What we talked about
Karthik Krishnamurthy is the CEO and Founder of Ascendion, an AI-native engineering company operating across 12 countries, including teams in Mexico, Brazil, and Colombia. Before starting Ascendion, he spent nearly two decades at Cognizant in senior leadership roles across digital, analytics, and AI. He is also the author of “AI Arbitrage Is The Next Frontier.”
Show notes
Karthik Krishnamurthy has been working in AI for 25 years and has watched every wave of excitement run into the same wall: nobody can show the ROI. A machine learning project he ran in 2015 to identify patients at risk of opioid drug-seeking behavior saved a healthcare company $60 million and created proactive interventions for 85,000 people, and that result has driven everything he has built since.
What we covered
- Karthik draws a sharp line between AI-first and AI-native companies. An AI-first company still spends time debating why and what to use AI for. An AI-native company only thinks about the how, because the why and what are already settled, baked into how people are hired, how marketing campaigns are run by agents, and how almost all code is written through internal agents.
- The wrong question when starting an AI transformation is “how can I implement AI?” The right question is “how do I deliver four quarters of revenue acceleration?” or “how do I hit 400 basis points of EBITDA improvement?” The power of AI today, Karthik said, is that many of those ambitious goals can only be achieved by applying an AI lens.
- Enterprises cannot simply hand every developer an AI tool and tell them to go. Ascendion’s approach is to define a new end-to-end software engineering process, covering quality engineering, test management, business discovery, design, that is built from the ground up to support an agentic framework, then physicalize that process on a platform so agents and humans work together inside existing compliance and security constraints.
- A client came to Ascendion having thought about building a data marketplace for months but never starting. In one platform session, no business people in the room, they produced a lean business canvas, must-have and nice-to-have requirements, and epics broken into stories. The client’s reaction: “My Lord, this is crazy.” Karthik frames this as the difference between incremental improvement (10-20% sprint velocity gains) and AI arbitrage: orders-of-magnitude shifts that actually move revenue lines.
- The rarest job in the next year, Karthik predicted, will be agentic delivery management, project managers who have actually delivered projects where human labor and agent labor worked together successfully, at speeds three to four times faster than human-only teams. He compared it to the moment data scientists became the hottest role in tech, and said those holding this experience will be “the hottest thing in the market.”
- Karthik told listeners to take screenplay writing workshops. Storytelling, being able to define the outcome you want, articulate it clearly, and bring others along, is now one of the most important skills in an AI-native organization, because guiding an LLM to the right result and then selling that result to stakeholders require exactly the same capability.
About Karthik
Karthik Krishnamurthy is the CEO and Founder of Ascendion, an AI-native engineering company. He spent nearly two decades at Cognizant in senior leadership roles across digital, analytics, and AI before founding Ascendion to help large enterprises connect AI to measurable business outcomes. He is the author of “AI Arbitrage Is The Next Frontier.”
- LinkedIn: https://www.linkedin.com/in/karthik-krishnamurthy-8117424
- Website: https://ascendion.com
Episode 124 of the PreVetted Podcast.
Full transcript
Federico Ramallo (00:00) Welcome back to the pre-vetted podcast where we spotlight extraordinary people and remarkable talent reshaping our world. Today, we are joined by Karthik Krishnamurti, CEO and founder of Ascendion, a company focused on disrupting and transforming how we build software by being AI native from day one.
He brings deep experience helping large enterprises connect technology to real outcomes. Earlier in his career, he spent only two decades at Cognizant. He had a senior leadership roles across digital analytics and AI. Last but not least, he’s the author of a book called AR Arbitrage is the Next Frontier. Karthik, welcome to the show.
KK (00:50) Glad to be here, Federico. Quite exciting.
Federico Ramallo (00:53) I am honored to have you here today. ⁓ So tell us what problem did you want to solve when you decided to start Ascendion?
KK (00:55) It’s my privilege.
Yeah, I think…
the problem that I really, really wanted to solve.
was to land the plane of AI.
All right. You know, I’ve been in AI for about 25 years, seen various sort of curves ups and downs and so on and so forth. And every time that we would get super excited about the journey that we were taking, the question of, okay, all this is great, but where is the return on investment? Where is the ROI would always come up? And you know,
I think it was Satya that said it beautifully. He said that, you know, if a lot of our, if let’s say a couple of years from now, if all of our conversations around AI still happens to be done by the people that are producing it, then I think we would have lost, we sort of lost it, right? Lost the whole point of it. And so I started Ascendia.
with the primary goal of helping organizations truly realize the potential and the outcome that AI promises. And I came to the conclusion that it’s not possible to do that with just building another fantastic piece of technology or two. The speed at which technology is moving right now, it’s just getting better by the day, better by the minute actually.
The challenge that needs to be solved is a business model, a process model, a culture model that one needs to put their head around and help clients and buyers and organizations figure out. And that’s what that was the primary reason I started Asenia.
