring
[00:24:08] to training three examples of how you
[00:24:11] did it before AI. Sterling examples,
[00:24:13] golden examples of here’s my report.
[00:24:15] This was perfect. My manager praised me
[00:24:17] for it or here’s here’s this output that
[00:24:19] I know is right and I made it myself.
[00:24:21] And then go into training, try to get
[00:24:22] the AI to make you a good report without
[00:24:25] it. And then give it those three reports
[00:24:26] and watch the difference. And I’ve
[00:24:28] watched that moment happen so many
[00:24:29] times. At that moment, they get they
[00:24:31] understand like, “Oh my god, this
[00:24:33] thing’s so powerful, but you’re
[00:24:35] responsible for choosing the right task
[00:24:37] and for giving it the best chance to
[00:24:39] succeed in helping you.” So, I see that
[00:24:42] piece missing a lot because that leads
[00:24:44] to so many things. If you if you take
[00:24:46] that approach, then one, the most
[00:24:48] important thing happens, which is the
[00:24:49] person uses it from that point forward.
[00:24:51] They’re like, “Oh my god, I feel like
[00:24:53] genuinely empowered. It’s not a buzzword
[00:24:55] anymore. I’m saving time and it’s
[00:24:57] producing really good output and I can
[00:24:58] put my mental faculties on this really
[00:25:00] gnarly thing over here that an AI is
[00:25:03] never going to solve. Thank you JPT. But
[00:25:05] the second thing is super powerful for
[00:25:07] the learning and development folks in
[00:25:08] the room for the the people who have
[00:25:09] been you know made responsible for
[00:25:11] rolling out AI and having it actually
[00:25:13] have return on investment and that’s
[00:25:14] this because you captured the before
[00:25:16] state in those interviews you know
[00:25:18] between manager and employee and then in
[00:25:20] the regular one-on ones after training
[00:25:22] you captured the after state. you better
[00:25:24] be collecting those. And so when you
[00:25:26] walk into a meeting being asked to
[00:25:28] justify your training initiatives, you
[00:25:31] walk in with the statistics, sure, but
[00:25:33] you walk in with a story. You find that
[00:25:34] one analyst who before was taking the
[00:25:37] first two days of every month, produced
[00:25:38] this incredibly gnarly report and now
[00:25:40] they’re getting it done in two and a
[00:25:41] half hours on the first day of the month
[00:25:42] and moving on to better things. You tell
[00:25:45] that story in the room and then you
[00:25:46] present the overall statistics and it’s
[00:25:49] true return on investment but that story
[00:25:51] speaks wonders and it gives it makes you
[00:25:53] trusted to do even more enablement
[00:25:56] empowerment helping people that’s what I
[00:25:57] found to be
[00:25:58] >> that was very insightful Eric so in your
[00:26:00] current work as a technology career
[00:26:02] adviser what is the single biggest skill
[00:26:05] gap you see in mid-career engineers who
[00:26:08] are trying to advance into architecture
[00:26:10] and leadership
[00:26:11] >> the gap is thinking it’s a skill gap and
[00:26:13] I’m not trying to be smart me or smart
[00:26:15] Alec about that. I was this is something
[00:26:17] that it took me a while to realize was
[00:26:19] occurring and when I realized it was
[00:26:20] happening it was a kind of surprise but
[00:26:22] it shouldn’t have been. I’ve worked with
[00:26:23] a lot of people helping them to try to
[00:26:24] to move up to whatever their next level
[00:26:26] is and that means whatever it means for
[00:26:28] the individual. It could be a money
[00:26:29] goal. It could be a position goal. It
[00:26:31] could be the kind of work that they’re
[00:26:32] doing they don’t like and they want to
[00:26:33] do different kinds of work. We all have
[00:26:35] goal. If you don’t have goals or
[00:26:36] purposes in life, I feel very bad for
[00:26:38] you. But often I would start attacking
[00:26:40] the problem from a skill viewpoint.
