WEBVTT
00:00:00.160 --> 00:00:10.880
Welcome to Agentic Edge, where we explore the frontier of AI agents, enterprise orchestration, and the architectures that are shaping tomorrow's intelligent enterprises.
00:00:11.039 --> 00:00:16.079
My name is Micah Smith, and joining me as always is my co-host, Kate Wrestler.
00:00:16.320 --> 00:00:20.000
And today we are diving into AI transformation.
00:00:20.160 --> 00:00:30.960
And it's one of those phrases that get used to describe absolutely everything from hey, we put a chat bot on our website to our company fundamentally has rewired how we operate.
00:00:31.199 --> 00:00:45.039
Organizations make big announcements about it, boards approve huge budgets for it, press releases go out, and then ultimately someone has to figure out what all of this means and how they can drive meaningful outcomes for the organization.
00:00:45.359 --> 00:00:50.240
To help us break all of that down, today's guest is Dennis Ganesh.
00:00:50.320 --> 00:01:01.840
He is CTO at RPG, and he's responsible for leading technology transformation in what's a very regulated and safety critical industry, which means he can't just fake his way through this.
00:01:02.079 --> 00:01:11.040
So Dennis is also the author of the Architect's Blueprint on Substack, which is honestly one of the few Substacks I read consistently.
00:01:11.359 --> 00:01:13.680
Dennis, welcome to Agentic Edge.
00:01:14.159 --> 00:01:14.879
Michael, thank you.
00:01:15.040 --> 00:01:15.920
Thank you for having me.
00:01:16.319 --> 00:01:26.319
Dennis, I want to start off with kind of a fundamental question here to ground the understanding for everyone in our audience, but how would you define AI transformation?
00:01:26.640 --> 00:01:28.079
Sure, that's a good question.
00:01:28.560 --> 00:01:38.640
For me, AI transformation begins where the presence of machine intelligence actually changes how the company operates.
00:01:38.799 --> 00:01:41.760
It's not just about making individual employees faster.
00:01:41.920 --> 00:01:43.280
We often talk about that a lot.
00:01:43.439 --> 00:01:45.120
There's tremendous amounts of traction there.
00:01:45.200 --> 00:01:47.599
You hear all the time in LinkedIn and other Substacks.
00:01:47.760 --> 00:02:08.800
It's really about how we change how we work, change our fundamental workflows, where the decisions are made, which activities are actually automated, which situations still require human in the loop or human judgment, potentially how many teams or how teams themselves are fundamentally structured or organized.
00:02:08.960 --> 00:02:12.400
That's, I think, is really the point of uh transformation.
00:02:12.560 --> 00:02:19.840
Now, giving employees co-pilot, GPT, Claude, any other AI assistance known to mankind is a very valuable asset for sure.
00:02:19.919 --> 00:02:24.639
I'm not saying that's not valuable, but to me, that's just uh primary adoption.
00:02:24.719 --> 00:02:26.240
That's just the the start.
00:02:26.560 --> 00:02:37.840
You could give 5,000 employees an AI tool, and you can still have the exact same process, the same handoffs, same approvals, same organizational structures.
00:02:38.000 --> 00:02:44.800
And so people become a little bit more productive for sure, but the company itself hasn't fundamentally transformed.
00:02:44.960 --> 00:02:47.360
There is no transformation that's really happening here.
00:02:47.599 --> 00:02:54.800
And so for me to answer your question, transformation really begins when you fundamentally ask it a different way.
00:02:54.960 --> 00:03:08.000
I would rather ask something like knowing what these AI capabilities and tool sets can do, knowing the capability and what you have at your fingertips today, would you design the existing workflow the same way?
00:03:08.240 --> 00:03:11.199
Take a financial process, for example.
00:03:11.439 --> 00:03:15.120
A lot of institutions that I've worked with, you know, there's multiple steps in there.
00:03:15.199 --> 00:03:22.639
There's human review, there's approvals, AI adoption might still be useful there and might expedite some of that.
00:03:22.719 --> 00:03:26.000
And it might help a financial analyst prepare something faster.
00:03:26.159 --> 00:03:28.159
But fundamentally, the workflow is still the same.
00:03:28.240 --> 00:03:47.919
You know, transformation might eliminate a step or two of that six-step process and automation may reduce some of that low-risk transactions, but continuously checking the data, sending the exceptions to the people only, fundamentally re-architecting, designing that workflow, I think that's the piece.
00:03:48.000 --> 00:03:58.960
And so if an organization is focused on uh AI transformation, they're fundamentally redesigning how the organization thinks, how the organization even uh looks.
00:03:59.039 --> 00:04:02.000
And that's a critical part of organizational intelligence.
00:04:02.319 --> 00:04:11.439
We're all on the same page here that pretty much every company out there right now is saying that they're on some sort of AI transformation journey.
00:04:11.680 --> 00:04:21.279
But in your experience, what is the difference between what they're saying and describing as their transformation and what's actually happening on the ground?
00:04:21.680 --> 00:04:22.879
No, that's a good follow-up.
00:04:23.120 --> 00:04:26.000
I think everyone says they're doing AI transformation.
00:04:26.079 --> 00:04:29.600
In fact, I think LinkedIn, I see nothing but AI activity there.
00:04:29.759 --> 00:04:30.879
I think it's a misnomer.
00:04:30.959 --> 00:04:38.240
When they talk about AI transformation, I think they're alluding to maybe some uh productivity enhancements they're seeing.
00:04:38.319 --> 00:04:41.759
And so, and I think they attribute that to word transformation.
00:04:41.920 --> 00:04:46.079
And that's not necessarily a criticism on that organization or their approach.
00:04:46.240 --> 00:04:47.439
I think that's the first stage.
00:04:47.600 --> 00:04:48.959
Productivity is critical.
00:04:49.040 --> 00:04:50.959
And so those enhancements are real.
00:04:51.120 --> 00:05:00.079
And so I don't want to say that they are not, but we should be really honest about the level of change that's actually occurring inside of an organization.
00:05:00.319 --> 00:05:07.759
Organizations measure frequently the number of AI licenses they issue out, what they have purchased, their overall stack.
00:05:07.839 --> 00:05:18.319
Uh, we also measure the number of pilots they've launched, the number of employees they have trained, uh, maybe the number of AI use cases that they even identified and that sits in their backlog.
00:05:18.720 --> 00:05:21.759
But those measurements tell you that AI is happening.
00:05:21.920 --> 00:05:31.839
There is an appetite for it, and maybe there's a um a mandate from leadership to push that along, but they don't tell you how you fundamentally changed.
00:05:31.920 --> 00:05:39.360
And so to measure that delta, I look at really four stages of maturity when I look at a particular organization.
00:05:39.600 --> 00:05:43.279
The first stage is really uh individual assistance.
00:05:43.519 --> 00:05:51.040
Does AI help a person summarize, write an email, allow us to research something, and maybe even write a piece of code?
00:05:51.120 --> 00:05:52.480
And I think that's the first stage.
00:05:52.639 --> 00:05:58.079
The second stage is really how are we reimagining and augmenting our existing workflows?
00:05:58.160 --> 00:06:05.120
And so AI becomes a part of that business process, but that basic process largely remains intact.
00:06:05.199 --> 00:06:10.639
And so we're peppering in AI there, but really not reimagining the activities itself.
00:06:10.879 --> 00:06:19.439
Third stage is when some organizations understand that they have squeezed as much as they can get out of the first few stages, they start looking at the fundamental process design.
00:06:19.519 --> 00:06:28.160
And so they think about the activities that they do and interactions they do with one another, um, the handoffs that they have today and the silos that they have built.
00:06:28.319 --> 00:06:35.439
When you see those disappearing because of AI and automation, you can then fundamentally you've changed the organization.
00:06:35.519 --> 00:06:36.800
You're behaving differently.
00:06:36.959 --> 00:06:42.639
And I think that will push you and lead you to the overall operating model redesign.
00:06:42.720 --> 00:06:49.360
And so I would measure, I would understand that there are stages to it, there are phases to it.
00:06:49.519 --> 00:06:51.519
Don't measure with these false metrics.
00:06:51.600 --> 00:06:56.879
Don't measure with the number of pilots you're running and how many folks are on your internal managed GPT.
00:06:57.120 --> 00:06:59.360
Instead, look at your business outcomes.
00:06:59.519 --> 00:07:02.079
Did your overall cycle time improve?
00:07:02.240 --> 00:07:03.680
Did your error rate decrease?
00:07:03.839 --> 00:07:08.079
Did your customers wait still that waiting time did that decrease?
00:07:08.319 --> 00:07:10.800
Those are critical things that I kind of look for.
00:07:10.959 --> 00:07:11.600
I like that.
00:07:11.759 --> 00:07:14.560
A lot of organizations get stuck measuring the wrong stuff.
00:07:14.800 --> 00:07:16.560
And you mentioned a lot of those cases.
00:07:16.639 --> 00:07:22.560
I see a lot of people who are talking about the number of tokens used per employee and a measurement of adoption.
00:07:22.720 --> 00:07:24.720
And I think that's missing it.
00:07:25.040 --> 00:07:36.639
If that basic measurement of adoption is kind of stage one, like you're talking about, the gold standard is probably something like ARR per employee, right?
00:07:36.800 --> 00:07:38.160
At stage four.
00:07:38.480 --> 00:07:42.000
How are you looking at measuring those individual stages?
00:07:42.160 --> 00:07:48.480
And how does the maturation mature as you go through, like, oh, we're moving from stage one to stage two?
00:07:48.959 --> 00:07:52.639
We know that everyone has the tools, we know that they're aware of what's available.
00:07:52.800 --> 00:07:53.120
Great.
00:07:53.279 --> 00:07:54.720
What are we looking for in stage two?
00:07:54.959 --> 00:08:01.360
What are we looking for in stage three that lead us up to ultimately something like ARR per employee?
00:08:02.319 --> 00:08:09.519
As you go through these different stages, I think your KPIs and your metrics should be adapted for those particular use cases for sure.
00:08:09.680 --> 00:08:12.879
And so there isn't a golden formula that you can apply there.
00:08:13.040 --> 00:08:15.759
But I would look at it from a use case-to-use case basis.
00:08:15.920 --> 00:08:27.920
And I think for employees that use and adopt these tool sets, I think one of the critical things organizations can do is provide rails, provide that common functionality that you can leverage to get on.
00:08:28.000 --> 00:08:34.639
And so what that does is will then empower individuals to move naturally matriculate from one stage to another.
00:08:34.879 --> 00:08:41.039
Without those rails, I think you're going to feel organizations are stuck and you'll see that they spin in governance.
00:08:41.120 --> 00:08:46.080
And so they'll make it to governance, they'll sit there and they'll review the same use cases over and over again.
00:08:46.320 --> 00:09:00.720
Maybe they'll sprinkle AI in there, some summarization, some basic summarization is in there, some basic transformation is in there, but you fundamentally haven't really disrupted that workflow and you haven't really asked yourself, I have four people managing that workflow today.
00:09:00.879 --> 00:09:01.840
Do I really need four?
00:09:02.000 --> 00:09:03.039
Can I have one?
00:09:03.279 --> 00:09:05.519
That's when you're actually making change.
00:09:05.600 --> 00:09:09.759
And you'll naturally you'll see that matriculate into your metrics and your AR.
00:09:10.480 --> 00:09:13.279
I just read recently about this company.
00:09:13.440 --> 00:09:43.440
I'm not gonna name names, but they saw some red flags in kind of their customer acquisition costs and decided kind of quickly that they were going to eliminate, I would assume was about 90 to 95% of their marketing organization to give themselves a chance to stop and rethink their processes and figure out how AI can assist or handle the vast majority of their processes.
00:09:43.919 --> 00:09:51.440
Do you think that is going to be a necessary step for a lot of organizations to really take it to one of the later stages that you're talking about?
00:09:51.600 --> 00:09:58.320
Or is there maybe a happy medium somewhere along that path where they don't have to take such drastic measures?
00:09:58.720 --> 00:10:02.000
I think those drastic measures are vastly performative.
00:10:02.080 --> 00:10:06.159
You know, I think that's there for the market to hear.
00:10:06.320 --> 00:10:14.559
What we have done is somehow mistaken the size of organizations for revenue that an organization can generate for the value they can produce.
00:10:14.639 --> 00:10:16.080
And you see that littered everywhere.
00:10:16.240 --> 00:10:20.159
For example, Snapchat still has thousand people plus for that app.
00:10:20.320 --> 00:10:21.279
I'm not really sure why.
00:10:21.440 --> 00:10:24.480
They've built the app, they've deployed it, they've built the models.
00:10:24.639 --> 00:10:26.960
I mean, you have the face filters, so on and so forth.
00:10:27.120 --> 00:10:28.320
I mean, I love Snapchat.
00:10:28.480 --> 00:10:31.200
That was actually the frontier of AI, your Snapchat filters.
00:10:31.279 --> 00:10:32.639
Let's not, let's not kid ourselves.
00:10:32.799 --> 00:10:35.759
But do you still need a thousand plus organization to manage that?
00:10:35.840 --> 00:10:36.639
I'm not really sure.
00:10:36.720 --> 00:10:41.840
And conversely, you have perplexity, you know, that came out in something like a 12-person organization.
00:10:42.000 --> 00:10:43.679
You know, why is that the case?
00:10:43.840 --> 00:10:46.559
And so there's a drastic gap there, you know.
00:10:46.639 --> 00:10:52.799
And if you look at that, I think what you're gonna see is the term AI native is often thrown around.
00:10:52.960 --> 00:10:56.960
AI native isn't someone that's using GPT and cloud right now.
00:10:57.200 --> 00:11:02.559
It's AI native is someone that's built their overall workflows around these tool sets.
00:11:02.720 --> 00:11:04.799
And that's how you can see those leverage.
00:11:04.960 --> 00:11:09.279
And so for someone to announce, hey, I'm gonna cut my marketing team by 99%, sure.
00:11:09.440 --> 00:11:11.120
I mean, that seems a little performative to me.
00:11:11.200 --> 00:11:15.039
I'd like to see the metrics and behind the scenes of what's actually going on.
00:11:15.200 --> 00:11:23.200
Um, but I think if you do follow that matriculation curve, snap the line and figure out where you are as an organization, first and foremost.
00:11:23.440 --> 00:11:27.279
The first order of magnitude is understand the complexity of your workflows.
00:11:27.360 --> 00:11:29.600
Do you even understand your workflows?
00:11:29.759 --> 00:11:31.039
Are they documented?
00:11:31.200 --> 00:11:32.960
Or are they institutional knowledge?
00:11:33.120 --> 00:11:34.480
Are they tribal knowledge?
00:11:34.639 --> 00:11:36.240
Do they sit in Excel spreadsheets?
00:11:36.480 --> 00:11:40.399
Do they sit in access databases and various legacy platforms?
00:11:40.480 --> 00:11:41.919
And I bet you it's the latter.