You know, somebody asked me the other day, know, five years from now, you want your, if somebody went to one of your clients and said, give me the words that would describe a Sandy on to me, there were only two words that came up. It was outcome assurance, right? We just want to be known by our clients as those that they would call when they were at a place where they really had to drive a return on investment on.
artificial intelligence, which means doing all of these various things and helping them do all of these various things.
Federico Ramallo (03:06) Right. I think that with all the hype that is around AI, it’s refreshing to see how you have a down-to-earth approach based on outcomes and not smoke.
KK (03:18) Yeah, listen,
you know, the reason I’ve been so obsessed with that part of the AI conversation sort of dates back to 2015, 2016, when in an earlier life we were working with a healthcare company to help them figure out a solution for the opioid crisis. If you remember, you know, 14, 15, 16, the opioid crisis was raging during that time.
And this was a healthcare company which was really trying to solve it at both ends, right? Both from a health delivery perspective as well as from a cost of health delivery perspective. And we created a machine learning algorithm through which we were able to proactively identify patients from falling into drug seeking behaviors.
based on refills and so on and so forth. It was a small project. It wasn’t much. It didn’t mean much to me from a revenue perspective, but it meant so much to me from an impact standpoint because it saved $60 million for the client, but more importantly, it saved the lives of 85,000 people. For 85,000 people, this company, this healthcare provider was able to create proactive interventions and stop them from becoming
Federico Ramallo (04:33) Wow.
KK (04:40) drug addicts, right? That thought has always stuck in my head, which is when artificial intelligence is leveraged in a way where you have a very clear idea of what is the outcome that you want to be able to produce and pointed in that direction and sort of enable the universe around it to make it possible, which is the business model, the process model, the culture model, the people model and so on and so forth.
we can truly impact lives, we can truly make the world a better place. And so that thought’s always been in my head. And so it sort of goes back to that, right? So helping organizations really sort of land this plane in a way where you are able to realize the business and the human impact that I think it can actually develop.
Federico Ramallo (05:26) Right, right. Yeah, what’s the point of building a bridge if nobody’s going to use it? If it’s not going to allow people to go to school or go with the lives, right? ⁓
KK (05:35) yeah. And what’s the point of building
a bridge to nowhere?
Federico Ramallo (05:38) Right,
right. So I think that as an engineer, our first purpose is to build cool stuff, But then a deeper, more meaningful purpose is to change people’s life, right? That they use what we build to improve their lives.
KK (05:55) Agreed. Agreed.
Federico Ramallo (05:57) So why did you build Ascendion AI native from day one? And what does AI native mean to you?
KK (06:05) AI native fundamentally means you’re not getting up every day trying to figure out why you need to use AI or.
what in AI you need to use. The only thing you’re thinking about is the how.
Right? That to me is AI native. A company that’s AI native is essentially wired with artificial intelligence mindset and thought from day one, which means that people inside the organization are leaning forward on using it. They look forward to using it. They are finding ways to use it as opposed to wondering what to use it for.
or why should I even use it? That is the difference between an AI native company and an AI first company. An AI first company, there’s still a lot of work to be done on the why and the what, and you would put AI as the first option, but then also be perfectly okay with the second and the third and the fourth options. But an AI native company is built with artificial intelligence.
at its core. What does that mean? You hire people that look forward to using artificial intelligence. You hire people that look forward to learning this. You hire people that come in with the core skill sets of curiosity, of communication, of math, of various different skills that are required for you to be able to really leverage artificial intelligence for generating outcomes.
And we’ve put it in the way that we build the company. Like for example, the day one when I joined the company, I wrote an AI ML algorithm that would instantly categorize my business into the various different types of things I was doing. Whether it was product engineering, whether it was platform engineering, it wasn’t just somebody sitting somewhere that decided what it would be. It was an ML algorithm.
that did that, right? And then you take it one step further where, you know, if we are running marketing campaigns today, my marketing campaigns are all run using agents. If you look at my own CIO, I’m not even talking about the work we do for clients, just the way that the company is being set up, the work that he does, almost all of the code that we write is written through our agents.
And so there is a lot of wiring that’s happened inside the organization across the various parts of the organization. So that ultimately we can confidently say to our clients that we eat our own dog food and you know, and these are the ways in which we would actually be delivering AI native solutions for you. Where everything that we deliver ultimately is either driven by AI or driven for AI. Right? For example, if you are
If you’re building an AI ML application or you’re leveraging an agentic process for you to be able to transform software engineering, that’s all where we using AI to drive something. Now, there are also many scenarios where clients ask us to come in and modernize platforms, modernize products, because it fits into a broader artificial intelligence journey that the client wants to get on, right? They want to build their platforms and their products to be ready for AI, which wasn’t for that as well, because of our AI native approach.
So when you design a platform, you design a product, you’re always thinking, how can I leverage that in a way where I make it possible for some AI ML algorithm to be able to use what is coming out of it. And so all of those pieces sort of go into what I think is AI native. But ultimately, it’s that. It is an organization that gets up every day only thinking about the how, not necessarily the why and not so much of the what because the why and the what.
to a large extent, we’ve already sort of taken care of. We’ve taken care of the why by bringing in people that are already on board with the why. And then we’ve done enough work inside the organization to define the what, where are the areas you could apply artificial intelligence and so on and so forth. Now it’s just about doing more and more of the how. That’s the AI native.
