Agentic AI: Moving Beyond Pilots to Enterprise Impact

Heat. Heat.

>> [music] [music] [music] [music] >> Heat. Heat.

>> Heat. Heat.

>> [music] [music] >> Heat. Heat.

>> Heat. Heat.

Heat. Heat.

[music] [music] Hallelujah.

[music] >> [music] [music] >> Heat. Heat.

>> Heat. Heat.

Heat.

Heat.

Heat. Heat.

[music] Heat.

Heat.

[music] >> [music] >> Heat. Heat.

>> Heat. Heat.

Yeah.

[music] Heat. [music]

Heat. [music] Heat. Heat.

Heat. Heat.

Heat. Heat.

Heat. Heat.

[music] >> [music] [music] >> Heat. Heat.

>> Heat. Heat.

Hallelujah.

[music] [music] Heat. Heat.

Heat. Heat.

Heat.

Heat.

>> [music] [music] >> In an unpredictable world, you can't simply wait for change to happen.

We created [music] McKenzie Live for leaders like you. Our global experts break [music] down the need to know on the issues that matter most to decision makers and to society.

We'll [music] introduce you to leading research, new and transformational ideas, and practical knowledge and [music] strategies.

All to empower the change makers of today, enabling you to step forward [music] boldly, tackle real challenges, and drive change

with clarity and purpose.

We're helping you [music] take control so you can drive the conversation. So

you can lead fearlessly while leaving no one behind. [music] So you can take the

one behind. [music] So you can take the reigns to create and accelerate growth.

So you [music] can build resilience for the future.

Together we can push towards [music] the opportunities of our era.

Let's get started [music] now.

Welcome to McKenzie Live. Today's

moderator is Lucia Rahilly.

Hello and welcome to McKenzie Live. I'm

Lucia Rahilly, Mackenzie's editorial director and your host for today's event on Agentic AI and how to move beyond discrete pilots to impact across the

enterprise enterprisewide. Before we

enterprise enterprisewide. Before we start, quit quick heads up that our goal with this series of live events is to give you the chance to engage with our

experts on our latest research.

Obviously, asking questions is a big part of that. So, if you are registered for this event, please do not hesitate to use the Q&A functionality within your

live stream. Given the topic, I will

live stream. Given the topic, I will caveat that we have a very large group in the audience today, but we will take your questions throughout the session

and answer as many as time allows.

Joining us today are two of our experts on aentic AI and technology more generally, Clemens Hiata and Dave Kerr.

Clemens is a senior partner and he is the global leader of McKenzie Technology where he supports prominent companies embarking on digital transformations across a range of industries, tech,

media, financial services, advanced industries, private equity and so forth.

Clemens is based in Copenhagen and Dave is a partner of Mackenzie's London office. He leads the firm's

London office. He leads the firm's software cyber cloud andure work in the Asia-Pacific region as well as our largest technology deliveries from build

by McKenzie. Dave also works with a

by McKenzie. Dave also works with a range of clients across industries to develop new products and services using the best of modern engineering practices and cloudnative technologies. He is also

a passionate open-source developer, tech blogger and writer and you can find his book effective shell on Amazon. David

Clemens, welcome to McKenzie Live.

>> Thanks for having us. Thanks, Lucia.

>> We're thrilled to have you both together with us today. Let's get rolling. So,

Genai has obviously had a seismic effect on the ways that organizations engage with technology on the day-to-day and also an incredibly rapid one. Our most

recent global survey on AI showed that more than three4ers of companies now use AI in at least one business function.

That's up from about half the year previous. Clemens, let's start with some

previous. Clemens, let's start with some context. Give us the high level, your

context. Give us the high level, your best highle quick and dirty on the state of AI today.

>> Thanks, Lucia. So, let's uh let's get [clears throat] into it. So um this is a really uh interesting time in uh in technology. Uh the the emergence of

technology. Uh the the emergence of genai and genic technologies has been a real inflection point in the technology

history. If you think about it, we have

history. If you think about it, we have a software and cloud uh industry that is probably around 900 billion in 2025.