[00:26:42] what’s the skill that’s holding them
[00:26:43] back? But it’s almost never the skill
[00:26:46] that holds them back. It’s a thing that
[00:26:47] we don’t talk about a lot in this
[00:26:49] industry, which is the person’s
[00:26:51] viewpoint about themselves. I mean, we
[00:26:53] talk about imposttor syndrome. We don’t
[00:26:55] break it down a lot and most people will
[00:26:56] acknowledge it in certain ways. They
[00:26:58] have some, but the things we say to
[00:27:00] ourselves about ourselves, we rarely
[00:27:02] voice out loud and they’re usually not
[00:27:04] kind. I’m not ready. Other people are
[00:27:06] brighter than me. I don’t have the gift
[00:27:08] of gab and these people are going to get
[00:27:09] promoted because they can talk to people
[00:27:10] and I can’t talk to people. Like that
[00:27:12] internal voice can say really really not
[00:27:15] nice things to you. In every major
[00:27:17] breakthrough I’ve had with a client. The
[00:27:19] the outcomes are great. On the surface
[00:27:20] it looks like oh I help them make more
[00:27:22] money. I I did some work just the other
[00:27:24] day in calculating this. Most people
[00:27:25] I’ve worked with it’s at least a 25% pay
[00:27:27] increase directly tied to us working
[00:27:29] together. That’s not why I do it but
[00:27:30] it’s a good marker of the fact that
[00:27:31] something changed. And what changed
[00:27:33] wasn’t a new skill. It was I got them to
[00:27:35] examine their true worth. And that
[00:27:37] sounds smarmy or like airy fairy or
[00:27:40] whatever. I’ve got a built-in excuse. I
[00:27:41] was raised by hippies in the redwood
[00:27:43] forest of Northern California. So, let
[00:27:45] me be a little bit airy fairy. But all
[00:27:46] joking aside, that stuff matters. In
[00:27:49] fact, your opinion about yourself
[00:27:50] matters almost more than anything in the
[00:27:52] world because when you change that
[00:27:54] opinion, everything about how you
[00:27:55] interact with your environment around
[00:27:56] you changes. So, that’s what I’ve
[00:27:58] actually found the most. The biggest
[00:27:59] lever to improve a person’s career isn’t
[00:28:01] the next skill. The other thing is that
[00:28:03] most engineers are trying to stay
[00:28:04] current with the skills that are popular
[00:28:06] anyway. Like most engineers right now
[00:28:07] are studying AI. It’s not a skill thing.
[00:28:09] It’s an internal viewpoint thing. So
[00:28:10] that’s how I feel about that.
[00:28:11] >> No, I mean um I’m I’m really glad that
[00:28:13] you talked about this, Eric, because
[00:28:15] it’s it’s very rare to uh you know,
[00:28:18] people who are working in tech to
[00:28:19] actually point out these things or even
[00:28:22] leaders to talk about these things at
[00:28:23] length. While I was looking for quotes
[00:28:25] to put at the top end of the chapters in
[00:28:27] my book, The Uncommon Engineer, in this
[00:28:29] book about this subject we just talked
[00:28:31] about, I found this amazing quote that
[00:28:32] said, “We all have a blind spot and it’s
[00:28:36] shaped exactly like us.” I really like
[00:28:38] the way they said that we are rarely
[00:28:40] objective observers of our own strength.
[00:28:43] We’re very accurate about our weaknesses
[00:28:44] and we can magnify them a lot. But when
[00:28:47] you ask another person to say what’s
[00:28:48] awesome about you, it’s almost always a
[00:28:50] surprise. I didn’t look at that. And
[00:28:52] maybe we’re wired to look at the
[00:28:53] negative. Whatever it is, you can get so
[00:28:55] far. You can help someone so much by
[00:28:57] just asking them what about you is
[00:28:59] awesome.
[00:28:59] >> You almost just don’t want to believe
[00:29:00] them.
[00:29:01] >> It’s fun work. It’s challenging work to
[00:29:03] help someone genuinely look at
[00:29:05] themselves and see how awesome they are.
[00:29:07] Cuz I’ve never met a person ever who
[00:29:09] have doesn’t have something about them
[00:29:11] that’s actually admirable. It can take
[00:29:12] some work with some people to look and
[00:29:14] find that thing that’s admirable. But
[00:29:15] with most people, it’s really clear.