Federico Ramallo (10:05) Right, and you’re using AI as a means, not as an end.
KK (10:10) AI is never the end. I think the, you’re right.
Federico Ramallo (10:12) Right, but that has been the, know, people ask,
how can I implement AI? And that’s probably the wrong question.
KK (10:21) That is the wrong question. How can I actually deliver on 400 bits of EBITDA? How can I deliver on four quarters of acceleration and revenue? Ask a tough question. Do not ask a question for which the answer is incremental. Ask a question for which the answer is substantial.
to me, we have to lead with that. the way the power of artificial intelligence today is that in many cases, the only way that you can solve that problem is by applying an AI lens.
Right?
Federico Ramallo (10:56) And also, as far as I understand, in enterprise, you also have the challenge that putting a generic LLM into the system and let it run wild doesn’t work. ⁓ You cannot ask higher level questions without providing the whole foundational infrastructure and tuning, right?
KK (11:09) That is very true.
Well there are two pieces to it right?
There are two pieces to it. think most enterprises are really really worried about LLM proliferation. There are lots of LLMs that are available and as you can, know, we all use LLMs in our personal lives. you know the truth is within enterprises there are many people that are leveraging LLMs already and what we come across when we speak to our clients is you know we need to have a
streamlined mechanism through which I can govern, standardize and create enterprise workflows around leveraging AI agents, specifically in the software engineering space. So clients really love that because it’s the type of discipline that they want, which is sort of where we really come in, right, to the point of outcomes. You cannot generate a software engineering outcome.
If the way that you’re thinking about AI is to give a tool to every developer and then say, go, you know, go crazy with it, right? It has to be done within the context of a software engineering process. It has to be done within the context of how you do quality engineering or test management frameworks. How you choose to do business discovery sessions. How do you connect one to the other and so on and so forth. So defining the software engineering process end to end.
Federico Ramallo (12:14) Right.
KK (12:34) in a manner where it is supportive of an agentic framework, where you can actually bring agents forward with humans in the loop is incredibly important and making that sit within your current definition of compliance, current definition of security, current definition of industry mandates. And of course, within your current budget, all of that is incredibly important for us to be able to make this real in an enterprise. So,
That is why we believe that enterprises really require this type of agentic streamlining, if you will.
Federico Ramallo (13:10) Right. I mean, it’s completely different when you’re doing a startup, you’re starting with a fresh code base. You can use a copilot agent for coding, and you can use the average coding that is in the world, and it’s going to work because you can follow the well-known patterns of the frameworks. But on enterprise, you have…
customized solutions throughout your organization, right? So you need to teach the agents how to work with those specific business use cases.
KK (13:51) 100%. In fact, one of the areas that we spend time on the most is contextualization. Contextualization is massively important in the context of enterprises, just given the many, many, many, many years of business definitions and ways in which things work and so on and so forth. And it is incredibly important that we get those domain models right.
And that becomes key to making something work at an enterprise level.
Federico Ramallo (14:15) Right.
How much of… How good are the LLMs now to be able to build abstractions of… When we build models, we build objects, we build classes, depending on the framework, you call it different names, but those are basically abstractions that allow us coders to… You have a problem in the world, you build a way to interpret that into…
something that we can abstract to it, right? How good are the LLMs at doing that now?
KK (14:51) They’re massively powerful federally, but that’s the reality. Every day they’re getting better. There’s absolutely no doubt about it. But the real question here is every enterprise works differently. So one cannot make the assumption that the way it works in one enterprise, it would just naturally work in the next enterprise. And so that is where I think there’s still a lot of work to be done, if you ask me.
Federico Ramallo (15:14) Right. So we’re so I’m wondering about the difference between consuming AI and producing AI within organizations. And what is how do you differentiate these two?
KK (15:28) Yeah, listen, think so producing AI is everybody that produces.
model or an application with a model in it, right? And so there are lots of people building agents. Many people are being taught how to build agents. But the real value is in how it’s consumed. So today, the way I see it is 95 % of almost all of the training that exists is about helping people build agents, helping people, you know, create workflows, helping people think about
You know how to use the tool this way and how to use the tool that way and so on and so forth But the truth of the matter is that In the real world Ten percent will produce it and 90 percent will consume it right and I’m talking about consumers in terms of product designers quality engineers Or the process the process itself, right? It may not just be the people
But even the process, if you decide that you want to have a human in the loop model, you may also have to define a software engineering process where the agents get consumed. we spend a lot more time on defining consumption patterns, especially across industries, across domains, understanding what types of process models can drive those consumption patterns and how can those process models be agentified.
through a platform based approach, right? So that, you know, I’m not in the business of building an agent. I’m in the business of delivering on an outcome for a client or helping a client deliver on an outcome by leveraging an agentified method. And so there, what we end up doing is we end up spending a ton of time with the people that are expected to consume AI agents or processes that are expected to consume AI agents.
and really making that ready for the consumer.