It is growing from 200 billion 10 years ago and it's expected to land somewhere close to 4,000 billions in 10 years from

now. So it's an immense growth threat

now. So it's an immense growth threat and um and this technology shift uh is monumental. Everybody in the industry is

monumental. Everybody in the industry is is jocking for for position. So, so kind of the the environment right now feels

like a battlefield.

So, um every bit everything is up for grabs. Um the the business model of of

grabs. Um the the business model of of SAS is something kind of of that is being discussed. The productivity

being discussed. The productivity promise in the companies is is kind of significant.

uh the uh whole uh drivers of uh traffic to e-commerce is bound to change. The

internet action model and and the consumer's percept or kind of expectation in in how they want to be

served is changing. So um this this year 2025 uh is a really really uh special time

and we can also see that uh in the behavior in the industry. the uh the we we are really kind of of in this place

where we are kind of of of jocking for for for positions and the enterprises are uh are gearing up uh

kind of of their their projects in 2025 going into 2026.

It's very very interesting time.

>> Great. So, Clemens, the discourse on AI often toggles between peril and promise

or between challenges and enthusiasm.

And on the challenge side of the le ledger, we're increasingly seeing headlines about AI not yet translating into tangible economic

results. You're out there talking to

results. You're out there talking to leaders every day. Why aren't

organizations seeing AI translate into more value in your view? Why this

paradox?

>> So, so we are seeing a a sequenced uh kind of of um action here. So, we we are seeing like a massive growth in some

of the most easy use cases. If you think about the consumer chat GPT that we we all use every day now that is the

largest part of this market uh by far.

The the second big use case is the whole software creation uh process uh which is kind of of is quite kind of deterministic and and and kind of of

well documented and we are seeing significant kind of of traction there but the story is changing during 2025.

Uh 2025 is um is a year where we begin to see a a real shift kind of the the top 20% of companies are beginning to

see real traction and they are setting very ambitious uh budget already for 2026.

So we are seeing it in in technology budgets. We seeing it in in selected

budgets. We seeing it in in selected uh areas of um of of pilots and we are

seeing the companies that have had a significant number of of pilots sometime and I'll talk about companies that have more pilots than Lufansa.

they are consolidating these efforts and saying we have some really big bets that we are going to consolidate about. So,

so we are beginning to see the leading companies move and and we think it's going to be very important to be in on that race. It is like being in the tour

that race. It is like being in the tour to France and missing out of the pelaton.

uh if if you're not kind of in the race, it's going to be difficult to catch up.

And in the last technology revolution, we can statistically prove that being a late mover was a a a bad strategy. So uh

so this is beginning to uh to to happen now.

Um and um but why don't we kind of or talk a little bit about kind of of kind of what are the use cases and and how we

are are uh seeing them shift.

>> Yeah, let's let's do that. So Dave,

let's bring you in here um and turn to you on the question of agentic in particular and the way it's going to intersect with or or sort of augment the

roles that we humans play today in our respective organizations. Talk to us about what we

organizations. Talk to us about what we might expect to change as a Gentai begins to gain traction and become more widely deployed.

>> Yeah, sure sir. I mean, and I work with lots of different organizations at different parts of their journey. Um,

it's interesting Clemens just said, uh, the palaton, if you missed the palaton, so you know, we've got the status quo, which is people who are, you know, potentially becoming more familiar, um, with things like chat GBT, but even for

the status quo, everyone else around you is going to start using AI to generate a lot of other stuff. So, if you are just here as a worker doing a certain role, there's a lot more things are actually

going to be coming at uh, coming to you.

So your productivity potential is actually probably going down from from 1x cuz just more stuff is coming in. You

need to process more stuff.