00:11:42.080 --> 00:11:44.879
It's scattered and fragmented all across the organization.
00:11:45.120 --> 00:11:49.600
It's never that, you know, 10 people in your team are actually executing.
00:11:49.759 --> 00:11:52.080
There's two of the 10 that are executing.
00:11:52.159 --> 00:11:53.679
There's eight sidewalk engineers.
00:11:53.759 --> 00:11:56.080
That's very common in larger organizations.
00:11:56.320 --> 00:11:56.799
You see that.
00:11:56.960 --> 00:12:00.000
And so in startups, you just don't have that luxury.
00:12:00.159 --> 00:12:03.919
You have to hire those two people that are keeping you, keeping you afloat, right?
00:12:04.000 --> 00:12:05.279
They're actually doing the work.
00:12:05.440 --> 00:12:18.799
But once you understand your workflows, document your workflows, then going back to the drawing board, reimagining them with these unique tool sets that's available, that's gonna write off the bat give you an understanding of what's residual.
00:12:18.960 --> 00:12:22.320
Then it's not a conversation of I'm gonna get rid of XYZ.
00:12:22.399 --> 00:12:24.639
It's well, how can I adapt these individuals?
00:12:24.879 --> 00:12:27.360
What other new roles do I have in this new world?
00:12:27.600 --> 00:12:30.879
Can I perhaps train them and move them over to that particular role?
00:12:31.120 --> 00:12:32.879
Are they fit for that role?
00:12:33.120 --> 00:12:34.960
Et cetera, having that conversation.
00:12:35.120 --> 00:12:47.919
And when you hear things of that sort, if you hear organizations, leaders talking in that language, I think more than less, you're gonna see a natural progression toward the native AI spectrum.
00:12:48.080 --> 00:12:53.759
But if you hear right off the bat, I'm gonna cut XYZ because we're dealing with agent development, I'd be cautious.
00:12:53.840 --> 00:12:55.200
I want to see the proof there.
00:12:55.440 --> 00:12:56.960
I like that perspective.
00:12:57.200 --> 00:13:06.960
I'm curious to hear your thoughts on the difference in the role that a CTO versus a CIO plays in AI transformation within a large organization.
00:13:07.200 --> 00:13:08.879
Are you guys thinking about the same things?
00:13:09.039 --> 00:13:10.240
Do you have the same goals?
00:13:10.480 --> 00:13:13.120
Is it a balance of one versus the other?
00:13:13.519 --> 00:13:18.960
I've worked in small, medium, large-size firms, startup, all of those in different organizations.
00:13:19.279 --> 00:13:20.879
That answer is going to change.
00:13:21.039 --> 00:13:21.360
Okay.
00:13:21.679 --> 00:13:28.320
In a larger organization, you're gonna have to have line of sight with respect to business master data.
00:13:28.480 --> 00:13:32.000
What are you doing with that master data for that particular line of business?
00:13:32.159 --> 00:13:33.679
How are you managing this?
00:13:33.919 --> 00:13:37.759
How are you solving customer problems for that business unit?
00:13:38.000 --> 00:13:43.279
So when you have to have that level of focus, I think a CIO role, that's their purview.
00:13:43.360 --> 00:13:44.399
And that's what they're looking at.
00:13:44.559 --> 00:13:46.639
The CTO role is an enabler role.
00:13:46.720 --> 00:13:56.080
You're making sure platforms are up and running, that irrespective of which CIO and which line of business, they can take advantage of those platform capabilities.
00:13:56.320 --> 00:13:58.480
What capability are you building today?
00:13:58.639 --> 00:14:05.360
How are you investing your firm's dollars in order to unblock fundamental capability down the road?
00:14:05.519 --> 00:14:09.840
Those are the things that you're managing and you're triaging on a day-to-day.
00:14:10.080 --> 00:14:20.879
And it's interesting, in my particular role, not only do I play a CTO role and manage that foundational capability, I also have the cyber role, the CISO role underneath me as well.
00:14:20.960 --> 00:14:29.039
And so I have to manage from a CISO perspective and a cyber perspective, not only do you push this capability out, how do I secure it?
00:14:29.200 --> 00:14:34.960
How do I keep up with the needs of my customers and the expectations that the market sets?
00:14:35.120 --> 00:14:39.279
And how do I make sure I put Rails in place for my organization?
00:14:39.519 --> 00:14:42.879
Micah, it reminds me of back in the day when we used to do mobile.
00:14:43.039 --> 00:14:49.759
We used to try to design these mobile sites and try to move from your web platform and your web interface to a mobile interface.
00:14:50.080 --> 00:14:59.840
iOS, Android, et cetera, Microsoft would try to be prescriptive with respect to how users would work in these platforms, how they would interact with these platforms.
00:15:00.000 --> 00:15:03.039
And we would try to prioritize features and functionality that way.
00:15:03.120 --> 00:15:12.159
But one of the things that we would always have to keep up with was the guy next door or a different app that already is pushing out this functionality.
00:15:12.320 --> 00:15:18.080
Why would I upload my driver's license when this guy just allows you to take a picture of it and digest it right off the bat?
00:15:18.159 --> 00:15:22.559
And so your customer expectation is set by other firms that are pushing the envelope.
00:15:22.720 --> 00:15:25.600
If you're not keeping up with them, they're going to find a way.
00:15:25.759 --> 00:15:41.679
And how I see that in the day-to-day is if I don't provide document intelligence, if I don't provide ways to automate, if I don't provide a community for my developers inside in order to perform this way, people will find a way.
00:15:41.840 --> 00:15:43.759
People will find other AI tools to use.
00:15:43.840 --> 00:15:49.519
They'll take pictures of your screen, send it to GPT, public GPT to do XYZ, come back.
00:15:49.600 --> 00:15:51.120
Your data has already left.
00:15:51.360 --> 00:15:52.720
Water will find a way, right?
00:15:52.879 --> 00:16:02.000
And so we have to put these rails and foundational capability in place in order to help and nudge people toward good behavior and incentivize that good behavior.
00:16:02.159 --> 00:16:08.000
But come back to your question, line of business, am I making sure that that data is clean?
00:16:08.159 --> 00:16:10.639
You know, can I get to my AI model better?
00:16:10.799 --> 00:16:14.320
Am I automating tasks and thinking about the workflows in my business line?
00:16:14.480 --> 00:16:17.440
That's a CIO's role in my eyes, the CTO role.
00:16:17.519 --> 00:16:23.200
It's an enabler broad, making sure all that capability is up and running and secure for adoption.
00:16:23.519 --> 00:16:29.440
I think it's incredibly interesting that you're owning both the technology and kind of the security side of things.
00:16:29.759 --> 00:16:33.679
How has that changed your perspectives on architecture?
00:16:33.840 --> 00:16:35.679
Because you've got a lot coming together here.
00:16:35.840 --> 00:16:42.320
There's the adoption of AI, there's the security aspect, but then there's the interoperability of the things that we're building.
00:16:42.559 --> 00:16:44.879
And so I've been building apps.
00:16:45.039 --> 00:16:49.840
I'm adding MCP capabilities for every single function that you can do in the UI.
00:16:50.080 --> 00:16:55.120
Because to me, that makes sense for an agent to be able to use the app the same way that a user can.
00:16:55.440 --> 00:17:01.360
How have you been challenged in the way you're thinking about architecture and working with your teams and the things that you're building?
00:17:01.519 --> 00:17:04.559
I know that's getting very specific all of a sudden, but No, no, no.
00:17:04.640 --> 00:17:05.680
I I love that question.
00:17:05.920 --> 00:17:07.599
So my background is really broad.
00:17:07.680 --> 00:17:11.920
And so I wrote front-end code for a long time, as soon as I got out of school.
00:17:12.160 --> 00:17:15.440
I left that and moved to a different team to write back end code.
00:17:15.519 --> 00:17:19.039
And I wrote a lot of crappy code for a long time before I wrote good code.
00:17:19.200 --> 00:17:24.960
And then I designed crappy systems for a while before I learned to do good systems and support these things long term.
00:17:25.039 --> 00:17:30.160
And every single turn, I learned a lesson and I put that lesson in the back and so on and so forth.
00:17:30.240 --> 00:17:36.960
And as I built this knowledge base, my internal knowledge base, my internal vector database that's inside me, right?
00:17:37.119 --> 00:17:43.039
As I went up, what I realized and it stood true was that architectural principle, keep it simple, stupid.
00:17:43.119 --> 00:17:44.960
You know, that is very, very true.
00:17:45.039 --> 00:17:53.920
And so if I can hold the architecture in my brain, if I can reduce it, if I can simplify it, chances are I can secure it properly.
00:17:54.079 --> 00:17:56.000
That's a core, core critical principle.
00:17:56.160 --> 00:18:06.240
And so every day my LinkedIn is flooded with vendors that want me to adopt AI products, security products, you name it, all over the place, free demos, this, that, and the other thing.
00:18:06.400 --> 00:18:08.000
Um, you have to resist.
00:18:08.079 --> 00:18:12.960
And I constantly see over-engineering of the overall stack.
00:18:13.119 --> 00:18:13.359
Okay.
00:18:13.759 --> 00:18:16.799
And I think where we're going is convergence.
00:18:16.880 --> 00:18:21.279
There's a great convergent event in the future over a horizon.
00:18:21.440 --> 00:18:26.319
And so we split up these roles, whether you're a CIO or a CTO or a CISO.
00:18:26.640 --> 00:18:35.680
Going in the future, especially with the gen TIC in mind, with delegation, with understanding context, one person can converge and manage all of this.
00:18:35.839 --> 00:18:40.559
The reason why we have that fragmentation inside of an organization is simply because information, right?
00:18:40.720 --> 00:18:42.880
One person doesn't have all that information.
00:18:43.119 --> 00:18:43.839
But think about it.
00:18:44.000 --> 00:18:45.200
If what if I did?
00:18:45.440 --> 00:18:48.640
What if I can scan every single log that's out there?
00:18:48.799 --> 00:18:59.839
If I understand what's happening in the edge of my network, if I understand all the holes and every single open source software that I have installed in my application, that's a tremendous amount of knowledge.
00:18:59.920 --> 00:19:05.839
I can discover vulnerabilities better, I can patch it better, I can do mitigation better, I can write software better.
00:19:06.000 --> 00:19:11.599
And then if I know your behavior itself, Micah, when you log in, what when do you drop your kids off and log in?
00:19:11.680 --> 00:19:13.200
And what's the first thing Micah does?
00:19:13.359 --> 00:19:13.759
Guess what?
00:19:13.839 --> 00:19:16.400
I can probably do a brief for you and just text it to you.
00:19:16.480 --> 00:19:17.440
Why even log in?
00:19:17.680 --> 00:19:19.440
How we think about work today.
00:19:19.599 --> 00:19:25.920
And so I think as we get more and more information, I think that these roles are very fluid and they change.
00:19:26.160 --> 00:19:36.480
I think keeping it simple, understanding and having a good sense of caution of here's a new model that's coming out, here's what it's promised to do, here's what it actually can do.
00:19:36.640 --> 00:19:38.640
What does a benchmark actually tell you?
00:19:38.799 --> 00:19:40.720
You know, understanding and asking that question.
00:19:40.799 --> 00:19:41.599
I think that's critical.
00:19:42.240 --> 00:19:52.720
But keeping your app stack very simple, keeping your cyber stack simple, making sure like workflows are simplified, and making sure you manage risk.
00:19:52.880 --> 00:19:54.240
And I think those things help.
00:19:54.400 --> 00:19:58.559
And that's how my CISO role influences my CTL role, vice versa.
00:19:58.640 --> 00:20:00.559
And my architecture brain comes in.
00:20:00.720 --> 00:20:05.759
And I really fight to make sure I don't over-engineer any problem.
00:20:06.079 --> 00:20:18.319
Even when I over-engineer, I put it down, go for a walk, go for a run, come back, and I try to attack it with this oversimplistic view and try to plead a devil's advocate for myself and say, you know, why can't this work?
00:20:18.480 --> 00:20:19.920
Why do I need A through Z?
00:20:20.000 --> 00:20:21.599
Why can't just ABC work?
00:20:21.759 --> 00:20:28.319
And so that's helped me manage my own context and make sure that I provide simpler solutions.
00:20:29.440 --> 00:20:36.160
Kind of on that note, you said a little bit ago that your inbox is flooded with vendors.
00:20:36.400 --> 00:20:36.640
Yes.
00:20:36.799 --> 00:20:40.079
And you have to resist and you are trying to keep things simple.
00:20:40.240 --> 00:20:45.680
So what is your perspective or your approach for build versus buy in this case?
00:20:46.000 --> 00:20:47.359
Oh, that's a great, great question.
00:20:47.759 --> 00:20:49.039
That's an architecture question.
00:20:49.119 --> 00:20:50.799
I feel like I'm being interviewed now.
00:20:50.960 --> 00:20:55.519
Um it always comes back to the business outcome.
00:20:55.680 --> 00:21:00.640
And so back in the day, we used to say, hey, we're not a tech company.
00:21:00.799 --> 00:21:04.160
I'm a financial services, I'm making it up, financial services company.
00:21:04.240 --> 00:21:06.640
I should work on this and leave the tech to the tech.
00:21:06.960 --> 00:21:08.720
Somehow something changed.
00:21:08.960 --> 00:21:13.519
Gardner probably put out talking points for CIOs, and everyone became tech companies.
00:21:13.680 --> 00:21:18.960
Didn't matter if you were a Dollar General or if you worked at Target, suddenly you're a tech firm.
00:21:19.119 --> 00:21:23.359
But how I look at it is I have a finite number of resources to my feel.
00:21:23.519 --> 00:21:30.319
I already have systems and capability that vendors build out for me or providers build out for me.
00:21:30.559 --> 00:21:33.279
And I use the architectural principle cut with the grain.
00:21:33.359 --> 00:21:35.839
You know, what can I compose with?
00:21:36.079 --> 00:21:41.440
And so I think the greatest thing that we do is I don't have to reinvent the wheel.
00:21:41.680 --> 00:21:45.200
I can go back and I can look at what capability do you have?
00:21:45.440 --> 00:21:47.680
Do you do very well today?
00:21:47.920 --> 00:21:50.960
And what can I get for the fraction of cost?
00:21:51.119 --> 00:21:52.960
I look for uni cost per that transaction.
00:21:53.039 --> 00:21:59.680
And so if I can slice and dice that and get that as low as possible for a good amount of value, sure, I'll I'll buy.
00:22:00.160 --> 00:22:01.599
I'll use that capability.
00:22:01.680 --> 00:22:03.680
I'll compose that into my workflow.
00:22:03.759 --> 00:22:10.480
I'll utilize that capability, move on until my workflow again transforms into something else.
00:22:10.559 --> 00:22:12.400
And then I'll look for a different solution.
00:22:12.720 --> 00:22:25.920
Only when there is a competitive advantage for my particular domain, for my company, where the vendor or the provider doesn't offer that, will I invest my own dollar in building that capability out?