Federico Ramallo (17:17) So you work more on helping people understand the technology than building custom technology, custom agents for the enterprises.
KK (17:30) No, I just reframe that. I don’t think this is just a training question. I think this is a helping organizations be able to get the most out of an agentified, out of the value of agentic AI. And what we do there is we help organizations define a new software engineering process, because the old software engineering process that they have of how they do
QE and how they do development and how they do design and how they do product ideation and so on and so forth is not built to be able to consume agents at an enterprise level. Individual people may do it differently. There’s not one standardized way of doing it. And so we help people sort of define that process model. And then we physicalize that process model on our platform. Our platform actually helps you physicalize that process model. It already has a golden set of agents that are made available. And you know, it is linked within.
custom-built workflows, pre-built workflows that can be deployed, and so on and so forth. And then the humans are in the loop, right? The humans are able to make sure that this thing happens, right? That to me is sort of the enterprise way of thinking about this. So there’s still a lot of agents that need to be built. There are people that will build the agents, right? You will have people building agents. You will have pre-built agents that we make available. We’ll have teams from our sides that will also build the agents, but we just don’t stop there.
Federico Ramallo (18:30) Bye.
KK (18:50) We also define the process model. We define a standardized process model. We then have our platform that physicalizes that process model, makes all of these agents that people are building available through process packs and so on and so forth. So it makes it, we really focus on the consumption side. You understand what I’m saying? So we really focus on the consumption side. A big part of this, like you just said, is also the change management. So it’s not just about the process. It’s not just about that. It’s also about the training. It’s about the cross-skilling.
We do that as well. We help clients sort of rethink their job families, rethink their career hierarchies. How should people get promoted in this new world? Right? What types of job roles should people have? Those are all very important questions that organizations need to think about. Startups don’t worry about that today. Organizations do, right? They have to worry about career hierarchies, they have to worry about promotion processes, they have to worry about all these things. How does all of this look in an agentified world?
Federico Ramallo (19:30) that’s a really interesting question.
KK (19:49) Right? So.
Federico Ramallo (19:50) Right, because
currently people, organizations and people in these organizations don’t know what they don’t know and they don’t know how to implement AI because there is no pattern to follow yet. It’s something new, right?
KK (20:05) Yeah, listen, very
recently, a very senior leader basically made an announcement that if you don’t learn AI, you won’t be with the organization anymore or something like that. I don’t want to quote names here. I think that’s a bold statement to make for a CEO. But what is underlying that, I think, is also what does the leadership actually do?
to enable the individual to be able to see a career for themselves in this journey. What are those career paths? What are those role families, job families? We got to define all of that so that we’re able to take the people that we want on the right. It’s not just about learning and using AI. The thing is there is a business model and a process model question here. We got to tackle all of that. And that is why I think that people, I think I was talking to somebody the other day, they were saying, it’s a people problem. I just don’t think it’s a people problem. I think it’s a leadership problem. I think…
Federico Ramallo (20:40) Right.
KK (20:55) We need to have leaders that are very, very comfortable with building organizations and delivering on the success metrics of an organization, knowing that a big part of their work will be done by digital labor. And they also have to build organizations where human labor and digital labor, agentic labor, are able to coexist together. Those are the leadership questions and challenges that we have to solve. And we spend a lot of time thinking about that.
Federico Ramallo (21:17) Right.
Right, right. And I agree. I I think it’s more than a people problem. It’s about how can we build this new culture and world dynamic.
KK (21:35) And that’s leadership.
Federico Ramallo (21:37) Yes. So one thing I’ve been trying to figure out, because I can see a difference for myself, but it’s getting harder and harder to see the difference from other people, is when is a person augmented by AI or, you know, like I’m thinking mostly when they…
KK (21:39) Right?
Federico Ramallo (22:00) use AI on a personal level, right? And when they’re using AI as crutches, right? So they project themselves as skillful, but in reality is, you know, they’re over relying on.
agents, but they don’t have the core skills. They actually don’t have the core skills. It’s hard to detect when somebody actually has the core skills or not because of that,
Yeah, sure. So I can tell you how I see the difference. So I can, for instance, if I’m writing an email, can put the ideas of the email. Then I can ask the agent, help me write this email better, help me fix the typos. And that’s a way I can augment AI to build a better email. The outcome is better.
Or I can do it in a shorter time, right? Or help me, you know, shorter this idea so I don’t use too many words, right? Or whatever that might be. So the outcome is better. But it’s me guiding, right? ⁓ It’s my ideas, just, you know, I get an assistant. But it’s different when I ask AI an open question and I use those concepts to…
KK (23:00) Mm-hmm.
Federico Ramallo (23:13) move forward, then it’s the AI that is actually driving with the strategy, with the ideas on how to move forward. So from a third party, from outside of me, you could see two responses, and it would be hard to detect if it’s one or the other. I’m wondering how can we start looking for those differences, and if you experience that.