Where a lot of people are and where a lot of people have been experimenting is the kind of augmentation of an

individual's work. So the kind of

individual's work. So the kind of co-pilot model where someone is coding and they're getting assistance from an agent or someone's doing marketing and they're getting uh the collateral that

they're building is is being generated and so on. But these augmented processes they don't really bring an enormous amount of value uh especially compared to the investment in tech that's

required to get them. The processes are already you know somewhat optimized or have been optimized for years. We've

been trying to improve these processes for a long time. So the real value that you can get from augmentation is is realistically quite limited and

it's probably only going to um just counterbalance the fact that a lot more stuff is coming in. You're doing a lot more marketing because a lot more contracts are coming in or you're reviewing a lot more code because a lot

more code is being written by models. So

you're just kind of keeping your head above the water. The organizations that that I'm working with and we're working with who are really starting to do something exciting are the organizations

who have been kind of rethinking processes entirely and building these digital agentic factories. So what do these factories mean? So you look at you look at a process and you look at what

it is you're actually trying to do and solve. Work out what can be offloaded to

solve. Work out what can be offloaded to agentic systems. So let's say that I'm working in a finance department and I'm doing reconciliation of purchase orders

to invoices. That's a large volume of

to invoices. That's a large volume of high toil work that actually follows a fairly predictable pattern. I have an invoice, I try to classify it, purchase order, I try to classify it. This is

ideal to be distributed. But I don't want to just um I don't want to take that and kind of run the process as is.

Instead, I want to get my experts to govern and organize this process, distribute it to a factory of Agente workers, and then, and I think this is where it's it's particularly interesting

and where a lot of people are not uh [snorts] looking at this yet. It's not

just running a factory or managing it.

Every time this factory is running, with every iteration, there's an opportunity to improve your institutional data or processes to create feedback loops. And

it's when you start to do that that you really really start to see the opportunities.

>> Super helpful. And let's just if we could get a little bit more concrete about what that looks like in practice, Dave.

>> Super. So this this is a screenshot of three iterations of a process that I've been working on with a financial institution. And it started quite

institution. And it started quite simple. It was a um development of of

simple. It was a um development of of payments technology uh process which is a digital factory processes. We have uh initial specifications and then we get

the agents to write technical stories, do technical designs, write code, do multiple iterations of the code, run tests and so on and this kind of fans out and then we do evaluations and bring this back in. The interesting thing

about this is this factory approach actually uh runs overnight. It's a ton of work. And then that leaves the team

of work. And then that leaves the team of experts the entire day, the next day to evaluate what's happened. But and and then this is where it gets really really interesting. Not just evaluate what's

interesting. Not just evaluate what's happened, but actually take this as an opportunity to improve the institutional knowledge. If something doesn't happen

knowledge. If something doesn't happen the way it's supposed to, rather than just kind of correcting here or there, actually improving things like standard operating procedures, processes,

documentation, and so on. So that the next time that factory runs overnight, the data is better, the standards are better and that's the standards that are not just shared across the agents but

vendors, internal teams and so on and so on. So with this factory running, not

on. So with this factory running, not only you actually running this process, getting the AI to work on much of the high toil tasks and distributing this across many many many many different

actors, but you're also creating this flywheel that improves your institutional data and kind of has a compound interest effect. Every time

you're running it, the data gets better.

>> Excellent. Clemens, do you want to add in here? Do you have a specific example

in here? Do you have a specific example you can share where you see this happening?

>> Yeah, h happy to. So, um well, we we will probably be showing you a little bit more technology in this call that you will typically see in a in a McKenzie call. But um but kind of just

McKenzie call. But um but kind of just to take one concrete example that is driven from uh from strategy in a in a large bank. Most of the large

large bank. Most of the large enterprises are now moving into doing this where it matters the most. In a in a normal bank uh you would be spending

up to six to 8% of the cost base on kind of financial crime. And the onboarding process uh is a a very good pro kind of

example of a long complicated process.

If you want to onboard a a a corporation, a wholesale kind of of relationship into a bank, this is something that can take months. And this

is an a process where you have kind of very very m many kind of complex steps.