00:22:26.160 --> 00:22:29.759
Otherwise, I'm pretty much into composing and orchestrating.
00:22:30.160 --> 00:22:30.480
Okay.
00:22:30.640 --> 00:22:34.720
So you're in the middle of AI transformation with your current organization.
00:22:34.880 --> 00:22:38.960
From what I understand, it's very safety focused, highly regulated.
00:22:39.759 --> 00:22:43.599
And that I assume creates a lot of complexity.
00:22:44.079 --> 00:22:49.599
So how did you decide or how have you made decisions on where to start and where to move next?
00:22:49.920 --> 00:22:51.039
Oh, that's a good question.
00:22:51.119 --> 00:22:54.400
I think it comes back to your operating foundation.
00:22:54.480 --> 00:22:58.640
And so even working at RPG, RPG is a heavily regulated industry.
00:22:58.880 --> 00:23:00.640
We're safety in critical industries.
00:23:00.720 --> 00:23:02.480
And so we're safety first, of course.
00:23:02.640 --> 00:23:09.440
And so one of my most important lessons here that I have learned is actually it didn't start with AI.
00:23:09.519 --> 00:23:10.960
I didn't start with the AI focus.
00:23:11.039 --> 00:23:13.759
I didn't start with implementing AI workflows.
00:23:13.920 --> 00:23:16.559
I actually started with the operating foundation.
00:23:16.640 --> 00:23:21.599
And so in a real enterprise, work rarely lives, like I said, inside of an application.
00:23:21.920 --> 00:23:30.079
What I found is that it lives with a lot of tribal knowledge, legacy platforms, sometimes hidden in systems that are old, that are fatigued.
00:23:30.319 --> 00:23:36.079
I found so many that were just sitting in people's workstations that were running and customer requirements were just fragmented.
00:23:36.160 --> 00:23:40.400
And so trying to go and collate and bring that all together, I think that was step one.
00:23:40.559 --> 00:23:50.880
And one of the other things that I pride myself doing is when I would meet with my stakeholders and do my initial round table, going around the organization trying to understand the problems that they have.
00:23:51.039 --> 00:23:52.480
Active listening actually helped.
00:23:52.640 --> 00:23:55.200
Just tell me what you're seeing day to day.
00:23:55.359 --> 00:23:57.279
What kind of experiences do you have, et cetera?
00:23:57.440 --> 00:24:01.599
And I would transcribe it, I would note it, I would go through it, refine it, so on and so forth.
00:24:01.680 --> 00:24:04.559
What is this individual actually trying to tell me?
00:24:04.720 --> 00:24:06.000
What are their pain points?
00:24:06.160 --> 00:24:07.039
I would take that.
00:24:07.200 --> 00:24:11.279
And that gave me a good understanding of how the organization was operating.
00:24:11.440 --> 00:24:11.759
Okay.
00:24:12.000 --> 00:24:22.880
And so whether it's a financial services company or a company such as RPG that works in a heavily regulated and a safety-focused industry, I think that's a common denominator.
00:24:23.039 --> 00:24:27.680
And so understanding how you operate, how you do your day-to-day is critical.
00:24:27.920 --> 00:24:37.920
And what often surprises me, and even the most advanced organizations, how many manual steps and human intervention is required for existing workflows.
00:24:38.079 --> 00:24:38.960
So that's really critical.
00:24:39.039 --> 00:24:40.160
And that opens my eyes.
00:24:40.240 --> 00:24:41.599
And then I go to simplify it.
00:24:41.680 --> 00:24:43.359
That's when I go and simplify it.
00:24:43.599 --> 00:24:49.920
I'm a very minimalist person, so I try to throw away all the extraneous stuff and reduce it to what it actually needs.
00:24:50.079 --> 00:24:53.920
Once I do that, I try to digitize the first aspect of it.
00:24:54.079 --> 00:24:54.400
Okay.
00:24:54.640 --> 00:25:01.039
Let me get and bring all that together and to provide a consolidated, structured data representation.
00:25:01.200 --> 00:25:02.400
Let me put that together.
00:25:02.640 --> 00:25:10.720
Once you have that, you have the basic foundations of doing any type of intelligent automation, data intelligence on top of that.
00:25:10.960 --> 00:25:14.720
Then of course, I look for deterministic processes.
00:25:14.880 --> 00:25:15.839
What's formulaic?
00:25:16.079 --> 00:25:17.839
What's A plus the B equals C?
00:25:18.000 --> 00:25:20.480
And I write that down and I take a jot of that.
00:25:20.720 --> 00:25:22.319
That's the overall wheel.
00:25:22.400 --> 00:25:23.920
You don't want to reinvent that wheel.
00:25:24.000 --> 00:25:24.640
That's working.
00:25:24.720 --> 00:25:26.160
I get that for a good cost.
00:25:26.319 --> 00:25:28.319
Again, it goes back to economics.
00:25:28.720 --> 00:25:31.440
That's a very good unit per transaction capability.
00:25:31.599 --> 00:25:32.960
And so I note that down.
00:25:33.279 --> 00:25:39.440
And then I start redesigning my process from the ground up with AI and any of the latest tool sets that I have.
00:25:39.680 --> 00:25:45.519
I use deterministic processes first, and then I sprinkle probabilistic toward the end.
00:25:45.599 --> 00:25:49.599
And so that pattern seemed to work out a lot.
00:25:49.759 --> 00:26:00.559
And so AI is really powerful when ambiguity exists, but traditional software and deterministic software is still superior in in most environments.
00:26:00.640 --> 00:26:05.920
And so I try to use that good balance and try to manage out when I'm trying to automate internally.
00:26:06.240 --> 00:26:17.759
I like that perspective and I like the sequencing there because you talked about I'm going to stick to deterministic where I can and only use AI where I need to for the probabilistic intelligence, right?
00:26:17.920 --> 00:26:32.160
I think a lot of organizations will sometimes just sprinkle AI everywhere and then all of a sudden costs go up because they're applying probabilistic logic where deterministic logic would have served perfectly fine.
00:26:32.400 --> 00:26:38.000
And I think that's a good segue into a quote that I'm going to pull from your Substack.
00:26:38.160 --> 00:26:46.880
You said that technology is essentially organizational design rendered in software, and that software doesn't solve complexity, it preserves it.
00:26:47.039 --> 00:26:56.880
So for the organizations that are sprinkling AI everywhere on top of broken processes, what ultimately happens and how would you guide or coach them?
00:26:57.279 --> 00:26:58.000
No, that's good.
00:26:58.079 --> 00:26:59.359
I forgot I wrote that actually.
00:26:59.519 --> 00:27:02.960
I had to go back to my Substack and read, and I'm like, oh man, that was profound.
00:27:03.119 --> 00:27:09.200
Um but when I said that back, I was seeing Conway's Law in effect all the time.
00:27:09.440 --> 00:27:15.039
I had one of these moments when I'm sitting in a meeting and I rose above myself and I saw the silos.
00:27:15.440 --> 00:27:18.559
And when I looked at the code like Neo, I saw the silos again.
00:27:18.640 --> 00:27:21.519
And so that's kind of how I came up with that whole thing.
00:27:21.680 --> 00:27:25.359
But when I said technology is organizational design rendered in software, I meant it.
00:27:25.519 --> 00:27:31.279
What I mean by that is really that software usually reflects how an organization is functioning underneath.
00:27:31.359 --> 00:27:47.119
The approvals, the silos, the handoffs, any controls that you have, even the politics that remains, the reporting relationships, the organization, and even historical baggage, they all eventually become encoded in the actual technology.
00:27:47.200 --> 00:27:49.039
And you can see it plain as day.
00:27:49.599 --> 00:27:56.559
And people will look at the tech back and then they look at those complicated systems and then they say this the software is too complex.
00:27:56.720 --> 00:27:59.680
But often the software did not create that complexity.
00:27:59.759 --> 00:28:00.480
It's preserved.
00:28:00.640 --> 00:28:03.839
That complexity already existed in that organization.
00:28:04.000 --> 00:28:05.759
People created that complexity.
00:28:06.000 --> 00:28:13.920
And so this is why simply taking like an AI tool and layering that on top of your existing systems is not very fruitful.
00:28:14.079 --> 00:28:16.319
And in fact, it can often be dangerous.
00:28:16.480 --> 00:28:24.720
And so if you don't challenge the underlying process, you can end up like automating corporate dysfunctions rather than actually transforming the company.
00:28:24.960 --> 00:28:35.119
Imagine like a factory where you have a physical layout that's poorly designed and materials cross the factory floor 10 times, 100 times unnecessarily.
00:28:35.359 --> 00:28:40.640
Using robots afterwards to actually move that freight back and forth, does that do anything?
00:28:40.799 --> 00:28:43.519
Yes, it does something, but are you actually fixing anything?
00:28:43.759 --> 00:28:44.240
Not really.
00:28:44.400 --> 00:28:48.079
You've just created robots that can go back and forth maturely faster.
00:28:48.400 --> 00:28:49.599
The mess happens faster now.
00:28:49.680 --> 00:28:49.839
Yeah.
00:28:50.079 --> 00:28:50.720
That's exactly right.
00:28:50.799 --> 00:28:54.720
It's like my my room boat when I go away and I come back to the mess that is created.
00:28:54.799 --> 00:28:56.079
It's very similar.
00:28:56.480 --> 00:29:00.640
And so AI doesn't absolve the leaders for sure from process design.
00:29:00.720 --> 00:29:02.000
You absolutely have to do that.
00:29:02.160 --> 00:29:07.599
But in some ways, what it does is it incentivizes good process design even more.
00:29:07.680 --> 00:29:13.839
Because what happens is AI allows both good and bad practices to scale more quickly.
00:29:14.000 --> 00:29:20.799
And so, in essence, AI basically ends up amplifying the quality of the systems underneath it.
00:29:20.880 --> 00:29:27.680
And so what that means is software, again, it preserves your organizational complexity and whatever existed there.
00:29:27.839 --> 00:29:31.039
Uh, when you put AI on top of it, it just scales it.
00:29:31.200 --> 00:29:42.240
And so before you can actually um accelerate and introduce AI into your processes, you should really ask yourself which processes deserve to live on their own?
00:29:42.400 --> 00:29:47.200
And are they at the most simplified form first before you uh introduce any AI?
00:29:47.279 --> 00:29:52.000
But that's what I really meant when I wrote that particular piece of that article.
00:29:52.400 --> 00:29:53.759
I like what you said there.
00:29:53.920 --> 00:29:59.920
For a lot of organizations, they've been on this digital transformation journey for 25 years now, right?
00:30:00.000 --> 00:30:01.920
And I don't think that journey ever really ended.
00:30:02.000 --> 00:30:04.640
It was kind of like, well, we were doing digital transformation.
00:30:04.799 --> 00:30:09.680
This leader came in, brought this workflow app, this leader came in, brought this case management app.
00:30:09.839 --> 00:30:11.839
Those people got a bonus and took off.
00:30:12.000 --> 00:30:20.319
And so they're left with this really fragmented architecture for the way that workflows through their environments.
00:30:20.799 --> 00:30:23.599
Are they able to use AI and where?
00:30:23.759 --> 00:30:37.920
Or do they need to step back and say, we need to redesign our entire workflows first and orient on one workflow application or one case management application, like we said we were going to do 15 years ago before we can start to adopt AI?
00:30:38.079 --> 00:30:39.359
How do you think of that?
00:30:39.759 --> 00:30:41.039
No, that's a good question.
00:30:41.200 --> 00:30:47.680
The question that you're asking is do you have to do an extensive six-month process engineering exercise?
00:30:48.000 --> 00:30:50.240
Can I um adopt something now?
00:30:50.319 --> 00:30:54.799
And so if we make that a prerequisite for every use case, I think you'll be blown out of the water.
00:30:54.880 --> 00:30:56.720
Everyone will innovate a lot faster than you.
00:30:56.799 --> 00:31:00.720
And most organizations will move too slowly and people will start working around them.
00:31:00.799 --> 00:31:02.799
You know, like I stated, water will find a way.
00:31:03.039 --> 00:31:06.799
But in some levels, you do have to have some level of discipline.
00:31:06.960 --> 00:31:15.359
And so I keep going back to before you do apply AI to some of this capability, I think you should fundamentally ask yourself a couple of questions.
00:31:15.519 --> 00:31:17.759
What outcomes are you actually trying to produce?
00:31:17.920 --> 00:31:22.400
Not what processes do you actually have, what results actually matter?
00:31:22.559 --> 00:31:23.839
I think that's really critical.
00:31:24.079 --> 00:31:26.480
What decisions happen along the way?
00:31:26.640 --> 00:31:30.160
And so AI frequently becomes useful at decision boundaries.
00:31:30.319 --> 00:31:36.319
Put somebody in the middle of an interpretation, in the middle of a classification, a prioritization.
00:31:36.559 --> 00:31:38.799
What actions do you recommend after the fact?
00:31:38.880 --> 00:31:39.599
That's really critical.
00:31:39.759 --> 00:31:42.079
And so again, what decisions happen along the way?
00:31:42.240 --> 00:31:43.359
Ask yourself that.
00:31:43.599 --> 00:31:47.359
What information or data is required to make those decisions?
00:31:47.519 --> 00:31:54.319
And so information is scattered and fragmented and often and resides in employees' brain.
00:31:54.480 --> 00:31:55.599
It's tribal knowledge.
00:31:55.759 --> 00:31:58.079
And so it's contradictory.
00:31:58.319 --> 00:32:00.240
It doesn't have any authoritative source.
00:32:00.400 --> 00:32:01.920
There is no master data sometimes.
00:32:02.000 --> 00:32:06.960
And so that is usually a data and architectural problem before it actually becomes an AI problem.
00:32:07.119 --> 00:32:09.599
And so that's a critical question you have to ask yourself.
00:32:09.839 --> 00:32:12.240
And what controls do we need?
00:32:12.319 --> 00:32:16.480
You know, what controls are absolutely mandatory and what is non-negotiable?
00:32:16.640 --> 00:32:27.839
And so that includes when you start touching domains like authorization, security, compliance, segregation of duties, privacy, auditability, and even accountability.
00:32:28.000 --> 00:32:32.960
And then I would look at what old constraints still have to be applied.
00:32:33.119 --> 00:32:34.559
Do they have to be applied?
00:32:34.799 --> 00:32:39.200
Maybe a step exists because two systems could not communicate before.
00:32:39.359 --> 00:32:42.319
Maybe now you have a better API or a better protocol.
00:32:42.559 --> 00:32:43.839
Do I still need that step?
00:32:44.000 --> 00:32:49.119
And maybe an approval exists because we didn't have that visibility before.
00:32:49.279 --> 00:32:49.839
Now we do.
00:32:50.000 --> 00:32:52.160
And so do I absolutely need that step?
00:32:52.319 --> 00:32:53.680
And so that's really critical.
00:32:53.839 --> 00:33:02.160
And so some manual reeking the data for integration sometimes will be built just because they didn't have that integration piece.