KK (23:39) The question really is why should we look for those differences? So the question that we have to ask ourselves is should we even look for those differences? Because ultimately, if we agree that artificial intelligence is an extension of ourselves and is a reality that’s going to exist in our life, then
Why should we even look for it? I think there is a very important trust equation here that one needs to sort of figure out. So to me, I sort of think about it a little differently. The thinking that we have in the organization as we say, listen, there’s an outcome that you ultimately have to achieve. If you’re able to deliver on that outcome,
through a mechanism that requires you to use artificial intelligence and agents. And as long as that is validated and compliant,
Why does one need to worry?
You get what I’m going with this? So.
Federico Ramallo (24:32) Right, because at the
end of the day, you are achieving the outcome.
KK (24:37) achieving
the outcome in a securitized and compliant manner and you’re able to do that. Now again, there is a change management around this. Now the question here to me, Federico, the question that you’re asking is a question of change management. It’s a question of how can I ensure that I’m not just making you use AI, but you yourself want to use AI, right? Or the people that are consuming it want to.
want to see you use AI. That’s a change management question. And for change management, a lot of this is about, you for example, you do contextualization, you provide contextualization lineage, you provide a lot of lineage, right? You have to have mechanisms through which you win somebody’s trust over during the change management process. Like how would you do that? Let’s say that you’re trying to go and tell somebody about something you’ve learned somewhere. What do you say? You say that and then you say what are the sources from which you got it?
You know, like when people make a presentation and they provide a certain output, they clarify the sources from which you got it. So to a large extent, lineage and contextualization, when you do contextualization and models is about you being able to to your beautiful podcast name, pre-vetted sources. Right? And so there are ways and means by which you can actually do that to deal with the change management process. But ultimately where…
you know, the highest value of the existence of this conversation will be one where people are not so worried about trying to make the differentiation, but except the fact that at the end of the day, this is one unit, human labor plus digital labor and measure it on the outcome that’s being delivered. Are you able to deliver the outcome or not? Right? And so that is the journey that we help.
organizations, enterprises, and people get on.
Federico Ramallo (26:28) Right, right. Yeah
KK (26:30) And it’s difficult, right?
Because you you talked about this notion of people using it for themselves versus people using it in an organization. When you’re using it for yourself, you’re ultimately accountable only to one person, which is you. But when you’re using it in an organization, you’re trying to make it happen through a structure or through some sort of a process or a discipline. And there it can get more
complicated. So we have to go the extra mile to drive change management, trust mechanisms, do contextualization lineage, do a whole bunch of things like that, that actually gets people more over the wall.
Federico Ramallo (27:07) right, right and by assuring the outcome then It doesn’t matter which tool we use whether we use an agent or whether we use a manual manual labor at the end of the day the assurance of the outcome reduce the risk the the process for the business, right
KK (27:30) Exactly. And so the outcome itself is listening and it depends on how you choose to define what outcome is, right? Your outcome could be speed, your outcome could be accuracy, your outcome could be any of these things, your outcome could be revenue, it could be EBITDA, it could be customer experiences, it could be anything that people are building software and technology today for. But ultimately, my only sort of conversation with clients is measure everything on the outcome.
Right? Yes, there is a change management path that one needs to take to be able to get to that point. That is a very important journey that we have to take personally and from a corporate perspective. But think outcome back.
Federico Ramallo (28:12) Right. And how do you manage the agents are, they can provide different answers, right? So they could hallucinate and they can come up with different, they can make decisions, right? Whether they could be.
higher quality or lower quality decision process, but at the end of the day, it’s a non-deterministic process, right? Whereas on the business, we want a more deterministic process, right? How do you manage that to reduce the risk for enterprise?
KK (28:50) It’s a good question. think it all boils down to how you choose to design your agentic framework. Like for example, in our world when we design and I can’t talk for all products here, but at least on our platform, right upfront when you design the agentic process, you get to define the degree to which you want it to be deterministic versus probabilistic. You can set your temperature controls. You can do all of that upfront. So it is a very, very important part of
how you choose to design it, which I think is incredibly important to be able to define deterministic versus probabilistic choices that you make based on the work you do. For example, you’re, you know, we’re talking about things that need to be deterministic. Yes, from a coding perspective, absolutely. But if you’re doing product discovery, you’re trying to do idea brainstorming with product managers, you don’t want to be too deterministic. You actually want a little bit of probabilistic there, right? And so,
I think the choices we make are very important. Having an agentic design framework through which you can actually manage and control these choices are very important. But ultimately, it’s going to come down to contextualization and learning. I just don’t feel like enough organizations that decide to go forward with agentic AI and LLMs spend enough of a time on contextualization.