[00:29:16] I’ve got a good friend who’s also a
[00:29:18] super talented engineer that had this
[00:29:19] kind of block in like how he saw
[00:29:21] himself. And I’m looking at the guy and
[00:29:22] going, “Man, your ability to think with
[00:29:24] the back-end data integration, all the
[00:29:27] movement, the state changes, how to keep
[00:29:29] track of it, how to keep it safe blows
[00:29:31] my mind. And I’ve worked in that area
[00:29:32] for years. You’re really good at that.”
[00:29:35] And he didn’t see it at all. Just nope,
[00:29:36] not there. And then you get him to look
[00:29:38] and look again and look again, and he’ll
[00:29:40] finally go, “Huh, wait. Doesn’t
[00:29:41] everybody do that?” No, everybody
[00:29:43] doesn’t do that. That’s what I’ve been
[00:29:45] trying to tell you. I do that kind of
[00:29:46] work with people because I’ve been
[00:29:48] fortunate enough at some key moments in
[00:29:49] my life to have someone who cared enough
[00:29:51] about me. I don’t know why, but they
[00:29:53] cared enough about me to one to get me
[00:29:55] to look at something that I was doing
[00:29:56] that I shouldn’t be and fix it. But just
[00:29:58] as important to have me look at
[00:29:59] something I was really good at that I
[00:30:00] never noticed. And when they did that,
[00:30:02] it’s just like I I kind of think of like
[00:30:04] the total lifetime value of an engineer.
[00:30:06] How much help can you deliver over your
[00:30:08] lifetime of your work? And there’s two
[00:30:09] things I think that really increase that
[00:30:11] total lifetime value. One is those
[00:30:12] moments where you recognize, oh, I’m
[00:30:14] actually really good at that, cuz then
[00:30:15] you’re going to lean into that and do
[00:30:16] even more effective work. But the other
[00:30:17] one is those magical moments when you
[00:30:19] find out you were wrong about something
[00:30:21] that you really thought you understood.
[00:30:23] Those moments, you find out so much
[00:30:24] about the character of an individual
[00:30:26] when they find out, oh, that’s wrong.
[00:30:28] Many people will bounce off of it, hold
[00:30:30] on to it, seek to justify, insist they
[00:30:32] were right. And I used to do that when I
[00:30:33] was in the Navy. I was very much that
[00:30:34] way, right? But I learned painfully over
[00:30:36] the years, those moments are beautiful.
[00:30:38] That’s actually really important because
[00:30:39] if you can be humble enough to go, okay,
[00:30:41] I genuinely didn’t get that. What’s the
[00:30:42] truth? And you drill down and you
[00:30:44] genuinely understand something for the
[00:30:46] first time that you misunderstood every
[00:30:48] moment of work you do from that second
[00:30:51] forward in your entire professional
[00:30:52] career, you’re better. You just
[00:30:54] increased your total lifetime value as
[00:30:55] an engineer. So I think they’re kind of
[00:30:56] related.
[00:30:57] >> Yeah, absolutely. And you, you know, my
[00:30:59] next question is related to your book,
[00:31:00] The Uncommon, you know. So you focus on
[00:31:03] in that particular book you focus on
[00:31:05] helping developers become trusted
[00:31:06] decision makers. So in an age where AI
[00:31:09] instantly generates standard code. What
[00:31:12] specific thinking skill makes an
[00:31:14] engineer truly unreplaceable?
[00:31:16] >> It’s really interesting. My my primary
[00:31:19] skill here has actually been around in
[00:31:21] the public discourse for about 2500
[00:31:24] years. And people talk about this
[00:31:25] online. It’s not like I have some, you
[00:31:27] know, hidden secret or anything, but the
[00:31:28] way I’ve applied it has been very
[00:31:30] effective with myself and others. And
[00:31:31] that’s first principles, capital F,
[00:31:33] capital P. It’s those principles that
[00:31:35] descended from the Greek work on how do
[00:31:38] you break something down to its most
[00:31:39] fundamental components? And look, we all
[00:31:41] kind of know that as engineers, we deal
[00:31:42] with modular systems, but how do you
[00:31:45] apply that across the board in
[00:31:46] everything you do? If there’s a specific
[00:31:48] development tool you’re using, what what
[00:31:50] are the building blocks underneath it?