00:33:02.319 --> 00:33:03.759
Now you have that integration piece.
00:33:03.920 --> 00:33:05.039
Do you still need to do that?
00:33:05.119 --> 00:33:07.359
And so that's a critical question you have to ask yourself.
00:33:07.519 --> 00:33:12.960
The objective is really not to document existing processes with preciseness, really.
00:33:13.119 --> 00:33:16.319
It's really to understand it well enough to challenge it.
00:33:16.400 --> 00:33:17.680
And that's a critical component.
00:33:17.759 --> 00:33:22.319
And so don't have to automate the full-blown process map.
00:33:22.559 --> 00:33:28.079
The opportunity with AI tools today is really, well, how do I challenge that process map?
00:33:28.160 --> 00:33:29.119
So that's really critical.
00:33:29.200 --> 00:33:36.960
And so I would say don't spend six months documenting your processes that you're you're gonna spend the next two weeks eliminating.
00:33:37.440 --> 00:33:44.880
Dennis, what's something about AI transformation that you think most people, most orgs have not figured out yet?
00:33:45.039 --> 00:33:50.160
Something that you're watching or thinking about, something that's sitting in the back of your mind, bugging you.
00:33:50.559 --> 00:33:51.839
No, that's a good question.
00:33:52.160 --> 00:33:56.640
I think most people have applied it to simplistic workflows today.
00:33:56.880 --> 00:34:00.319
And I think there are two things that I don't think they fully appreciate.
00:34:00.480 --> 00:34:05.680
One, I think AI is likely to change your overall organizational structure, not just personal productivity.
00:34:05.839 --> 00:34:14.079
Like I said, you know, historically we have created these vertical hierarchies because one person does not have enough context, they can't digest all that information.
00:34:14.239 --> 00:34:15.760
Maybe they can, maybe they can't.
00:34:15.920 --> 00:34:17.920
And so we've created these verticalized layers.
00:34:18.079 --> 00:34:19.199
AI is gonna squish that.
00:34:19.280 --> 00:34:30.239
So many AI discussions assume that company remains essentially the same and everyone simply does become that 20% or 30% productive, again, against with the productivity suite.
00:34:30.480 --> 00:34:32.159
I'm not really convinced that's where it ends.
00:34:32.320 --> 00:34:35.360
I think it has to really attack your organizational structure.
00:34:35.599 --> 00:34:41.199
Organizations today and historically they exist to coordinate information and decisions.
00:34:41.360 --> 00:34:49.360
In fact, management layers exist specifically for that, just to make sure you address human limitations with respect to capacity and bandwidth.
00:34:49.519 --> 00:34:53.920
Now, as information moves up the organization, decision moves down.
00:34:54.079 --> 00:35:01.199
Managers monitor activity, they summarize what's happening, they resolve any exceptions, and they coordinate work across teams.
00:35:01.360 --> 00:35:03.119
Now, AI changes all that.
00:35:03.280 --> 00:35:09.760
Now I can look at that data, I can centralize that data, I can democratize that specialized knowledge.
00:35:09.920 --> 00:35:21.760
And so when systems can continue, summarize what's happening, they can monitor for exceptions, they can recommend next best actions, and they can even coordinate some of these activities and execute a routine playbook.
00:35:21.920 --> 00:35:26.400
The amount of information a person or team needs, they can absolutely manage that, right?
00:35:26.559 --> 00:35:30.079
And so what that will do is essentially lead to a flatter organization.
00:35:30.239 --> 00:35:37.760
And so you have new roles with wider spans of control, you have smaller, multidisciplinary teams, you have fewer handoffs.
00:35:37.840 --> 00:35:41.519
And so managers may eventually supervise systems of people.
00:35:41.760 --> 00:35:50.639
Today they manage people, but I think in the future, or even now, it's gonna start changing to managing agents that are supervising tasks.
00:35:50.800 --> 00:35:51.760
I think that's really critical.
00:35:51.840 --> 00:35:54.079
I don't think that people are paying enough attention to them.
00:35:54.320 --> 00:35:57.360
Now, that doesn't mean like leadership absolutely disappears.
00:35:57.599 --> 00:36:02.320
I think it more alludes to the fact that the nature of that relationship changes.
00:36:02.559 --> 00:36:07.199
Secondarily, I think execution becomes less expensive.
00:36:07.360 --> 00:36:10.639
However, I think judgment becomes a lot more valuable.
00:36:10.800 --> 00:36:25.440
And so AI can generate like the first draft for you, it can create optionality for you, it can even perform some analysis for you, it can even assemble and aggregate information for you, and it can even execute and delegate parts of your workflow and execute those.
00:36:25.599 --> 00:36:31.840
The differentiating human capability starts becoming well, am I solving the right problem?
00:36:32.079 --> 00:36:34.400
Am I considering the right context?
00:36:34.559 --> 00:36:36.559
Am I setting the right boundaries and guardrails?
00:36:36.880 --> 00:36:39.679
And am I evaluating the trade-offs properly?
00:36:39.840 --> 00:36:42.239
And so judgment is really critical.
00:36:42.400 --> 00:36:46.159
And fundamentally, of course, accepting accountability because AI can't do that.
00:36:46.320 --> 00:36:58.880
So talking about judgment and the guardrails that you talked about earlier, Brenda from Finance just got Claude Pro and she's vibe coding all kinds of apps and deploying them everywhere.
00:36:59.119 --> 00:37:01.679
What is her role in AI transformation?
00:37:01.840 --> 00:37:04.000
Is what she's doing destructive?
00:37:04.159 --> 00:37:06.480
Is it introducing more risk to the organization?
00:37:06.639 --> 00:37:08.320
Is it accelerating outcomes?
00:37:08.480 --> 00:37:09.760
How do you think about that?
00:37:10.000 --> 00:37:17.599
And how would you guide someone like Brenda who's excited about AI, but maybe building things that you don't want built?
00:37:17.920 --> 00:37:19.840
Shadow IT on steroids now.
00:37:20.079 --> 00:37:21.199
Yeah, no, absolutely.
00:37:21.360 --> 00:37:27.360
I think Brenda's probably the most important person in our AI transformation story.
00:37:27.519 --> 00:37:29.840
So for decades, it's really common, right?
00:37:29.920 --> 00:37:36.480
And so there's been a tremendous amount of gap between the person who understands the business problem and the person that builds software.
00:37:36.559 --> 00:37:39.920
So analysts have sat in the middle trying to merge both environments.
00:37:40.079 --> 00:37:43.039
Architects have sat and trying to merge both environments.
00:37:43.199 --> 00:37:46.400
But I think the power of someone like Brenda is they understand the work.
00:37:46.480 --> 00:37:47.280
She understands the work.
00:37:47.360 --> 00:37:54.320
She knows what frustrates her, where the process breaks, what people repeatedly enter, which reports take four hours to assemble.
00:37:54.480 --> 00:37:56.800
That is institutional knowledge that's sitting with her.
00:37:56.960 --> 00:38:00.239
Now, historically, she would have documented that problem.
00:38:00.400 --> 00:38:11.360
She send it over to technology, technology would pick it up, you would do a scrum planning session for hours and score it and all these things, and eventually, six months later, build something that doesn't solve the problem.
00:38:11.519 --> 00:38:13.039
You know, that's really common, right?
00:38:13.199 --> 00:38:18.480
And so uh AI, I think, is beginning to collapse and converge that chasm.
00:38:18.639 --> 00:38:26.159
And so Brenda can now turn around and take her understanding of the actual business problem that she's solving and create something that's functional.
00:38:26.239 --> 00:38:27.760
And so that's really incredibly powerful.
00:38:27.840 --> 00:38:29.280
I think that should be celebrated.
00:38:29.440 --> 00:38:33.199
And she's a critical piece of this transformation for any organization.
00:38:33.440 --> 00:38:55.360
However, the other side of that, the risk that she's entering into is that I don't think Brenda's probably cognizant of the fact that now she's not only a business um guru, uh, now she's a developer, she's a product owner, she's an architect, a security engineer, a data engineer, a QA person, a whole plethora of things that she was not assuming that she would be.
00:38:55.519 --> 00:38:59.760
And so assuming all those responsibilities gets very heavy very quick.
00:39:00.000 --> 00:39:01.519
But I don't think the problem is Brenda.
00:39:02.079 --> 00:39:10.639
I think the problem is whether the organization has created a governed path for what Brenda can build to become enterprise software.
00:39:10.800 --> 00:39:15.519
And so I tend to think that governance really here should scale with risk.
00:39:15.599 --> 00:39:24.719
So Brenda's Brenda's experimenting with something that's non-sensitive in a sandbox environment, no one should depend on those results, then you know, let it let it go.
00:39:24.880 --> 00:39:26.000
Let her experiment.
00:39:26.239 --> 00:39:27.599
Governance should be very light.
00:39:27.679 --> 00:39:31.679
And so she should be able to do many integrations at light speed.
00:39:31.920 --> 00:39:43.679
Now, when her team starts beginning to rely on critical data, the organization should allow Rails, should introduce testing, documentation, version control, support, those critical capabilities.
00:39:43.840 --> 00:39:55.920
If it touches production systems, PII data, customer information, so on and so forth, then it should have architecture, security controls, monitoring, all that capability that should be integrated for her.
00:39:56.239 --> 00:40:06.639
I think this is one of the areas where you look for something like a platform engineering team where they stand up that critical capability that the organization can rely on.
00:40:06.800 --> 00:40:11.280
They enable things like an MCP server where you can discover domain capability.
00:40:11.519 --> 00:40:19.199
They enable things like an API gateway or excuse me, an AI gateway where you can provision rate limiting, so on and so forth, things of that sort.
00:40:19.679 --> 00:40:26.639
Very analogous to the journey that we had back in the day when we moved from monolithic to distributed applications.
00:40:26.800 --> 00:40:29.360
It's essentially that same journey that's happening now.
00:40:29.679 --> 00:40:38.079
Put in those rails for Brenda, let Brenda leverage those existing patterns, enable her to get to the end zone a lot faster.
00:40:38.239 --> 00:40:46.239
And so my punchline there really is that I wouldn't put do anything to stop Brenda, but I would want 10 Brendas, 100 Brendas.
00:40:46.400 --> 00:40:53.599
I just need an architecture and platform that allows Brenda, people like Brenda to innovate safely.
00:40:54.000 --> 00:40:58.400
Well, we always like to end episodes with something actionable.
00:40:58.800 --> 00:41:05.920
So for someone who's listening to this right now and they want to do this right, where do they actually start?
00:41:06.079 --> 00:41:08.239
What resources, departments, concepts?
00:41:08.400 --> 00:41:09.519
What's your recommendation?
00:41:09.840 --> 00:41:13.760
I would start by creating a hundred-page enterprise AI strategy.
00:41:15.760 --> 00:41:19.119
No, and I'd be and hiring both of you to do the work, right?
00:41:19.199 --> 00:41:19.760
Absolutely, absolutely.
00:41:19.920 --> 00:41:20.880
You know, put that in there.
00:41:20.960 --> 00:41:35.440
Um, no, what I would actually do is I would tell them to pause and I would tell them to look actively listen inside their organization, look for things that their users say, things such as, you know, we do this manually every day.
00:41:35.599 --> 00:41:37.440
When only one person knows how this works.
00:41:37.599 --> 00:41:40.880
We copy this information from one system to another.
00:41:41.119 --> 00:41:44.320
Uh, we have to read every one of these documents.
00:41:44.480 --> 00:41:47.599
You know, there is all these triggers that you could look for.
00:41:47.679 --> 00:41:48.880
I would look for those.
00:41:49.039 --> 00:41:51.599
Those are signals, those are signals for opportunity.
00:41:51.760 --> 00:42:02.000
Then I would choose one particular workflow where the pain is significant, where you're impacting a lot of people, and where you can actually measure that pain.
00:42:02.159 --> 00:42:05.119
And the initial risk is also manageable.
00:42:05.280 --> 00:42:06.639
I think those are really critical.
00:42:06.800 --> 00:42:10.000
Pain high, value high, risk low.
00:42:10.159 --> 00:42:11.760
You know, I would look for that.
00:42:11.920 --> 00:42:19.440
I would dig in at that point, I understand that workflow from beginning to end and make sure you separate the work in three categories.
00:42:19.599 --> 00:42:23.760
I would look for deterministic capability and calculations there.
00:42:24.079 --> 00:42:27.280
What rules do you have in flight, validation rules?
00:42:27.519 --> 00:42:31.440
Do you have structured systems and transactions that can support that deterministic work?
00:42:31.599 --> 00:42:33.440
I would look to tease those out.
00:42:33.679 --> 00:42:35.760
I would look for cognitive work.
00:42:35.840 --> 00:42:42.000
And so things that are interpretive, things that you do with respect to summarization, classification, pattern recognition.
00:42:42.159 --> 00:42:43.280
That's also opportunity.
00:42:43.360 --> 00:42:45.440
I would look for that category of work as well.
00:42:45.599 --> 00:42:53.519
And then I would look for consequential work, things that are regulatory, contractual, safety, security, those are the third bucket.
00:42:53.760 --> 00:43:10.159
And then I would identify authoritative data, determine which system is the source for that information, map those out, and build one end-to-end simplified flow, not 20 different use cases, not 20 different disconnected paths.
00:43:10.320 --> 00:43:12.079
You know, take a simple workflow.
00:43:12.159 --> 00:43:21.840
Again, look for that signal, break it out into three different pieces, look to automate the deterministic aspects of it, look for deterministic capability to do that.
00:43:22.079 --> 00:43:30.000
Sprinkle AI at the tour, the last mile, just to make sure that you're interpretive and adding the charm that's necessary to communicate back to the user.
00:43:30.159 --> 00:43:38.880
And then make sure you add analytics to it so you can measure the business outcome, not in terms of its volume of use, in terms of am I reducing errors?
00:43:39.039 --> 00:43:41.039
Am I reducing those manual handoffs?
00:43:41.280 --> 00:43:47.840
Did that person that once had that institutional knowledge only can I survive when he goes on vacation, when he she goes on vacation?
00:43:47.920 --> 00:43:48.639
That's really critical.
00:43:48.800 --> 00:43:50.159
And so I would start there.
00:43:50.320 --> 00:43:55.119
I would leave the team by saying start with the actual work, not the model.
00:43:55.280 --> 00:43:57.519
That's kind of how I would lead into it.
00:43:57.840 --> 00:43:58.400
Very cool.
00:43:58.639 --> 00:44:01.519
Dennis, thank you so much for joining us on Agentic Edge.
00:44:01.599 --> 00:44:09.440
For anyone who wants to read more of Dennis's work, the Architect's Blueprint at Substack, you can check out more of his content there.
00:44:09.519 --> 00:44:10.719
We'll have that in the show notes.
00:44:10.880 --> 00:44:15.360
Dennis, thank you so much for joining us, and we'll catch you next time on Agenc Edge.