We need to spend a lot of time on contextualization. people have to define their own domain context and domain models through which they’re able to drive this. With the client that we’re working with, we’re spending a ton of time getting the contextualization right. We’ve got 25 to 30 work streams through which we are implementing contextualization, even before we open up the platform for everybody to use. To your point,
for some reason, we’re all party to it. Humans tend to hold technology to a higher standard than they hold themselves, which is true. And so, you know, it’s okay if the human makes mistakes. We give them like a couple of months, but the first time that the model hallucinates, it’s out. And so, unfortunately, that’s the reality that we learn to deal with. But, you know, the only way to deal with that is to spend a tremendous amount of time.
Federico Ramallo (30:48) You
KK (30:58) contextualization, cross contextualization and so on.
Federico Ramallo (31:02) Right, right. It could be statistically more accurate than humans, but one mistake and it’s aha.
KK (31:10) I
always joke about this, mean, are, you know, and I mean this only, I don’t mean this in the wrong way, you know, somebody drives a car and rams into five people, that’s on page five or six. A self-driving car happens to touch some person walking with a little bag, that’s front page news.
Federico Ramallo (31:32) Right. Right.
KK (31:33) So yeah, so we do tend to hold technology to a higher standard than we hold ourselves.
Federico Ramallo (31:38) Yeah, the self-driving cars are, I’ve seen they start driving more aggressively because now they’re more, I feel like I could be wrong on this, but I felt that they now have been able to figure out how to keep being safe and they’re safer than human drivers, right? I love driving, so you know.
I want to keep driving myself, but I also love the idea of pushing a button and on the boring sections just, you know, drive me home, right?
KK (32:07) 100 %
Federico Ramallo (32:09) And
that experience would be like getting into an airplane, It’s safer to fly than drive, right? And I think that’s what’s going on. Yeah.
KK (32:15) Agreed, Statistically that is correct. That
is correct. It’s safer to fly than to cross the road.
Federico Ramallo (32:22) Right, right, yeah, which is, it’s amazing. And there’s a lot of effort to make that happen. It’s not by magic, right? It’s just systemic process improvement that happen over and over time, right? ⁓
KK (32:34) Yeah, and I think
that will happen with artificial intelligence as well. think it’s, I mean, if you look at the cloud industry, it’s fascinating, right? I mean, the first few years of cloud, everybody was worried about security issues with cloud. Then it came to a point where people felt very comfortable going to the cloud because it would be more secure there, right? Because of the way that the hyperscalers implemented security controls and so on and so forth, people just felt more comfortable being there than having it in their own data centers.
The same thing could happen with artificial intelligence, where if you’re about biases, for example, I mean, the investment in AI, the investment in tech is so deep that one would actually realize that in an AI world, things may be more secure, things may be more bias-free and higher quality than sort of a world which is non-AI. And what was the thing that was holding people back could actually become the thing that accelerates adoption in the first place.
Right. And I think those days are not very far from us.
Federico Ramallo (33:35) Yes, this thing is evolution. The evolution of these AI tools is getting faster and faster every day, and I’m greatly surprised by that.
KK (33:45) Great.
Federico Ramallo (33:46) So when a new customer asks you to do a transformation, what is usually the advice or the steps that you advise them to do?
KK (33:58) Well, the first question is, first question we try and understand, we really try and understand a lot of things upfront, right. First is.
why why are you even doing it and the answer has to be something that’s way better than or way different not better way different than I’ve been asked to do it or those people are doing it you know things like that it has to be something that one is able to tie to a strategic outcome that that matters to the
the right? So one has to be able to tie to that. Otherwise, the transformation is just not worth getting on. The reason is because you’re not just talking about implementing a new technology. You’re talking about capturing minds and hearts. You have to change minds and hearts here. And so for that, the journey really, really needs to be worth it. It’s not going to be easy. It needs to be worth it. Right? And so how do you know that it’s worth it?
Federico Ramallo (34:38) Great.
KK (34:56) You know it’s worth it only when you know that if you do that, you’re be able to deliver on that outcome. And so the why question of the transformation is something is where we always ask people to start. Why, why not? Is there another way to do this? And what is really the potential out of the possible that you could actually deliver using this? Like for example, we’re talking to a client who, for whom we’re doing a.
massive transformation and for that person, the Steve customer officer is able to deliver revenue. Four, five quarters ahead of plan. Five quarters, right? That is something worth going after.
Federico Ramallo (35:30) Wow.
KK (35:34) And when you set that true north, that ultimately as a company, you’re able to start your revenue plan five quarters earlier and what that would then mean for the growth of the company and what that would then mean for the health of everybody in the company and so on and so forth is sort of the why behind getting on the transformation. so situating yourself in that is incredibly important, incredibly important. Then the second thing becomes
the conversation around the model, the business and the process model, right? Specifically the process model. The business model to a large extent is tied to the why, the outcome. The process model is almost always what gets in the way. That’s where you hear the conversations of, but we always used to it this way and we’ve always done it that way. And you know, the whole thing where change is at its highest.
is in the process model. Instead of brushing your teeth this way, I have to ask you to brush your teeth this way. Or instead of wearing your whatever this way, I’m asking, everything changes. And so understanding that process model and making sure that people understand what is that new process model that people need to move to is sort of the second thing that we spent a ton of time
Federico Ramallo (36:32) Right.