[00:31:51] But let’s focus on the the work that
[00:31:53] matters, the system you’re building. A
[00:31:54] really great question that can lead
[00:31:56] someone into thinking what this first
[00:31:58] principles area is to look at a part of
[00:32:00] a system and just ask a simple question
[00:32:02] on what else in the system does this
[00:32:04] depend and if the answer isn’t nothing
[00:32:06] keep looking because if you look long
[00:32:08] enough you’ll get to a thing inside
[00:32:09] there where it doesn’t depend on
[00:32:11] anything else. Everything else depends
[00:32:13] on it. That isn’t like something you can
[00:32:14] look up in a tutorial. You can’t go
[00:32:16] online and find the answer to that
[00:32:18] because your system is your system and
[00:32:21] it’s unique. But if you look long enough
[00:32:22] and you find that thing, two things
[00:32:24] happen. One, you can either build, if
[00:32:26] it’s a new system you’re designing, you
[00:32:27] can build that system better, or if it’s
[00:32:29] a system that already exists, you can
[00:32:31] maintain it and repair it better. But
[00:32:32] the other thing is that you can tell
[00:32:34] people about the technologies involved
[00:32:35] in that system so much better,
[00:32:38] especially to a non-technical audience.
[00:32:40] If you broke it down to that fundamental
[00:32:41] building block and then what’s layered
[00:32:43] on top of that and on top of that and on
[00:32:44] top of that, when you go to talk to an
[00:32:46] executive, especially a non-technical
[00:32:48] executive about what this thing is and
[00:32:49] what it does, there’s going to be zero
[00:32:51] technical jargon in your description.
[00:32:52] You’re not going to be that nerd in the
[00:32:54] room that makes that mistake of throwing
[00:32:55] out, I’m not even going to say, you
[00:32:57] know, you hear these words and a
[00:32:58] non-technical person just their eyes
[00:33:00] glaze over and they’re done. You’re not
[00:33:01] going to be that person. You’re going to
[00:33:02] describe what this thing does in their
[00:33:04] language because you understand it
[00:33:05] better than anyone in the room. So I
[00:33:07] think that’s probably the thing that AI
[00:33:09] is putting in danger the most because to
[00:33:11] to learn how to think that way and to
[00:33:13] actually apply it. AI doesn’t help with
[00:33:14] that. That’s you diving in and looking
[00:33:17] at some philosophy.
[00:33:18] >> So what is one core framework from this
[00:33:20] particular book that tech leaders can
[00:33:22] use right now to coach their teams into
[00:33:24] that mindset and what is the single
[00:33:27] biggest takeaway you want a reader to
[00:33:29] ponder at when they put the book down?
[00:33:31] tell you the one that I think would be
[00:33:33] the most valuable for the leader and be
[00:33:34] the most challenging for them is there’s
[00:33:36] a section in the book called the idea
[00:33:38] factory where I discuss um a way of
[00:33:41] thinking that I’ve been doing for
[00:33:42] decades even though my wife would make
[00:33:44] fun of me for it a lot and I didn’t
[00:33:45] quite understand why it worked but it
[00:33:46] did and then later found was based on
[00:33:48] research been done been done like 70
[00:33:50] years before and it’s essentially this
[00:33:52] when you’re struggling to solve some
[00:33:53] sort of a problem it’s not one where you
[00:33:55] just walk in you’re like oh you do this
[00:33:56] this and this and you’re done you’re
[00:33:57] like I can’t quite get this or like
[00:33:59] what’s the best way to refactor this
[00:34:01] complex system or I’m going to present
[00:34:03] you know at an industry conference
[00:34:04] what’s the core thesis I really want to
[00:34:07] you’re struggling to solve this thing
[00:34:08] right and you can do the following
[00:34:10] really this is greatly expanded in the
[00:34:11] book but first is to just gather data
[00:34:13] two kinds of data data about the problem
[00:34:15] you’re trying to solve in particular but
[00:34:17] also use general data in anything even
[00:34:20] tangentially related life human behavior
[00:34:22] the area of science that it’s related to
[00:34:24] whatever you just specific things about
[00:34:26] the problem and then more general data
[00:34:28] and you’re just gathering this data and
[00:34:29] then you’d sit down and you’d grab any
[00:34:31] one of those pieces of data and start
[00:34:32] looking at it. What does this mean? How
[00:34:34] does it compare and contrast to the
[00:34:35] other pieces of data I was looking at?