00:00:00.160 --> 00:00:10.880
Welcome to Agentic Edge, where we explore the frontier of AI agents, enterprise orchestration, and the architectures that are shaping tomorrow's intelligent enterprises.
00:00:11.039 --> 00:00:16.079
My name is Micah Smith, and joining me as always is my co-host, Kate Wrestler.
00:00:16.320 --> 00:00:20.000
And today we are diving into AI transformation.
00:00:20.160 --> 00:00:30.960
And it's one of those phrases that get used to describe absolutely everything from hey, we put a chat bot on our website to our company fundamentally has rewired how we operate.
00:00:31.199 --> 00:00:45.039
Organizations make big announcements about it, boards approve huge budgets for it, press releases go out, and then ultimately someone has to figure out what all of this means and how they can drive meaningful outcomes for the organization.
00:00:45.359 --> 00:00:50.240
To help us break all of that down, today's guest is Dennis Ganesh.
00:00:50.320 --> 00:01:01.840
He is CTO at RPG, and he's responsible for leading technology transformation in what's a very regulated and safety critical industry, which means he can't just fake his way through this.
00:01:02.079 --> 00:01:11.040
So Dennis is also the author of the Architect's Blueprint on Substack, which is honestly one of the few Substacks I read consistently.
00:01:11.359 --> 00:01:13.680
Dennis, welcome to Agentic Edge.
00:01:14.159 --> 00:01:14.879
Michael, thank you.
00:01:15.040 --> 00:01:15.920
Thank you for having me.
00:01:16.319 --> 00:01:26.319
Dennis, I want to start off with kind of a fundamental question here to ground the understanding for everyone in our audience, but how would you define AI transformation?
00:01:26.640 --> 00:01:28.079
Sure, that's a good question.
00:01:28.560 --> 00:01:38.640
For me, AI transformation begins where the presence of machine intelligence actually changes how the company operates.
00:01:38.799 --> 00:01:41.760
It's not just about making individual employees faster.
00:01:41.920 --> 00:01:43.280
We often talk about that a lot.
00:01:43.439 --> 00:01:45.120
There's tremendous amounts of traction there.
00:01:45.200 --> 00:01:47.599
You hear all the time in LinkedIn and other Substacks.
00:01:47.760 --> 00:02:08.800
It's really about how we change how we work, change our fundamental workflows, where the decisions are made, which activities are actually automated, which situations still require human in the loop or human judgment, potentially how many teams or how teams themselves are fundamentally structured or organized.
00:02:08.960 --> 00:02:12.400
That's, I think, is really the point of uh transformation.
00:02:12.560 --> 00:02:19.840
Now, giving employees co-pilot, GPT, Claude, any other AI assistance known to mankind is a very valuable asset for sure.
00:02:19.919 --> 00:02:24.639
I'm not saying that's not valuable, but to me, that's just uh primary adoption.
00:02:24.719 --> 00:02:26.240
That's just the the start.
00:02:26.560 --> 00:02:37.840
You could give 5,000 employees an AI tool, and you can still have the exact same process, the same handoffs, same approvals, same organizational structures.
00:02:38.000 --> 00:02:44.800
And so people become a little bit more productive for sure, but the company itself hasn't fundamentally transformed.
00:02:44.960 --> 00:02:47.360
There is no transformation that's really happening here.
00:02:47.599 --> 00:02:54.800
And so for me to answer your question, transformation really begins when you fundamentally ask it a different way.
00:02:54.960 --> 00:03:08.000
I would rather ask something like knowing what these AI capabilities and tool sets can do, knowing the capability and what you have at your fingertips today, would you design the existing workflow the same way?
00:03:08.240 --> 00:03:11.199
Take a financial process, for example.
00:03:11.439 --> 00:03:15.120
A lot of institutions that I've worked with, you know, there's multiple steps in there.
00:03:15.199 --> 00:03:22.639
There's human review, there's approvals, AI adoption might still be useful there and might expedite some of that.
00:03:22.719 --> 00:03:26.000
And it might help a financial analyst prepare something faster.
00:03:26.159 --> 00:03:28.159
But fundamentally, the workflow is still the same.
00:03:28.240 --> 00:03:47.919
You know, transformation might eliminate a step or two of that six-step process and automation may reduce some of that low-risk transactions, but continuously checking the data, sending the exceptions to the people only, fundamentally re-architecting, designing that workflow, I think that's the piece.
00:03:48.000 --> 00:03:58.960
And so if an organization is focused on uh AI transformation, they're fundamentally redesigning how the organization thinks, how the organization even uh looks.
00:03:59.039 --> 00:04:02.000
And that's a critical part of organizational intelligence.
00:04:02.319 --> 00:04:11.439
We're all on the same page here that pretty much every company out there right now is saying that they're on some sort of AI transformation journey.
00:04:11.680 --> 00:04:21.279
But in your experience, what is the difference between what they're saying and describing as their transformation and what's actually happening on the ground?
00:04:21.680 --> 00:04:22.879
No, that's a good follow-up.
00:04:23.120 --> 00:04:26.000
I think everyone says they're doing AI transformation.
00:04:26.079 --> 00:04:29.600
In fact, I think LinkedIn, I see nothing but AI activity there.
00:04:29.759 --> 00:04:30.879
I think it's a misnomer.
00:04:30.959 --> 00:04:38.240
When they talk about AI transformation, I think they're alluding to maybe some uh productivity enhancements they're seeing.
00:04:38.319 --> 00:04:41.759
And so, and I think they attribute that to word transformation.
00:04:41.920 --> 00:04:46.079
And that's not necessarily a criticism on that organization or their approach.
00:04:46.240 --> 00:04:47.439
I think that's the first stage.
00:04:47.600 --> 00:04:48.959
Productivity is critical.
00:04:49.040 --> 00:04:50.959
And so those enhancements are real.
00:04:51.120 --> 00:05:00.079
And so I don't want to say that they are not, but we should be really honest about the level of change that's actually occurring inside of an organization.
00:05:00.319 --> 00:05:07.759
Organizations measure frequently the number of AI licenses they issue out, what they have purchased, their overall stack.
00:05:07.839 --> 00:05:18.319
Uh, we also measure the number of pilots they've launched, the number of employees they have trained, uh, maybe the number of AI use cases that they even identified and that sits in their backlog.
00:05:18.720 --> 00:05:21.759
But those measurements tell you that AI is happening.
00:05:21.920 --> 00:05:31.839
There is an appetite for it, and maybe there's a um a mandate from leadership to push that along, but they don't tell you how you fundamentally changed.
00:05:31.920 --> 00:05:39.360
And so to measure that delta, I look at really four stages of maturity when I look at a particular organization.
00:05:39.600 --> 00:05:43.279
The first stage is really uh individual assistance.
00:05:43.519 --> 00:05:51.040
Does AI help a person summarize, write an email, allow us to research something, and maybe even write a piece of code?
00:05:51.120 --> 00:05:52.480
And I think that's the first stage.
00:05:52.639 --> 00:05:58.079
The second stage is really how are we reimagining and augmenting our existing workflows?
00:05:58.160 --> 00:06:05.120
And so AI becomes a part of that business process, but that basic process largely remains intact.
00:06:05.199 --> 00:06:10.639
And so we're peppering in AI there, but really not reimagining the activities itself.
00:06:10.879 --> 00:06:19.439
Third stage is when some organizations understand that they have squeezed as much as they can get out of the first few stages, they start looking at the fundamental process design.
00:06:19.519 --> 00:06:28.160
And so they think about the activities that they do and interactions they do with one another, um, the handoffs that they have today and the silos that they have built.
00:06:28.319 --> 00:06:35.439
When you see those disappearing because of AI and automation, you can then fundamentally you've changed the organization.
00:06:35.519 --> 00:06:36.800
You're behaving differently.
00:06:36.959 --> 00:06:42.639
And I think that will push you and lead you to the overall operating model redesign.
00:06:42.720 --> 00:06:49.360
And so I would measure, I would understand that there are stages to it, there are phases to it.
00:06:49.519 --> 00:06:51.519
Don't measure with these false metrics.
00:06:51.600 --> 00:06:56.879
Don't measure with the number of pilots you're running and how many folks are on your internal managed GPT.
00:06:57.120 --> 00:06:59.360
Instead, look at your business outcomes.
00:06:59.519 --> 00:07:02.079
Did your overall cycle time improve?
00:07:02.240 --> 00:07:03.680
Did your error rate decrease?
00:07:03.839 --> 00:07:08.079
Did your customers wait still that waiting time did that decrease?
00:07:08.319 --> 00:07:10.800
Those are critical things that I kind of look for.
00:07:10.959 --> 00:07:11.600
I like that.
00:07:11.759 --> 00:07:14.560
A lot of organizations get stuck measuring the wrong stuff.
00:07:14.800 --> 00:07:16.560
And you mentioned a lot of those cases.
00:07:16.639 --> 00:07:22.560
I see a lot of people who are talking about the number of tokens used per employee and a measurement of adoption.
00:07:22.720 --> 00:07:24.720
And I think that's missing it.
00:07:25.040 --> 00:07:36.639
If that basic measurement of adoption is kind of stage one, like you're talking about, the gold standard is probably something like ARR per employee, right?
00:07:36.800 --> 00:07:38.160
At stage four.
00:07:38.480 --> 00:07:42.000
How are you looking at measuring those individual stages?
00:07:42.160 --> 00:07:48.480
And how does the maturation mature as you go through, like, oh, we're moving from stage one to stage two?
00:07:48.959 --> 00:07:52.639
We know that everyone has the tools, we know that they're aware of what's available.
00:07:52.800 --> 00:07:53.120
Great.
00:07:53.279 --> 00:07:54.720
What are we looking for in stage two?
00:07:54.959 --> 00:08:01.360
What are we looking for in stage three that lead us up to ultimately something like ARR per employee?
00:08:02.319 --> 00:08:09.519
As you go through these different stages, I think your KPIs and your metrics should be adapted for those particular use cases for sure.
00:08:09.680 --> 00:08:12.879
And so there isn't a golden formula that you can apply there.
00:08:13.040 --> 00:08:15.759
But I would look at it from a use case-to-use case basis.
00:08:15.920 --> 00:08:27.920
And I think for employees that use and adopt these tool sets, I think one of the critical things organizations can do is provide rails, provide that common functionality that you can leverage to get on.
00:08:28.000 --> 00:08:34.639
And so what that does is will then empower individuals to move naturally matriculate from one stage to another.
00:08:34.879 --> 00:08:41.039
Without those rails, I think you're going to feel organizations are stuck and you'll see that they spin in governance.
00:08:41.120 --> 00:08:46.080
And so they'll make it to governance, they'll sit there and they'll review the same use cases over and over again.
00:08:46.320 --> 00:09:00.720
Maybe they'll sprinkle AI in there, some summarization, some basic summarization is in there, some basic transformation is in there, but you fundamentally haven't really disrupted that workflow and you haven't really asked yourself, I have four people managing that workflow today.
00:09:00.879 --> 00:09:01.840
Do I really need four?
00:09:02.000 --> 00:09:03.039
Can I have one?
00:09:03.279 --> 00:09:05.519
That's when you're actually making change.
00:09:05.600 --> 00:09:09.759
And you'll naturally you'll see that matriculate into your metrics and your AR.
00:09:10.480 --> 00:09:13.279
I just read recently about this company.
00:09:13.440 --> 00:09:43.440
I'm not gonna name names, but they saw some red flags in kind of their customer acquisition costs and decided kind of quickly that they were going to eliminate, I would assume was about 90 to 95% of their marketing organization to give themselves a chance to stop and rethink their processes and figure out how AI can assist or handle the vast majority of their processes.
00:09:43.919 --> 00:09:51.440
Do you think that is going to be a necessary step for a lot of organizations to really take it to one of the later stages that you're talking about?
00:09:51.600 --> 00:09:58.320
Or is there maybe a happy medium somewhere along that path where they don't have to take such drastic measures?
00:09:58.720 --> 00:10:02.000
I think those drastic measures are vastly performative.
00:10:02.080 --> 00:10:06.159
You know, I think that's there for the market to hear.
00:10:06.320 --> 00:10:14.559
What we have done is somehow mistaken the size of organizations for revenue that an organization can generate for the value they can produce.
00:10:14.639 --> 00:10:16.080
And you see that littered everywhere.
00:10:16.240 --> 00:10:20.159
For example, Snapchat still has thousand people plus for that app.
00:10:20.320 --> 00:10:21.279
I'm not really sure why.
00:10:21.440 --> 00:10:24.480
They've built the app, they've deployed it, they've built the models.
00:10:24.639 --> 00:10:26.960
I mean, you have the face filters, so on and so forth.
00:10:27.120 --> 00:10:28.320
I mean, I love Snapchat.
00:10:28.480 --> 00:10:31.200
That was actually the frontier of AI, your Snapchat filters.
00:10:31.279 --> 00:10:32.639
Let's not, let's not kid ourselves.
00:10:32.799 --> 00:10:35.759
But do you still need a thousand plus organization to manage that?
00:10:35.840 --> 00:10:36.639
I'm not really sure.
00:10:36.720 --> 00:10:41.840
And conversely, you have perplexity, you know, that came out in something like a 12-person organization.
00:10:42.000 --> 00:10:43.679
You know, why is that the case?
00:10:43.840 --> 00:10:46.559
And so there's a drastic gap there, you know.
00:10:46.639 --> 00:10:52.799
And if you look at that, I think what you're gonna see is the term AI native is often thrown around.
00:10:52.960 --> 00:10:56.960
AI native isn't someone that's using GPT and cloud right now.
00:10:57.200 --> 00:11:02.559
It's AI native is someone that's built their overall workflows around these tool sets.
00:11:02.720 --> 00:11:04.799
And that's how you can see those leverage.
00:11:04.960 --> 00:11:09.279
And so for someone to announce, hey, I'm gonna cut my marketing team by 99%, sure.
00:11:09.440 --> 00:11:11.120
I mean, that seems a little performative to me.
00:11:11.200 --> 00:11:15.039
I'd like to see the metrics and behind the scenes of what's actually going on.
00:11:15.200 --> 00:11:23.200
Um, but I think if you do follow that matriculation curve, snap the line and figure out where you are as an organization, first and foremost.
00:11:23.440 --> 00:11:27.279
The first order of magnitude is understand the complexity of your workflows.
00:11:27.360 --> 00:11:29.600
Do you even understand your workflows?
00:11:29.759 --> 00:11:31.039
Are they documented?
00:11:31.200 --> 00:11:32.960
Or are they institutional knowledge?
00:11:33.120 --> 00:11:34.480
Are they tribal knowledge?
00:11:34.639 --> 00:11:36.240
Do they sit in Excel spreadsheets?
00:11:36.480 --> 00:11:40.399
Do they sit in access databases and various legacy platforms?
00:11:40.480 --> 00:11:41.919
And I bet you it's the latter.