KK (36:56) So making sure that the why is clear and people are able to see examples of the why in the pilots and so on and so forth. And then being able to get to a conversation around the process model is sort of the first many things that we do. And as part of the process model, when you start to get a sense of culture and so on and so forth, those things become very important. What LLM choices you make is incredibly important, but it’s not the first thing.
It should not be the first. Right?
Because I’ll also tell you this, ultimately it’s in the interest of the LLM companies as well because at the end of the day, even if you think about it, LLMs make their money in consumption. When they’re consumed, they make more money. so consumption doesn’t happen if you’ve not figured these two things out.
Federico Ramallo (37:45) Right. So it’s change management with consulting, with cultural transformation to implement this tool that a lot of people have a lot of misconceptions about.
KK (37:59) And platform, and it’s change management, it’s consulting, it’s platform-driven transformation. I think that’s the most important thing. Because it’s not just about doing PowerPoint presentations anymore. People want to be able to instantiate and physicalize a lot of what we’re talking about. So that’s where our platform becomes very important.
Federico Ramallo (38:18) Right. And you mentioned something about that the promotions and the hierarchy is also refactored with this new approach. How do you see, what is your vision for how that should be approached? what would be the outcome, the ideal outcome?
KK (38:29) Yeah?
Look, the real outcome for people to me is nothing new. It’s just that this to me is an incredible way to accelerate your career, accelerate your financial health, especially because organizations and enterprises are getting very, very centered around trying to figure out ways by which they can make this real. And so what that would mean is it is a
It is a skill transformation question. It’s a mindset transformation question for people. And also from a talent standpoint, organizations need to start to think about new role definitions. Like for example, think, you know, I don’t want to give away too much of my secret sauce, but let’s just say, I’ll give you an example, right? So we believe that one of the rarest jobs in about a year from now would be agentic delivery management.
What do I mean by that? I’m talking about project managers, delivery managers that are taking responsibility for delivering on a particular project that have the mindset, the confidence, the skill sets, and the experience to deliver projects on
using an agentified method with humans and agents. And there aren’t going to be too many of them. But that to me is going to be a very rarefied skill. You’re not going to find a lot of people that would have actually been able to pull that off in such a short time. But those that have, it’s almost like the new data scientists. Remember, there was a time when data scientists was the hottest thing out there. I think agentic delivery managers would be one of the hottest things out
And so delivery managers have to start to think about themselves as agent delivery managers. And the people today that are being given the responsibility of delivering on a project where you have people and agents and you’re being made to use, people are using cloud or people are using platform or whatever they may be using. And those that are delivering those projects successfully, that’s a rare skill.
Because you know, are the things they’ve learned? They’ve learned what does it mean to make humans and agents exist together. They’ve learned what does it mean to work with a scenario, the point that you said, Federico, where the first time an agent hallucinates and the developer goes, my God, I’m just going to go back to my old way. The guy says, no, don’t do that. Stay at it. Let’s go back to contextualization. I will take care of the outcome. Don’t worry. It’s all of that, right? Understanding how the process workflows will work.
Federico Ramallo (41:02) Right.
KK (41:11) understanding where to apply the agents, all of this is going to be between agentic architecture design and agentic delivery management, an incredibly important skill set, a management skill set. To me, that’s going to be one of the rarest skills out there, right? Because organizations and enterprises are going to look for people that have delivered agentic projects successfully with human labor and digital labor together.
Federico Ramallo (41:35) It’s being able to coordinate humans and agents to a specific outcome, to a specific timeline. That’s going to be super rare.
KK (41:44) Yeah, today a
project manager builds a Gantt chart, builds an MPP, builds a project plan, or in the context of Agile, they define their scrum definitions. But now, you have to do that and you should have been able to pull that off in the new software engineering model with the agents in it. And if you’re somebody that’s been able to deliver projects, and all of these projects are going to have expectations on truly accelerated outcomes, so if a human-based approach takes you
10 months and a human plus agent in an enterprise takes you, you know, three months and you’ve been able to deliver four or five of that very successfully.
you are the hottest thing in the market. And so, agent-like delivery management for some time will be incredibly important. Now, technology will get to a point where you could have agents orchestrating agents and so on and so forth. Right? I don’t know if you know this, but we all started driving, we all drive our electric cars, right? So, self-driving car capability that came into my car,
Federico Ramallo (42:25) interesting.
KK (42:49) was maybe about five or six years back, we got it. The technology has been there for years.
Right? But for it to become truly real in the world in which it was being applied took that time. And it’s the same scenario. So agent tech delivery management to me is going to be an incredibly important skill set in enterprises for the next many years.
Federico Ramallo (43:09) Right, right. And even when the models evolve, it’s going to be, there’s going to be a need for it, for them.
KK (43:16) Yeah, and people that are producing like the journey that people will take now with job families and roles is going to be you go from being a creator, a builder to a designer, to a system thinker. You’re thinking holistically about systems, right? So there is an evolution there that I think will happen, absolutely happen. I actually, I worry a lot for the senior developers now, right?