It's everything from profile initiation, enrichment of of of kind of the customer profile, the initial risk assessment.

You will have to do many complicated uh searches to kind of create the ownership structures uh of beneficial owners uh kind of do

all kinds of external adversal media screening internal blacklist lists and and there's a number of things that we actually have to kind of go go through

and this needs to be done at a a very high accuracy. You need to call on

high accuracy. You need to call on humans in the right kind of of process skills. But but this is is no longer a

skills. But but this is is no longer a process where you're just using a a tool to help you in in individual steps. This

is a real end to end transformation of a process. And this is complicated to do.

process. And this is complicated to do.

And the industry has not yet given large enterprises all of the tools needed to do this. And and that is why we are

do this. And and that is why we are seeing in the very largest institutions they are building their platforms to to do uh these kind of things to be able to

create processes that kind of go across these things. And if you think about any

these things. And if you think about any enterprise um today and this is is going to be the reality for for any enterprise they need

to break out of siloed and horizontal implementations. They will all be

implementations. They will all be relying on um existing applications of kind of things that are doing their CRM or things that are doing their ticketing

systems or their ERP system. They will

have a number of of own developed systems and for any end to end process you'll have to use resources from many

of these these things and and this is what has been holding back many large enterprises and has caused a little bit the lack that we have seen in in

adoption in businesses compared to consumers. this is just much much much

consumers. this is just much much much much more complex and uh that has forced us to to really innovate and say how do we actually solve this this problem? I

mean how do you break it down? How do

you make sure that it's secure? How do

you make sure that you get the right evaluations and and and kind of for something that really needs to be enterprise?

Um that is that is that's an unsolved problem and uh that's why we would like to introduce you this concept of the gen AI mass. We will be showing you a bit of

AI mass. We will be showing you a bit of technology the next uh five minutes. Uh

but but this is actually the realities of what is now being implemented in in some very very uh vast institutions.

>> Yeah, that's right. Uh it's a topic that's close to my heart and because maybe six months ago, even 12 months ago, a lot of the clients we were working with and the the technologists I

was working with were saying, should we use platform X or platform Y for ouric work? Um,

work? Um, >> and that was just never really the right question because one platform, one technical solution, one product is is never going to solve everything in a

complex enterprise. Like like Clemens

complex enterprise. Like like Clemens mentioned, you know, to break out of of silos, you have to distribute work across many many different systems. There is not I mean, I'm talking from a

tech perspective as well. There's never

going to be one system that solves everything. In fact, the complex stuff

everything. In fact, the complex stuff of course is across many, many different systems. So this idea that I should pick one platform or another platform or kind of have a bake off between their

capabilities, it's kind of always absurd and and there's always one that the you know software suppliers are always trying to push for. You know,

there's one one software solution that will rule them all, but it's never one solution. It's it's how you bring all

solution. It's it's how you bring all this stuff together, orchestrate it, govern it, make it observable, manage security across the stack. And if you if you're not thinking about this kind of

distributed architecture, um you're never going to be able to break out of these little silos. You're never going to be able to run workloads that that span different parts of the of the organization. And that's what the idea

organization. And that's what the idea is behind a kind of aentic mesh, which is really that the technical systems that we work with have always been big

complex distributed systems, message cues, CRM solutions, banking systems, uh, [clears throat] all of these different things, the systems you're building yourself, the off-the-shelf

solutions, and so on. We've always been having to try and find ways to integrate them. As soon as we add Aentic AI, we're

them. As soon as we add Aentic AI, we're just adding more nodes to this big complex mesh and we're making it less predictable and uh less deterministic.

So obviously, you know, more distributed and more complex.