00:11:42.080 --> 00:11:44.879
It's scattered and fragmented all across the organization.
00:11:45.120 --> 00:11:49.600
It's never that, you know, 10 people in your team are actually executing.
00:11:49.759 --> 00:11:52.080
There's two of the 10 that are executing.
00:11:52.159 --> 00:11:53.679
There's eight sidewalk engineers.
00:11:53.759 --> 00:11:56.080
That's very common in larger organizations.
00:11:56.320 --> 00:11:56.799
You see that.
00:11:56.960 --> 00:12:00.000
And so in startups, you just don't have that luxury.
00:12:00.159 --> 00:12:03.919
You have to hire those two people that are keeping you, keeping you afloat, right?
00:12:04.000 --> 00:12:05.279
They're actually doing the work.
00:12:05.440 --> 00:12:18.799
But once you understand your workflows, document your workflows, then going back to the drawing board, reimagining them with these unique tool sets that's available, that's gonna write off the bat give you an understanding of what's residual.
00:12:18.960 --> 00:12:22.320
Then it's not a conversation of I'm gonna get rid of XYZ.
00:12:22.399 --> 00:12:24.639
It's well, how can I adapt these individuals?
00:12:24.879 --> 00:12:27.360
What other new roles do I have in this new world?
00:12:27.600 --> 00:12:30.879
Can I perhaps train them and move them over to that particular role?
00:12:31.120 --> 00:12:32.879
Are they fit for that role?
00:12:33.120 --> 00:12:34.960
Et cetera, having that conversation.
00:12:35.120 --> 00:12:47.919
And when you hear things of that sort, if you hear organizations, leaders talking in that language, I think more than less, you're gonna see a natural progression toward the native AI spectrum.
00:12:48.080 --> 00:12:53.759
But if you hear right off the bat, I'm gonna cut XYZ because we're dealing with agent development, I'd be cautious.
00:12:53.840 --> 00:12:55.200
I want to see the proof there.
00:12:55.440 --> 00:12:56.960
I like that perspective.
00:12:57.200 --> 00:13:06.960
I'm curious to hear your thoughts on the difference in the role that a CTO versus a CIO plays in AI transformation within a large organization.
00:13:07.200 --> 00:13:08.879
Are you guys thinking about the same things?
00:13:09.039 --> 00:13:10.240
Do you have the same goals?
00:13:10.480 --> 00:13:13.120
Is it a balance of one versus the other?
00:13:13.519 --> 00:13:18.960
I've worked in small, medium, large-size firms, startup, all of those in different organizations.
00:13:19.279 --> 00:13:20.879
That answer is going to change.
00:13:21.039 --> 00:13:21.360
Okay.
00:13:21.679 --> 00:13:28.320
In a larger organization, you're gonna have to have line of sight with respect to business master data.
00:13:28.480 --> 00:13:32.000
What are you doing with that master data for that particular line of business?
00:13:32.159 --> 00:13:33.679
How are you managing this?
00:13:33.919 --> 00:13:37.759
How are you solving customer problems for that business unit?
00:13:38.000 --> 00:13:43.279
So when you have to have that level of focus, I think a CIO role, that's their purview.
00:13:43.360 --> 00:13:44.399
And that's what they're looking at.
00:13:44.559 --> 00:13:46.639
The CTO role is an enabler role.
00:13:46.720 --> 00:13:56.080
You're making sure platforms are up and running, that irrespective of which CIO and which line of business, they can take advantage of those platform capabilities.
00:13:56.320 --> 00:13:58.480
What capability are you building today?
00:13:58.639 --> 00:14:05.360
How are you investing your firm's dollars in order to unblock fundamental capability down the road?
00:14:05.519 --> 00:14:09.840
Those are the things that you're managing and you're triaging on a day-to-day.
00:14:10.080 --> 00:14:20.879
And it's interesting, in my particular role, not only do I play a CTO role and manage that foundational capability, I also have the cyber role, the CISO role underneath me as well.
00:14:20.960 --> 00:14:29.039
And so I have to manage from a CISO perspective and a cyber perspective, not only do you push this capability out, how do I secure it?
00:14:29.200 --> 00:14:34.960
How do I keep up with the needs of my customers and the expectations that the market sets?
00:14:35.120 --> 00:14:39.279
And how do I make sure I put Rails in place for my organization?
00:14:39.519 --> 00:14:42.879
Micah, it reminds me of back in the day when we used to do mobile.
00:14:43.039 --> 00:14:49.759
We used to try to design these mobile sites and try to move from your web platform and your web interface to a mobile interface.
00:14:50.080 --> 00:14:59.840
iOS, Android, et cetera, Microsoft would try to be prescriptive with respect to how users would work in these platforms, how they would interact with these platforms.
00:15:00.000 --> 00:15:03.039
And we would try to prioritize features and functionality that way.
00:15:03.120 --> 00:15:12.159
But one of the things that we would always have to keep up with was the guy next door or a different app that already is pushing out this functionality.
00:15:12.320 --> 00:15:18.080
Why would I upload my driver's license when this guy just allows you to take a picture of it and digest it right off the bat?
00:15:18.159 --> 00:15:22.559
And so your customer expectation is set by other firms that are pushing the envelope.
00:15:22.720 --> 00:15:25.600
If you're not keeping up with them, they're going to find a way.
00:15:25.759 --> 00:15:41.679
And how I see that in the day-to-day is if I don't provide document intelligence, if I don't provide ways to automate, if I don't provide a community for my developers inside in order to perform this way, people will find a way.
00:15:41.840 --> 00:15:43.759
People will find other AI tools to use.
00:15:43.840 --> 00:15:49.519
They'll take pictures of your screen, send it to GPT, public GPT to do XYZ, come back.
00:15:49.600 --> 00:15:51.120
Your data has already left.
00:15:51.360 --> 00:15:52.720
Water will find a way, right?
00:15:52.879 --> 00:16:02.000
And so we have to put these rails and foundational capability in place in order to help and nudge people toward good behavior and incentivize that good behavior.
00:16:02.159 --> 00:16:08.000
But come back to your question, line of business, am I making sure that that data is clean?
00:16:08.159 --> 00:16:10.639
You know, can I get to my AI model better?
00:16:10.799 --> 00:16:14.320
Am I automating tasks and thinking about the workflows in my business line?
00:16:14.480 --> 00:16:17.440
That's a CIO's role in my eyes, the CTO role.
00:16:17.519 --> 00:16:23.200
It's an enabler broad, making sure all that capability is up and running and secure for adoption.
00:16:23.519 --> 00:16:29.440
I think it's incredibly interesting that you're owning both the technology and kind of the security side of things.
00:16:29.759 --> 00:16:33.679
How has that changed your perspectives on architecture?
00:16:33.840 --> 00:16:35.679
Because you've got a lot coming together here.
00:16:35.840 --> 00:16:42.320
There's the adoption of AI, there's the security aspect, but then there's the interoperability of the things that we're building.
00:16:42.559 --> 00:16:44.879
And so I've been building apps.
00:16:45.039 --> 00:16:49.840
I'm adding MCP capabilities for every single function that you can do in the UI.
00:16:50.080 --> 00:16:55.120
Because to me, that makes sense for an agent to be able to use the app the same way that a user can.
00:16:55.440 --> 00:17:01.360
How have you been challenged in the way you're thinking about architecture and working with your teams and the things that you're building?
00:17:01.519 --> 00:17:04.559
I know that's getting very specific all of a sudden, but No, no, no.
00:17:04.640 --> 00:17:05.680
I I love that question.
00:17:05.920 --> 00:17:07.599
So my background is really broad.
00:17:07.680 --> 00:17:11.920
And so I wrote front-end code for a long time, as soon as I got out of school.
00:17:12.160 --> 00:17:15.440
I left that and moved to a different team to write back end code.
00:17:15.519 --> 00:17:19.039
And I wrote a lot of crappy code for a long time before I wrote good code.
00:17:19.200 --> 00:17:24.960
And then I designed crappy systems for a while before I learned to do good systems and support these things long term.
00:17:25.039 --> 00:17:30.160
And every single turn, I learned a lesson and I put that lesson in the back and so on and so forth.
00:17:30.240 --> 00:17:36.960
And as I built this knowledge base, my internal knowledge base, my internal vector database that's inside me, right?
00:17:37.119 --> 00:17:43.039
As I went up, what I realized and it stood true was that architectural principle, keep it simple, stupid.
00:17:43.119 --> 00:17:44.960
You know, that is very, very true.
00:17:45.039 --> 00:17:53.920
And so if I can hold the architecture in my brain, if I can reduce it, if I can simplify it, chances are I can secure it properly.
00:17:54.079 --> 00:17:56.000
That's a core, core critical principle.
00:17:56.160 --> 00:18:06.240
And so every day my LinkedIn is flooded with vendors that want me to adopt AI products, security products, you name it, all over the place, free demos, this, that, and the other thing.
00:18:06.400 --> 00:18:08.000
Um, you have to resist.
00:18:08.079 --> 00:18:12.960
And I constantly see over-engineering of the overall stack.
00:18:13.119 --> 00:18:13.359
Okay.
00:18:13.759 --> 00:18:16.799
And I think where we're going is convergence.
00:18:16.880 --> 00:18:21.279
There's a great convergent event in the future over a horizon.
00:18:21.440 --> 00:18:26.319
And so we split up these roles, whether you're a CIO or a CTO or a CISO.
00:18:26.640 --> 00:18:35.680
Going in the future, especially with the gen TIC in mind, with delegation, with understanding context, one person can converge and manage all of this.
00:18:35.839 --> 00:18:40.559
The reason why we have that fragmentation inside of an organization is simply because information, right?
00:18:40.720 --> 00:18:42.880
One person doesn't have all that information.
00:18:43.119 --> 00:18:43.839
But think about it.
00:18:44.000 --> 00:18:45.200
If what if I did?
00:18:45.440 --> 00:18:48.640
What if I can scan every single log that's out there?
00:18:48.799 --> 00:18:59.839
If I understand what's happening in the edge of my network, if I understand all the holes and every single open source software that I have installed in my application, that's a tremendous amount of knowledge.
00:18:59.920 --> 00:19:05.839
I can discover vulnerabilities better, I can patch it better, I can do mitigation better, I can write software better.
00:19:06.000 --> 00:19:11.599
And then if I know your behavior itself, Micah, when you log in, what when do you drop your kids off and log in?
00:19:11.680 --> 00:19:13.200
And what's the first thing Micah does?
00:19:13.359 --> 00:19:13.759
Guess what?
00:19:13.839 --> 00:19:16.400
I can probably do a brief for you and just text it to you.
00:19:16.480 --> 00:19:17.440
Why even log in?
00:19:17.680 --> 00:19:19.440
How we think about work today.
00:19:19.599 --> 00:19:25.920
And so I think as we get more and more information, I think that these roles are very fluid and they change.
00:19:26.160 --> 00:19:36.480
I think keeping it simple, understanding and having a good sense of caution of here's a new model that's coming out, here's what it's promised to do, here's what it actually can do.
00:19:36.640 --> 00:19:38.640
What does a benchmark actually tell you?
00:19:38.799 --> 00:19:40.720
You know, understanding and asking that question.
00:19:40.799 --> 00:19:41.599
I think that's critical.
00:19:42.240 --> 00:19:52.720
But keeping your app stack very simple, keeping your cyber stack simple, making sure like workflows are simplified, and making sure you manage risk.
00:19:52.880 --> 00:19:54.240
And I think those things help.
00:19:54.400 --> 00:19:58.559
And that's how my CISO role influences my CTL role, vice versa.
00:19:58.640 --> 00:20:00.559
And my architecture brain comes in.
00:20:00.720 --> 00:20:05.759
And I really fight to make sure I don't over-engineer any problem.
00:20:06.079 --> 00:20:18.319
Even when I over-engineer, I put it down, go for a walk, go for a run, come back, and I try to attack it with this oversimplistic view and try to plead a devil's advocate for myself and say, you know, why can't this work?
00:20:18.480 --> 00:20:19.920
Why do I need A through Z?
00:20:20.000 --> 00:20:21.599
Why can't just ABC work?
00:20:21.759 --> 00:20:28.319
And so that's helped me manage my own context and make sure that I provide simpler solutions.
00:20:29.440 --> 00:20:36.160
Kind of on that note, you said a little bit ago that your inbox is flooded with vendors.
00:20:36.400 --> 00:20:36.640
Yes.
00:20:36.799 --> 00:20:40.079
And you have to resist and you are trying to keep things simple.
00:20:40.240 --> 00:20:45.680
So what is your perspective or your approach for build versus buy in this case?
00:20:46.000 --> 00:20:47.359
Oh, that's a great, great question.
00:20:47.759 --> 00:20:49.039
That's an architecture question.
00:20:49.119 --> 00:20:50.799
I feel like I'm being interviewed now.
00:20:50.960 --> 00:20:55.519
Um it always comes back to the business outcome.
00:20:55.680 --> 00:21:00.640
And so back in the day, we used to say, hey, we're not a tech company.
00:21:00.799 --> 00:21:04.160
I'm a financial services, I'm making it up, financial services company.
00:21:04.240 --> 00:21:06.640
I should work on this and leave the tech to the tech.
00:21:06.960 --> 00:21:08.720
Somehow something changed.
00:21:08.960 --> 00:21:13.519
Gardner probably put out talking points for CIOs, and everyone became tech companies.
00:21:13.680 --> 00:21:18.960
Didn't matter if you were a Dollar General or if you worked at Target, suddenly you're a tech firm.
00:21:19.119 --> 00:21:23.359
But how I look at it is I have a finite number of resources to my feel.
00:21:23.519 --> 00:21:30.319
I already have systems and capability that vendors build out for me or providers build out for me.
00:21:30.559 --> 00:21:33.279
And I use the architectural principle cut with the grain.
00:21:33.359 --> 00:21:35.839
You know, what can I compose with?
00:21:36.079 --> 00:21:41.440
And so I think the greatest thing that we do is I don't have to reinvent the wheel.
00:21:41.680 --> 00:21:45.200
I can go back and I can look at what capability do you have?
00:21:45.440 --> 00:21:47.680
Do you do very well today?
00:21:47.920 --> 00:21:50.960
And what can I get for the fraction of cost?
00:21:51.119 --> 00:21:52.960
I look for uni cost per that transaction.
00:21:53.039 --> 00:21:59.680
And so if I can slice and dice that and get that as low as possible for a good amount of value, sure, I'll I'll buy.
00:22:00.160 --> 00:22:01.599
I'll use that capability.
00:22:01.680 --> 00:22:03.680
I'll compose that into my workflow.
00:22:03.759 --> 00:22:10.480
I'll utilize that capability, move on until my workflow again transforms into something else.
00:22:10.559 --> 00:22:12.400
And then I'll look for a different solution.
00:22:12.720 --> 00:22:25.920
Only when there is a competitive advantage for my particular domain, for my company, where the vendor or the provider doesn't offer that, will I invest my own dollar in building that capability out?