[00:34:36] And you’re looking for connections that
[00:34:38] you didn’t see before. And you do that
[00:34:39] and you’ll often start to get some
[00:34:40] pretty cool ideas about this. But after
[00:34:42] a while, that kind of work is going to
[00:34:43] just you’re going to hit this wave like
[00:34:45] okay, all right, I’ve looked at this
[00:34:46] enough. And you step away for a minute
[00:34:47] and then come back to it. And finally,
[00:34:48] you’re just going to be that’s it. I’m
[00:34:50] done looking at this. I’ve looked at all
[00:34:51] this data. I’m not seeing anything. I
[00:34:53] still can’t solve the problem. And
[00:34:54] that’s when you do the thing that is
[00:34:56] hardest for the engineer and would be
[00:34:58] hard for a leader to tell people to do,
[00:35:00] which is you literally stop working on
[00:35:02] the problem. You stop thinking about it.
[00:35:03] >> Yeah.
[00:35:04] >> You walk away. And to an outside
[00:35:05] observer and sometimes even yourself, it
[00:35:07] feels like you’re giving up. But you
[00:35:08] aren’t. You’re actually having a part of
[00:35:09] your mind that is expert at it. Chew on
[00:35:12] that problem. I’ve used this over and
[00:35:14] over again. And some point in the near
[00:35:17] future, the idea is going to go bang. If
[00:35:19] you’re lucky, it’ll happen while you’re
[00:35:20] awake. Sometimes it happens while you’re
[00:35:21] asleep and then you’re awake. And you
[00:35:23] better get up and write it down. But it
[00:35:24] happens. And then the last thing you
[00:35:25] want to do is go, “Okay, wow. Holy moly,
[00:35:27] that’s a great idea.” Show it to other
[00:35:29] experts in that area. People who are
[00:35:31] qualified to judge it and say, “Pick
[00:35:32] this apart. What did I miss? How can we
[00:35:34] make it better?” If a leader does that
[00:35:36] with gnarly problems, either does it
[00:35:38] themselves or empowers their people to
[00:35:40] do it. We talk about a culture of
[00:35:41] innovation and you see it in the media
[00:35:43] all the time and the CEO says, “We’re an
[00:35:44] innovative company.” That is a tool for
[00:35:46] innovation that has never failed me. And
[00:35:48] if a leader can practice that themselves
[00:35:50] and give people the space to do it,
[00:35:52] especially during that quiet period
[00:35:54] where it looks like they’re not actually
[00:35:55] working on the problem, you can get
[00:35:57] amazing solutions out of people.
[00:35:58] >> What is that one hard learned lesson
[00:36:00] from your career that completely changed
[00:36:03] how you approach life, technology, or
[00:36:05] leadership?
[00:36:06] >> It’s always it’s always the right thing
[00:36:07] to do to be kind. I don’t often succumb
[00:36:09] to being unkind. I can mentally
[00:36:11] inventory the moment because they were
[00:36:13] so painful to the other person, but they
[00:36:15] were painful to me. They stick. It’s not
[00:36:17] our natural state as humans to be
[00:36:20] unkind. So that has that’s not a
[00:36:21] technical thing. That’s not a business
[00:36:23] thing. It’s one human being to another.
[00:36:25] It’s always the right thing to do to be
[00:36:26] kind.
[00:36:26] >> Thank you so much, you know, for your
[00:36:28] energy, for for your truthfulness, for
[00:36:30] your unhind, you know, for sharing it
[00:36:33] all with our audience.
[00:36:34] >> Absolutely. My pleasure. Thank you so
[00:36:35] much.
[00:36:37] [music]