Um so I just recorded yesterday a little bit of kind of some of the stuff that I'm working on with my team at the moment that we're that we're building with clients, which I think kind of brings

this to life a little bit. It doesn't

really matter the technical details of what you're about to see. Um, we'll just kind of like briefly talk over it, but I think it's important to say that this is this is real. It's happening. It's what

we're building with clients. And in no cases where it's valuable is it locked into one system. So, what I'm seeing on the screen at the moment is is Google

Cloud. Google Cloud I've loaded in a

Cloud. Google Cloud I've loaded in a load of data um lots and lots of unstructured data. And I I use this in

unstructured data. And I I use this in my day-to-day work because I ask my agents that are running on Google Cloud to actually analyze some of the things I'm doing against protocol documentation, hundreds of pages of

specifications.

That's great. Google's very, very good at searching through lots of unstructured data. Um, it's far easier

unstructured data. Um, it's far easier than me building, say, rag systems from scratch. But that's not going to solve

scratch. But that's not going to solve all of my problems. It's great for the use case that I'm using it for. It's

great for many other use cases as well.

Um, but I'm also running a lot of infrastructure. I create lots and lots

infrastructure. I create lots and lots of test environments. Um, so I use Azure. I use AWS. So I've got AWS here.

Azure. I use AWS. So I've got AWS here.

So I also run agents on AWS. I run them to manage my AWS infrastructure. Um, AWS

is very good at infrastructure. Um,

except in the last 48 hours. Uh, so I'm I'm going to be using AWS agents as well. I'm going to be running things

well. I'm going to be running things across different cloud environments. I'm

going to be running things on data centers. I'm going to be running things

centers. I'm going to be running things on airgapped environments. So, what does that mean? It means that the tech I'm

that mean? It means that the tech I'm using, it needs to be portable and it needs to be standards focused because if I can't run things in different environments or if I can't interconnect

things, I'm never going to be able to do anything that gets outside of these little silos of these uh little horizontal solutions. To do something

horizontal solutions. To do something that's truly transformative, I need to span all sorts of different types of data. Much of my data is going to be

data. Much of my data is going to be resident in certain locations. I may not be able to move it around. That doesn't

mean that all these different technologies I have to pick and choose one or the other, but I need to be able to bring them together and that means I need to be able to govern them. What

we're seeing here is some open-source technology um that's available and some of our clients are using and many people are using. Um we open source this

are using. Um we open source this because we don't really think that the kind of software that integrates technology is um is as important as a methodology like people to go and use

this software. This is treating

this software. This is treating different agents running across different hyperscalers and different uh environments as as resources in a distributed system. We're actually using

distributed system. We're actually using a technology called Kubernetes which most enterprises have which is really for running hardcore enterprise

workloads resource managed um isolated from a security perspective governed from a security perspective. It's for

managing kind of micro service chaos where people are building many many different systems. We don't necessarily trust all of them. We need to govern them. We need to manage them. This

them. We need to manage them. This

provides a great kind of lowest common denominator environment to run Aentic workloads across different cloud providers using all sorts of different

technologies. Um, and this has been

technologies. Um, and this has been incredibly powerful because, you know, any institution who's working with with this tech and trying to solve problems is going to be using lots of different

tech. And if you're boxed into just one

tech. And if you're boxed into just one thing or another, you'll always be dependent, beholden, um, and so on. And

and also you need to build this stuff.

You need to get good at building this technology. Uh because you just can't

technology. Uh because you just can't buy these capabilities at the moment.

They have to be built institutionally.

Um which is something that you know obviously uh super super passionate about.

>> Great. Okay. That is so interesting to see sort of in action. Um, and I think it points to this issue that you've both

alluded to that Gentic really changes the calculus on AI transformation efforts, some of which or many of which may already have been in motion at

different organizations. Clemens, you've

different organizations. Clemens, you've said in your research that a reset is necessary to realize the value of AI at

scale. Bring that to life a little bit

scale. Bring that to life a little bit for us.

So uh we kind of we see that that to do this effectively we we first of all this technology is going to change our lives

in the next five years. It is going to do a fundamental shift in many industries.

uh it is uh it is something that uh we really really have to go into and and for many industries this is the main productivity promise that is on the

horizon for the next two three years.