00:22:26.160 --> 00:22:29.759
Otherwise, I'm pretty much into composing and orchestrating.
00:22:30.160 --> 00:22:30.480
Okay.
00:22:30.640 --> 00:22:34.720
So you're in the middle of AI transformation with your current organization.
00:22:34.880 --> 00:22:38.960
From what I understand, it's very safety focused, highly regulated.
00:22:39.759 --> 00:22:43.599
And that I assume creates a lot of complexity.
00:22:44.079 --> 00:22:49.599
So how did you decide or how have you made decisions on where to start and where to move next?
00:22:49.920 --> 00:22:51.039
Oh, that's a good question.
00:22:51.119 --> 00:22:54.400
I think it comes back to your operating foundation.
00:22:54.480 --> 00:22:58.640
And so even working at RPG, RPG is a heavily regulated industry.
00:22:58.880 --> 00:23:00.640
We're safety in critical industries.
00:23:00.720 --> 00:23:02.480
And so we're safety first, of course.
00:23:02.640 --> 00:23:09.440
And so one of my most important lessons here that I have learned is actually it didn't start with AI.
00:23:09.519 --> 00:23:10.960
I didn't start with the AI focus.
00:23:11.039 --> 00:23:13.759
I didn't start with implementing AI workflows.
00:23:13.920 --> 00:23:16.559
I actually started with the operating foundation.
00:23:16.640 --> 00:23:21.599
And so in a real enterprise, work rarely lives, like I said, inside of an application.
00:23:21.920 --> 00:23:30.079
What I found is that it lives with a lot of tribal knowledge, legacy platforms, sometimes hidden in systems that are old, that are fatigued.
00:23:30.319 --> 00:23:36.079
I found so many that were just sitting in people's workstations that were running and customer requirements were just fragmented.
00:23:36.160 --> 00:23:40.400
And so trying to go and collate and bring that all together, I think that was step one.
00:23:40.559 --> 00:23:50.880
And one of the other things that I pride myself doing is when I would meet with my stakeholders and do my initial round table, going around the organization trying to understand the problems that they have.
00:23:51.039 --> 00:23:52.480
Active listening actually helped.
00:23:52.640 --> 00:23:55.200
Just tell me what you're seeing day to day.
00:23:55.359 --> 00:23:57.279
What kind of experiences do you have, et cetera?
00:23:57.440 --> 00:24:01.599
And I would transcribe it, I would note it, I would go through it, refine it, so on and so forth.
00:24:01.680 --> 00:24:04.559
What is this individual actually trying to tell me?
00:24:04.720 --> 00:24:06.000
What are their pain points?
00:24:06.160 --> 00:24:07.039
I would take that.
00:24:07.200 --> 00:24:11.279
And that gave me a good understanding of how the organization was operating.
00:24:11.440 --> 00:24:11.759
Okay.
00:24:12.000 --> 00:24:22.880
And so whether it's a financial services company or a company such as RPG that works in a heavily regulated and a safety-focused industry, I think that's a common denominator.
00:24:23.039 --> 00:24:27.680
And so understanding how you operate, how you do your day-to-day is critical.
00:24:27.920 --> 00:24:37.920
And what often surprises me, and even the most advanced organizations, how many manual steps and human intervention is required for existing workflows.
00:24:38.079 --> 00:24:38.960
So that's really critical.
00:24:39.039 --> 00:24:40.160
And that opens my eyes.
00:24:40.240 --> 00:24:41.599
And then I go to simplify it.
00:24:41.680 --> 00:24:43.359
That's when I go and simplify it.
00:24:43.599 --> 00:24:49.920
I'm a very minimalist person, so I try to throw away all the extraneous stuff and reduce it to what it actually needs.
00:24:50.079 --> 00:24:53.920
Once I do that, I try to digitize the first aspect of it.
00:24:54.079 --> 00:24:54.400
Okay.
00:24:54.640 --> 00:25:01.039
Let me get and bring all that together and to provide a consolidated, structured data representation.
00:25:01.200 --> 00:25:02.400
Let me put that together.
00:25:02.640 --> 00:25:10.720
Once you have that, you have the basic foundations of doing any type of intelligent automation, data intelligence on top of that.
00:25:10.960 --> 00:25:14.720
Then of course, I look for deterministic processes.
00:25:14.880 --> 00:25:15.839
What's formulaic?
00:25:16.079 --> 00:25:17.839
What's A plus the B equals C?
00:25:18.000 --> 00:25:20.480
And I write that down and I take a jot of that.
00:25:20.720 --> 00:25:22.319
That's the overall wheel.
00:25:22.400 --> 00:25:23.920
You don't want to reinvent that wheel.
00:25:24.000 --> 00:25:24.640
That's working.
00:25:24.720 --> 00:25:26.160
I get that for a good cost.
00:25:26.319 --> 00:25:28.319
Again, it goes back to economics.
00:25:28.720 --> 00:25:31.440
That's a very good unit per transaction capability.
00:25:31.599 --> 00:25:32.960
And so I note that down.
00:25:33.279 --> 00:25:39.440
And then I start redesigning my process from the ground up with AI and any of the latest tool sets that I have.
00:25:39.680 --> 00:25:45.519
I use deterministic processes first, and then I sprinkle probabilistic toward the end.
00:25:45.599 --> 00:25:49.599
And so that pattern seemed to work out a lot.
00:25:49.759 --> 00:26:00.559
And so AI is really powerful when ambiguity exists, but traditional software and deterministic software is still superior in in most environments.
00:26:00.640 --> 00:26:05.920
And so I try to use that good balance and try to manage out when I'm trying to automate internally.
00:26:06.240 --> 00:26:17.759
I like that perspective and I like the sequencing there because you talked about I'm going to stick to deterministic where I can and only use AI where I need to for the probabilistic intelligence, right?
00:26:17.920 --> 00:26:32.160
I think a lot of organizations will sometimes just sprinkle AI everywhere and then all of a sudden costs go up because they're applying probabilistic logic where deterministic logic would have served perfectly fine.
00:26:32.400 --> 00:26:38.000
And I think that's a good segue into a quote that I'm going to pull from your Substack.
00:26:38.160 --> 00:26:46.880
You said that technology is essentially organizational design rendered in software, and that software doesn't solve complexity, it preserves it.
00:26:47.039 --> 00:26:56.880
So for the organizations that are sprinkling AI everywhere on top of broken processes, what ultimately happens and how would you guide or coach them?
00:26:57.279 --> 00:26:58.000
No, that's good.
00:26:58.079 --> 00:26:59.359
I forgot I wrote that actually.
00:26:59.519 --> 00:27:02.960
I had to go back to my Substack and read, and I'm like, oh man, that was profound.
00:27:03.119 --> 00:27:09.200
Um but when I said that back, I was seeing Conway's Law in effect all the time.
00:27:09.440 --> 00:27:15.039
I had one of these moments when I'm sitting in a meeting and I rose above myself and I saw the silos.
00:27:15.440 --> 00:27:18.559
And when I looked at the code like Neo, I saw the silos again.
00:27:18.640 --> 00:27:21.519
And so that's kind of how I came up with that whole thing.
00:27:21.680 --> 00:27:25.359
But when I said technology is organizational design rendered in software, I meant it.
00:27:25.519 --> 00:27:31.279
What I mean by that is really that software usually reflects how an organization is functioning underneath.
00:27:31.359 --> 00:27:47.119
The approvals, the silos, the handoffs, any controls that you have, even the politics that remains, the reporting relationships, the organization, and even historical baggage, they all eventually become encoded in the actual technology.
00:27:47.200 --> 00:27:49.039
And you can see it plain as day.
00:27:49.599 --> 00:27:56.559
And people will look at the tech back and then they look at those complicated systems and then they say this the software is too complex.
00:27:56.720 --> 00:27:59.680
But often the software did not create that complexity.
00:27:59.759 --> 00:28:00.480
It's preserved.
00:28:00.640 --> 00:28:03.839
That complexity already existed in that organization.
00:28:04.000 --> 00:28:05.759
People created that complexity.
00:28:06.000 --> 00:28:13.920
And so this is why simply taking like an AI tool and layering that on top of your existing systems is not very fruitful.
00:28:14.079 --> 00:28:16.319
And in fact, it can often be dangerous.
00:28:16.480 --> 00:28:24.720
And so if you don't challenge the underlying process, you can end up like automating corporate dysfunctions rather than actually transforming the company.
00:28:24.960 --> 00:28:35.119
Imagine like a factory where you have a physical layout that's poorly designed and materials cross the factory floor 10 times, 100 times unnecessarily.
00:28:35.359 --> 00:28:40.640
Using robots afterwards to actually move that freight back and forth, does that do anything?
00:28:40.799 --> 00:28:43.519
Yes, it does something, but are you actually fixing anything?
00:28:43.759 --> 00:28:44.240
Not really.
00:28:44.400 --> 00:28:48.079
You've just created robots that can go back and forth maturely faster.
00:28:48.400 --> 00:28:49.599
The mess happens faster now.
00:28:49.680 --> 00:28:49.839
Yeah.
00:28:50.079 --> 00:28:50.720
That's exactly right.
00:28:50.799 --> 00:28:54.720
It's like my my room boat when I go away and I come back to the mess that is created.
00:28:54.799 --> 00:28:56.079
It's very similar.
00:28:56.480 --> 00:29:00.640
And so AI doesn't absolve the leaders for sure from process design.
00:29:00.720 --> 00:29:02.000
You absolutely have to do that.
00:29:02.160 --> 00:29:07.599
But in some ways, what it does is it incentivizes good process design even more.
00:29:07.680 --> 00:29:13.839
Because what happens is AI allows both good and bad practices to scale more quickly.
00:29:14.000 --> 00:29:20.799
And so, in essence, AI basically ends up amplifying the quality of the systems underneath it.
00:29:20.880 --> 00:29:27.680
And so what that means is software, again, it preserves your organizational complexity and whatever existed there.
00:29:27.839 --> 00:29:31.039
Uh, when you put AI on top of it, it just scales it.
00:29:31.200 --> 00:29:42.240
And so before you can actually um accelerate and introduce AI into your processes, you should really ask yourself which processes deserve to live on their own?
00:29:42.400 --> 00:29:47.200
And are they at the most simplified form first before you uh introduce any AI?
00:29:47.279 --> 00:29:52.000
But that's what I really meant when I wrote that particular piece of that article.
00:29:52.400 --> 00:29:53.759
I like what you said there.
00:29:53.920 --> 00:29:59.920
For a lot of organizations, they've been on this digital transformation journey for 25 years now, right?
00:30:00.000 --> 00:30:01.920
And I don't think that journey ever really ended.
00:30:02.000 --> 00:30:04.640
It was kind of like, well, we were doing digital transformation.
00:30:04.799 --> 00:30:09.680
This leader came in, brought this workflow app, this leader came in, brought this case management app.
00:30:09.839 --> 00:30:11.839
Those people got a bonus and took off.
00:30:12.000 --> 00:30:20.319
And so they're left with this really fragmented architecture for the way that workflows through their environments.
00:30:20.799 --> 00:30:23.599
Are they able to use AI and where?
00:30:23.759 --> 00:30:37.920
Or do they need to step back and say, we need to redesign our entire workflows first and orient on one workflow application or one case management application, like we said we were going to do 15 years ago before we can start to adopt AI?
00:30:38.079 --> 00:30:39.359
How do you think of that?
00:30:39.759 --> 00:30:41.039
No, that's a good question.
00:30:41.200 --> 00:30:47.680
The question that you're asking is do you have to do an extensive six-month process engineering exercise?
00:30:48.000 --> 00:30:50.240
Can I um adopt something now?
00:30:50.319 --> 00:30:54.799
And so if we make that a prerequisite for every use case, I think you'll be blown out of the water.
00:30:54.880 --> 00:30:56.720
Everyone will innovate a lot faster than you.
00:30:56.799 --> 00:31:00.720
And most organizations will move too slowly and people will start working around them.
00:31:00.799 --> 00:31:02.799
You know, like I stated, water will find a way.
00:31:03.039 --> 00:31:06.799
But in some levels, you do have to have some level of discipline.
00:31:06.960 --> 00:31:15.359
And so I keep going back to before you do apply AI to some of this capability, I think you should fundamentally ask yourself a couple of questions.
00:31:15.519 --> 00:31:17.759
What outcomes are you actually trying to produce?
00:31:17.920 --> 00:31:22.400
Not what processes do you actually have, what results actually matter?
00:31:22.559 --> 00:31:23.839
I think that's really critical.
00:31:24.079 --> 00:31:26.480
What decisions happen along the way?
00:31:26.640 --> 00:31:30.160
And so AI frequently becomes useful at decision boundaries.
00:31:30.319 --> 00:31:36.319
Put somebody in the middle of an interpretation, in the middle of a classification, a prioritization.
00:31:36.559 --> 00:31:38.799
What actions do you recommend after the fact?
00:31:38.880 --> 00:31:39.599
That's really critical.
00:31:39.759 --> 00:31:42.079
And so again, what decisions happen along the way?
00:31:42.240 --> 00:31:43.359
Ask yourself that.
00:31:43.599 --> 00:31:47.359
What information or data is required to make those decisions?
00:31:47.519 --> 00:31:54.319
And so information is scattered and fragmented and often and resides in employees' brain.
00:31:54.480 --> 00:31:55.599
It's tribal knowledge.
00:31:55.759 --> 00:31:58.079
And so it's contradictory.
00:31:58.319 --> 00:32:00.240
It doesn't have any authoritative source.
00:32:00.400 --> 00:32:01.920
There is no master data sometimes.
00:32:02.000 --> 00:32:06.960
And so that is usually a data and architectural problem before it actually becomes an AI problem.
00:32:07.119 --> 00:32:09.599
And so that's a critical question you have to ask yourself.
00:32:09.839 --> 00:32:12.240
And what controls do we need?
00:32:12.319 --> 00:32:16.480
You know, what controls are absolutely mandatory and what is non-negotiable?
00:32:16.640 --> 00:32:27.839
And so that includes when you start touching domains like authorization, security, compliance, segregation of duties, privacy, auditability, and even accountability.
00:32:28.000 --> 00:32:32.960
And then I would look at what old constraints still have to be applied.
00:32:33.119 --> 00:32:34.559
Do they have to be applied?
00:32:34.799 --> 00:32:39.200
Maybe a step exists because two systems could not communicate before.
00:32:39.359 --> 00:32:42.319
Maybe now you have a better API or a better protocol.
00:32:42.559 --> 00:32:43.839
Do I still need that step?
00:32:44.000 --> 00:32:49.119
And maybe an approval exists because we didn't have that visibility before.
00:32:49.279 --> 00:32:49.839
Now we do.
00:32:50.000 --> 00:32:52.160
And so do I absolutely need that step?
00:32:52.319 --> 00:32:53.680
And so that's really critical.
00:32:53.839 --> 00:33:02.160
And so some manual reeking the data for integration sometimes will be built just because they didn't have that integration piece.