So you’ve got entry level and you’ve got senior people with deep experience. The people that I genuinely do worry about are the senior developers, the people that have held on to their Python code for too long, who haven’t spent enough time understanding domain, who haven’t spent enough time understanding the functions in which they belong so that they can actually validate the outputs of the agents more effectively and so on and so forth.
And so it is a very important career transition that many of who we call sort of senior developers, tech leads really need to be able to make.
Federico Ramallo (44:14) I want to talk about your book, AR Arbitrage is the Next Frontier. Can you tell us a little bit what AR Arbitrage means to you?
KK (44:22) it’s sort of everything that we’ve talked about so far, which is that, you know, the old model of providing services on top of technology was one called the labor arbitrage model, where you typically found labor pools in, you know, economically convenient societies, countries, ecosystems, wherever it may be for you to be able to deliver on.
what I would call the faster, better, cheaper alternative. And AI arbitrage at some level basically is about creating a new agentified process method through which AI can unlock the next wave of efficiency and the next wave of solutioning so that you’re able to deliver faster, better, cheaper, but now on steroids. Right?
And so to me, this notion of AI arbitrage is when our organizations start to leverage artificial intelligence at the core of their workforce transformation and their workplace transformation, you will be able to unlock new levels of productivity, new levels of efficiency, and essentially new levels of outcome assurance. And that’s what we sort of call AI arbitrage. Because you’re seeing material movement.
in work. And you know, for the first time, you’re able to produce engineering outcomes, software engineering outcomes that translate to material business outcomes. Let’s say for example, that you had a software engineering process, or you had a certain set of KRAs that you would deliver on building software. You know, so far all of the tools and techniques that we had improved that incrementally, like you know, improve sprint velocity by 10 % or
15 % or 20 % or improve your code quality by 20 % 30%. Those are all incremental benefits. Those don’t necessarily move lines on revenue or lines on EBITDA. But with AI and with the agentified model, we’re talking about massive shifts, right? We’re talking about things being done at speeds that could change the business trajectory of an organization.
especially for enterprises, right? And so that to me is for the first time we’ve got real tech that can deliver significant software outcomes that can materially change business outcomes. And that’s what AI arbitrage is. Labor arbitrage didn’t do that. Labor arbitrage helped you deliver 30, 40%. But now you’re talking about percentages which go over 100 now, right?
I mean, I don’t even think it’s mathematically possible, but it’s up there.
Federico Ramallo (47:00) Right, but its order of magnitudes much higher.
KK (47:03) Massive order of magnitude, massive. Orders of magnitude of acceleration of how fast you can develop something or how soon you can create an idea. I was in a conversation with a client the other day and we were showing them our platform and he was trying to understand the platform. So I told him, listen, what are you trying to figure out? And he said, listen, we’ve been trying to build a data marketplace. We’ve thought about it. We haven’t really moved on it because we…
We haven’t been able to get the business together, know, blah, this, blue, that, and all that. And then I just said, okay, let’s just play with my platform. Let’s see where you get. And then we were able to get to a lean business canvas, must have, nice have requirements, epics to stories in one conversation without any business person in the room. And the guy’s like, my Lord, this is crazy. And so think about the weeks and weeks and weeks of time that was crashed by the gentleman using my platform, right?
Federico Ramallo (47:52) Yeah.
KK (47:59) And so you are talking about significant shifts that are creating material business outcomes. That is the part, that is why I do want us to lose the importance of focusing on the outcomes.
Federico Ramallo (48:11) Right, right. That was my thinking about thinking on core skills. You know, people with experience that can define what the outcomes should be so they can fit that to the LLMs.
KK (48:23) Yeah.
100 % storytelling. Federico is a very important skill set now.
Federico Ramallo (48:31) amazing.
KK (48:32) If I ask people, you know, what’s the one skill set that you learn that you need to learn? I tell them, go and take screenplay writing workshops. You need to know how to tell the story.
Federico Ramallo (48:41) Right. Because now it’s about having a conversation with the LLM to guide the outcome that you want.
KK (48:49) the outcome that you want and then be able to tell that story to somebody where that is now your role.
Federico Ramallo (48:55) Right, right, and get buying to move forward.
KK (48:59) Exactly.
Federico Ramallo (49:00) That’s amazing. I truly appreciate you being here today. Any final remarks before we wrap it up?
KK (49:05) No, I love that you’re doing this, Federico. I think this is incredible. You know, I think I get very, very excited about a lot of the tech that we develop. We’re developing amazing tech ourselves. The only thing that I would tell you to listeners is I think we are, this is not a short part. This is a long part. I mean, I’m using golf analogy here.
This is a long part where the ball is going to travel at five times the speed or 10 times the speed that you thought. But still remember, it’s a long part. So I think patience is virtue that we could all gain a lot from as we truly try and move the bar within enterprises.
Federico Ramallo (49:43) KK. Thank you for joining us today
KK (49:46) Fantastic. Thank you so much, Federico.