So, so embedding this in your strategy, how are you going to actually do this?

What is the productivity promise that you have? What are the big areas where

you have? What are the big areas where this can help you? That can be in how you deal with your customers. It can be in some of the major cost areas or it

can be in your innovation. Personally,

the whole product creation innovation is probably where I am most enthusiastic these days and I I think it's going to be a front runner. But also, if you're

e-commerce uh company, uh this is going to change everything around how customers are going to find you and potentially be with you and and and pay.

So, so kind of of make the strategy, consolidate your bets, introduce some uh platforms that you work off. It

is not going to work for any company to try to kind of innovate on 50 different platforms. So, consolidate the bats.

uh do the workforce planning kind of of of it's not like uh the world is full of people that have done this before but kind of you'll have to kind of train some and bring in in some uh it is going

to change the operating model. It's

going to change how we task teams but also kind of the size of teams are are simply becoming kind of smaller. Uh I

don't buy into the future where we will have a a workforce that is much much smaller. I actually think and I already

smaller. I actually think and I already see that that we will be able to do so much more but it is going to be a huge reinvention technology. This technology is

technology. This technology is difficult. It is not an easy technology.

difficult. It is not an easy technology.

we we begin to see patterns that this is a real extension of compute. So kind of of using all of the best practices from enterprise technology that we showed you

a bit in this call is going to be the core to kind of get it manageable and and secure and safe. Um, and then lastly, data is is it can help you with

structuring your data, but but kind of ultimately the data that you you put in is is kind of what you and then lastly the the whole adoption and the scaling

paradigm is uh is is is going to be super super important. So treat this as a core element of a strategy, do the

prioritization and be ambitious.

that that would be my my parting advice.

>> Thank you, Clemens. Okay, I'm mindful of time. I would love to fit in at least

time. I would love to fit in at least one question from our live audience.

Let's take this one. Um, Clemens, you mentioned that this is not an easy technology.

What are some of the biggest risks that organizations should be watching out for as they embark on these transformations?

So I mean there are very concrete risks if you're relying on this technology that in in nature is probabilistic and you don't do uh kind of your

implementation in a in a in a managed way. Uh you can be putting kind of of

way. Uh you can be putting kind of of yourself and customers and employees at real risk. We have seen cases already

real risk. We have seen cases already where there has been physical risk on people because of of bad AI translations.

uh and uh and and therefore kind of of this needs to be implemented in a in a secure and evaluated way and in all regulated industries uh this will not

fly if you're not having the right governance kind of policies the right process control both in process and and and out of process. So um it's a it's an

amazing promise but it needs to be done right.

>> Dave, anything to add quickly there?

Yeah, I mean I' I'd kind of almost echo both points. You one, be ambitious.

both points. You one, be ambitious.

You've got to go into prod and build things. Um, it's the only way you're

things. Um, it's the only way you're going to get good at doing this. But to

treat every agent and AI system as a potential bad actor that is incredibly creative at finding uh ways to do things. um use the learnings we've had

things. um use the learnings we've had for a long time around how you you think about zero trust systems, how you limit the blast radius of potential incidents,

how you quantify the statistical correctness of a process because a process doesn't have to be perfect, but it has to be quantifiably better than what was there before. And the risk

needs to be managed. Um and you will only do this by building things and getting it out there. So limit the blast radius but force yourself to build something something real.

>> Okay, that brings us to time. This was a super interesting discussion. Obviously

a hugely consequential topic. Clemens

and Dave, thanks again for joining us.

And as always, many thanks to all of you in our Mackenzie live audience for being together with us today. Please visit our website mckenzie.com for a replay of this event and of all previous McKenzie

lives as well as for more on our research and work on Agentic AI. And a

heads up that for our November event, we are going deeper into the dynamics of Agentic within a particular industry.

Join us November 18th for Agentic AI and the future of travel with Jules Cely and Kelly Angererman. See you there. Have a

Kelly Angererman. See you there. Have a

great day or evening and be well.