00:33:02.319 --> 00:33:03.759
Now you have that integration piece.
00:33:03.920 --> 00:33:05.039
Do you still need to do that?
00:33:05.119 --> 00:33:07.359
And so that's a critical question you have to ask yourself.
00:33:07.519 --> 00:33:12.960
The objective is really not to document existing processes with preciseness, really.
00:33:13.119 --> 00:33:16.319
It's really to understand it well enough to challenge it.
00:33:16.400 --> 00:33:17.680
And that's a critical component.
00:33:17.759 --> 00:33:22.319
And so don't have to automate the full-blown process map.
00:33:22.559 --> 00:33:28.079
The opportunity with AI tools today is really, well, how do I challenge that process map?
00:33:28.160 --> 00:33:29.119
So that's really critical.
00:33:29.200 --> 00:33:36.960
And so I would say don't spend six months documenting your processes that you're you're gonna spend the next two weeks eliminating.
00:33:37.440 --> 00:33:44.880
Dennis, what's something about AI transformation that you think most people, most orgs have not figured out yet?
00:33:45.039 --> 00:33:50.160
Something that you're watching or thinking about, something that's sitting in the back of your mind, bugging you.
00:33:50.559 --> 00:33:51.839
No, that's a good question.
00:33:52.160 --> 00:33:56.640
I think most people have applied it to simplistic workflows today.
00:33:56.880 --> 00:34:00.319
And I think there are two things that I don't think they fully appreciate.
00:34:00.480 --> 00:34:05.680
One, I think AI is likely to change your overall organizational structure, not just personal productivity.
00:34:05.839 --> 00:34:14.079
Like I said, you know, historically we have created these vertical hierarchies because one person does not have enough context, they can't digest all that information.
00:34:14.239 --> 00:34:15.760
Maybe they can, maybe they can't.
00:34:15.920 --> 00:34:17.920
And so we've created these verticalized layers.
00:34:18.079 --> 00:34:19.199
AI is gonna squish that.
00:34:19.280 --> 00:34:30.239
So many AI discussions assume that company remains essentially the same and everyone simply does become that 20% or 30% productive, again, against with the productivity suite.
00:34:30.480 --> 00:34:32.159
I'm not really convinced that's where it ends.
00:34:32.320 --> 00:34:35.360
I think it has to really attack your organizational structure.
00:34:35.599 --> 00:34:41.199
Organizations today and historically they exist to coordinate information and decisions.
00:34:41.360 --> 00:34:49.360
In fact, management layers exist specifically for that, just to make sure you address human limitations with respect to capacity and bandwidth.
00:34:49.519 --> 00:34:53.920
Now, as information moves up the organization, decision moves down.
00:34:54.079 --> 00:35:01.199
Managers monitor activity, they summarize what's happening, they resolve any exceptions, and they coordinate work across teams.
00:35:01.360 --> 00:35:03.119
Now, AI changes all that.
00:35:03.280 --> 00:35:09.760
Now I can look at that data, I can centralize that data, I can democratize that specialized knowledge.
00:35:09.920 --> 00:35:21.760
And so when systems can continue, summarize what's happening, they can monitor for exceptions, they can recommend next best actions, and they can even coordinate some of these activities and execute a routine playbook.
00:35:21.920 --> 00:35:26.400
The amount of information a person or team needs, they can absolutely manage that, right?
00:35:26.559 --> 00:35:30.079
And so what that will do is essentially lead to a flatter organization.
00:35:30.239 --> 00:35:37.760
And so you have new roles with wider spans of control, you have smaller, multidisciplinary teams, you have fewer handoffs.
00:35:37.840 --> 00:35:41.519
And so managers may eventually supervise systems of people.
00:35:41.760 --> 00:35:50.639
Today they manage people, but I think in the future, or even now, it's gonna start changing to managing agents that are supervising tasks.
00:35:50.800 --> 00:35:51.760
I think that's really critical.
00:35:51.840 --> 00:35:54.079
I don't think that people are paying enough attention to them.
00:35:54.320 --> 00:35:57.360
Now, that doesn't mean like leadership absolutely disappears.
00:35:57.599 --> 00:36:02.320
I think it more alludes to the fact that the nature of that relationship changes.
00:36:02.559 --> 00:36:07.199
Secondarily, I think execution becomes less expensive.
00:36:07.360 --> 00:36:10.639
However, I think judgment becomes a lot more valuable.
00:36:10.800 --> 00:36:25.440
And so AI can generate like the first draft for you, it can create optionality for you, it can even perform some analysis for you, it can even assemble and aggregate information for you, and it can even execute and delegate parts of your workflow and execute those.
00:36:25.599 --> 00:36:31.840
The differentiating human capability starts becoming well, am I solving the right problem?
00:36:32.079 --> 00:36:34.400
Am I considering the right context?
00:36:34.559 --> 00:36:36.559
Am I setting the right boundaries and guardrails?
00:36:36.880 --> 00:36:39.679
And am I evaluating the trade-offs properly?
00:36:39.840 --> 00:36:42.239
And so judgment is really critical.
00:36:42.400 --> 00:36:46.159
And fundamentally, of course, accepting accountability because AI can't do that.
00:36:46.320 --> 00:36:58.880
So talking about judgment and the guardrails that you talked about earlier, Brenda from Finance just got Claude Pro and she's vibe coding all kinds of apps and deploying them everywhere.
00:36:59.119 --> 00:37:01.679
What is her role in AI transformation?
00:37:01.840 --> 00:37:04.000
Is what she's doing destructive?
00:37:04.159 --> 00:37:06.480
Is it introducing more risk to the organization?
00:37:06.639 --> 00:37:08.320
Is it accelerating outcomes?
00:37:08.480 --> 00:37:09.760
How do you think about that?
00:37:10.000 --> 00:37:17.599
And how would you guide someone like Brenda who's excited about AI, but maybe building things that you don't want built?
00:37:17.920 --> 00:37:19.840
Shadow IT on steroids now.
00:37:20.079 --> 00:37:21.199
Yeah, no, absolutely.
00:37:21.360 --> 00:37:27.360
I think Brenda's probably the most important person in our AI transformation story.
00:37:27.519 --> 00:37:29.840
So for decades, it's really common, right?
00:37:29.920 --> 00:37:36.480
And so there's been a tremendous amount of gap between the person who understands the business problem and the person that builds software.
00:37:36.559 --> 00:37:39.920
So analysts have sat in the middle trying to merge both environments.
00:37:40.079 --> 00:37:43.039
Architects have sat and trying to merge both environments.
00:37:43.199 --> 00:37:46.400
But I think the power of someone like Brenda is they understand the work.
00:37:46.480 --> 00:37:47.280
She understands the work.
00:37:47.360 --> 00:37:54.320
She knows what frustrates her, where the process breaks, what people repeatedly enter, which reports take four hours to assemble.
00:37:54.480 --> 00:37:56.800
That is institutional knowledge that's sitting with her.
00:37:56.960 --> 00:38:00.239
Now, historically, she would have documented that problem.
00:38:00.400 --> 00:38:11.360
She send it over to technology, technology would pick it up, you would do a scrum planning session for hours and score it and all these things, and eventually, six months later, build something that doesn't solve the problem.
00:38:11.519 --> 00:38:13.039
You know, that's really common, right?
00:38:13.199 --> 00:38:18.480
And so uh AI, I think, is beginning to collapse and converge that chasm.
00:38:18.639 --> 00:38:26.159
And so Brenda can now turn around and take her understanding of the actual business problem that she's solving and create something that's functional.
00:38:26.239 --> 00:38:27.760
And so that's really incredibly powerful.
00:38:27.840 --> 00:38:29.280
I think that should be celebrated.
00:38:29.440 --> 00:38:33.199
And she's a critical piece of this transformation for any organization.
00:38:33.440 --> 00:38:55.360
However, the other side of that, the risk that she's entering into is that I don't think Brenda's probably cognizant of the fact that now she's not only a business um guru, uh, now she's a developer, she's a product owner, she's an architect, a security engineer, a data engineer, a QA person, a whole plethora of things that she was not assuming that she would be.
00:38:55.519 --> 00:38:59.760
And so assuming all those responsibilities gets very heavy very quick.
00:39:00.000 --> 00:39:01.519
But I don't think the problem is Brenda.
00:39:02.079 --> 00:39:10.639
I think the problem is whether the organization has created a governed path for what Brenda can build to become enterprise software.
00:39:10.800 --> 00:39:15.519
And so I tend to think that governance really here should scale with risk.
00:39:15.599 --> 00:39:24.719
So Brenda's Brenda's experimenting with something that's non-sensitive in a sandbox environment, no one should depend on those results, then you know, let it let it go.
00:39:24.880 --> 00:39:26.000
Let her experiment.
00:39:26.239 --> 00:39:27.599
Governance should be very light.
00:39:27.679 --> 00:39:31.679
And so she should be able to do many integrations at light speed.
00:39:31.920 --> 00:39:43.679
Now, when her team starts beginning to rely on critical data, the organization should allow Rails, should introduce testing, documentation, version control, support, those critical capabilities.
00:39:43.840 --> 00:39:55.920
If it touches production systems, PII data, customer information, so on and so forth, then it should have architecture, security controls, monitoring, all that capability that should be integrated for her.
00:39:56.239 --> 00:40:06.639
I think this is one of the areas where you look for something like a platform engineering team where they stand up that critical capability that the organization can rely on.
00:40:06.800 --> 00:40:11.280
They enable things like an MCP server where you can discover domain capability.
00:40:11.519 --> 00:40:19.199
They enable things like an API gateway or excuse me, an AI gateway where you can provision rate limiting, so on and so forth, things of that sort.
00:40:19.679 --> 00:40:26.639
Very analogous to the journey that we had back in the day when we moved from monolithic to distributed applications.
00:40:26.800 --> 00:40:29.360
It's essentially that same journey that's happening now.
00:40:29.679 --> 00:40:38.079
Put in those rails for Brenda, let Brenda leverage those existing patterns, enable her to get to the end zone a lot faster.
00:40:38.239 --> 00:40:46.239
And so my punchline there really is that I wouldn't put do anything to stop Brenda, but I would want 10 Brendas, 100 Brendas.
00:40:46.400 --> 00:40:53.599
I just need an architecture and platform that allows Brenda, people like Brenda to innovate safely.
00:40:54.000 --> 00:40:58.400
Well, we always like to end episodes with something actionable.
00:40:58.800 --> 00:41:05.920
So for someone who's listening to this right now and they want to do this right, where do they actually start?
00:41:06.079 --> 00:41:08.239
What resources, departments, concepts?
00:41:08.400 --> 00:41:09.519
What's your recommendation?
00:41:09.840 --> 00:41:13.760
I would start by creating a hundred-page enterprise AI strategy.
00:41:15.760 --> 00:41:19.119
No, and I'd be and hiring both of you to do the work, right?
00:41:19.199 --> 00:41:19.760
Absolutely, absolutely.
00:41:19.920 --> 00:41:20.880
You know, put that in there.
00:41:20.960 --> 00:41:35.440
Um, no, what I would actually do is I would tell them to pause and I would tell them to look actively listen inside their organization, look for things that their users say, things such as, you know, we do this manually every day.
00:41:35.599 --> 00:41:37.440
When only one person knows how this works.
00:41:37.599 --> 00:41:40.880
We copy this information from one system to another.
00:41:41.119 --> 00:41:44.320
Uh, we have to read every one of these documents.
00:41:44.480 --> 00:41:47.599
You know, there is all these triggers that you could look for.
00:41:47.679 --> 00:41:48.880
I would look for those.
00:41:49.039 --> 00:41:51.599
Those are signals, those are signals for opportunity.
00:41:51.760 --> 00:42:02.000
Then I would choose one particular workflow where the pain is significant, where you're impacting a lot of people, and where you can actually measure that pain.
00:42:02.159 --> 00:42:05.119
And the initial risk is also manageable.
00:42:05.280 --> 00:42:06.639
I think those are really critical.
00:42:06.800 --> 00:42:10.000
Pain high, value high, risk low.
00:42:10.159 --> 00:42:11.760
You know, I would look for that.
00:42:11.920 --> 00:42:19.440
I would dig in at that point, I understand that workflow from beginning to end and make sure you separate the work in three categories.
00:42:19.599 --> 00:42:23.760
I would look for deterministic capability and calculations there.
00:42:24.079 --> 00:42:27.280
What rules do you have in flight, validation rules?
00:42:27.519 --> 00:42:31.440
Do you have structured systems and transactions that can support that deterministic work?
00:42:31.599 --> 00:42:33.440
I would look to tease those out.
00:42:33.679 --> 00:42:35.760
I would look for cognitive work.
00:42:35.840 --> 00:42:42.000
And so things that are interpretive, things that you do with respect to summarization, classification, pattern recognition.
00:42:42.159 --> 00:42:43.280
That's also opportunity.
00:42:43.360 --> 00:42:45.440
I would look for that category of work as well.
00:42:45.599 --> 00:42:53.519
And then I would look for consequential work, things that are regulatory, contractual, safety, security, those are the third bucket.
00:42:53.760 --> 00:43:10.159
And then I would identify authoritative data, determine which system is the source for that information, map those out, and build one end-to-end simplified flow, not 20 different use cases, not 20 different disconnected paths.
00:43:10.320 --> 00:43:12.079
You know, take a simple workflow.
00:43:12.159 --> 00:43:21.840
Again, look for that signal, break it out into three different pieces, look to automate the deterministic aspects of it, look for deterministic capability to do that.
00:43:22.079 --> 00:43:30.000
Sprinkle AI at the tour, the last mile, just to make sure that you're interpretive and adding the charm that's necessary to communicate back to the user.
00:43:30.159 --> 00:43:38.880
And then make sure you add analytics to it so you can measure the business outcome, not in terms of its volume of use, in terms of am I reducing errors?
00:43:39.039 --> 00:43:41.039
Am I reducing those manual handoffs?
00:43:41.280 --> 00:43:47.840
Did that person that once had that institutional knowledge only can I survive when he goes on vacation, when he she goes on vacation?
00:43:47.920 --> 00:43:48.639
That's really critical.
00:43:48.800 --> 00:43:50.159
And so I would start there.
00:43:50.320 --> 00:43:55.119
I would leave the team by saying start with the actual work, not the model.
00:43:55.280 --> 00:43:57.519
That's kind of how I would lead into it.
00:43:57.840 --> 00:43:58.400
Very cool.
00:43:58.639 --> 00:44:01.519
Dennis, thank you so much for joining us on Agentic Edge.
00:44:01.599 --> 00:44:09.440
For anyone who wants to read more of Dennis's work, the Architect's Blueprint at Substack, you can check out more of his content there.
00:44:09.519 --> 00:44:10.719
We'll have that in the show notes.
00:44:10.880 --> 00:44:15.360
Dennis, thank you so much for joining us, and we'll catch you next time on Agenc Edge.
