Kyndryl CEO On Launch Of Advanced Agentic AI Initiative

‘We launched our Kyndryl Agentic AI Framework back in July. We’re now bringing the set of capabilities to bring it to life. Our customers are enterprises with mission-critical needs and very complex hybrid infrastructures. We now have the ability to get these customers to leverage agentic AI in a way that is secure. We help them orchestrate and scale it so they not only can do the things they’re reading about, but do it faster, do it safely, do it in the right governance models, etc. It’s very exciting,’ says Kyndryl Chairman and CEO Martin Schroeter.

Kyndryl’s July launch of its Kyndryl Agentic AI Framework laid the groundwork for the global enterprise technology services provider to start working with customers to bring agentic AI into their operations.

Less than three months later, Kyndryl is putting that initiative into action with the launch of an advanced agentic initiative aimed at accelerating the adoption of AI by customers across industries and at scale, said Kyndryl Chairman and CEO Martin Schroeter.

Schroeter, in an exclusive meeting with CRN, said the Kyndryl Agentic AI Framework brings his company’s agentic AI capabilities to life.

[Related: Kyndryl CEO On Driving ‘3As’ Strategy: More Alliances, More Advanced Delivery, Focused Accounts]

“Our customers are enterprises with mission-critical needs and very complex hybrid infrastructures,” he said. “We now have the ability to get these customers to leverage agentic AI in a way that is secure. We help them orchestrate and scale it so they not only can do the things they’re reading about, but do it faster, do it safely, do it in the right governance models, etc. It’s very exciting.”

Schroeter said Kyndryl is “customer zero” when it comes to agentic AI, including using its experience with Kyndryl Bridge, the company’s platform for integrating and orchestrating complex IT infrastructures to better meet customers’ mission-critical requirements.

“At same time, we use our agentic ingestion unit to make sure we understand how we’re operating today,” he said. “We reverse-engineer and extract and analyze our own infrastructure so we can again build the policies and make sure they’re all being enforced, build the security rules, all the things that you would think about from a guardian perspective. So yes, we use this today to build our own agents. We then deploy our agents and manager agents and coordinate our agents internally, just like we help our customers.”

There’s a lot going on with agentic AI at Kyndryl. For the complete overview of the company’s activities in this area, read the full conversation with Schroeter, which has been lightly edited for clarity.

What’s your latest definition of Kyndryl? How do you describe the company?

My latest definition of Kyndryl is the same as the company when we spun out of IBM, and that is, we run the world’s mission critical systems, we modernize them, and we transform them. We run and transform them better than anybody else, given the deep engineering talent we have.

What’s new with Kyndryl in agentic AI?

We launched our Kyndryl Agentic AI Framework back in July. We’re now bringing the set of capabilities to bring it to life. Our customers are enterprises with mission-critical needs and very complex hybrid infrastructures. We now have the ability to get these customers to leverage agentic AI in a way that is secure. We help them orchestrate and scale it so they not only can do the things they’re reading about, but do it faster, do it safely, do it in the right governance models, etc. It’s very exciting.

Given that this is such a new technology, please define what Kyndryl means by the term ‘agentic AI.’

Think about the different kinds of AI that businesses use internally. Agentic AI is part of the continuum of AI. We use a lot of machine learning because it’s how we identify patterns to automate things. It’s all rules-based. We can predict things, we can observe things, we can give customers insights based on what we’ve seen. That’s at one end of AI. Machine learning has been here in Kyndryl for a long, long time. We’ve been using it very successfully. In fact, I think, as we sit here today, we automate something like 170 million things per month for our customers.

Then there’s generative AI, which we’re obviously helping our customers with. It creates new things. It creates text, images, code. It’s quite good at writing programs. And that’s all based on large language models. We use also that internally to help us understand a bunch of different functions.

When we think about agentic AI, we think about relentless goal-seeking bits of code that have intelligence, that can act autonomously. They can learn from their environment and can collaborate both with other agents and with humans. So while they don’t generate content, they do make decisions and take actions, and they can adapt in real time, just like the way we run all of our systems. They’re designed to be our eyes and ears, if you will, to take action, confer with the experts—our engineers—to make sure they’re doing the right thing, and to watch what other agents are doing as well. And they’re designed to operate across these complex IT environments with—importantly for us, we know this better than anybody—the right oversight, the right accountability, the right governance, so that they’re not stepping out of line. They are relentless goal-seeking things. You have to orchestrate and manage and control them, just like you would any person or anything that’s acting within your system. This for us is a way to deliver to our customers a lot of what we do ourselves in a way that recognizes their requirements, recognizes their reference architecture for whatever their industry is, and makes sure they know that it’s going to be controlled, be governed in the right way, be complete, et cetera. That’s a big part of our news. The other part is about how we’re moving our people, the employees of Kyndryl, to get them ready to take this on through training and skilling so they can be confident that their skills will be enhanced. Their oversight and their judgment are critical in how we do this, but they also need to be ready and comfortable using AI so that they can bring that to our customers in a very fulsome way.

Is what’s being unveiled now something that Kyndryl is already working on with customers, or is this a new capability you are bringing customers?

We released our Kyndryl framework back in July, and that described how companies can do this. We have been working with some of our customers on this, and now we’re making those capabilities available for all of our customers. We have examples already. We’re working with a government in the Middle East to help them be an AI-first government so they can reach their citizenry using agentic AI. We have examples in some other industries. So this is what I’ll call the broad release of the tools and the things our Kyndryl agentic AI framework described. These are now the tools so that they can start to do this themselves. For instance, through our ingestion engine, we reverse engineer, extract, and analyze an enterprise’s current technology estate, applications, data, policies, process, the logic that applications have, the data schemes, their topology, the way people talk to machines and machines talk to machines, the workflows, and the interdependencies. We get all of that, and then add insights from Kyndryl Bridge. And that allows our agent builder with our own industry and domain reference architectures to create and catalog and design and engineer and deploy forward engineers for their systems so they can start to leverage agents in the way they operate. It’s a faster way to get the benefit of AI in their systems and help them start to become agentic AI enterprises. But it’s only the beginning for them.

Does Kyndryl use these capabilities for its own processes? Give us an example of what Kyndryl is doing internally.

We are Customer Zero here. As I mentioned, we have a lot of learning from Kyndryl Bridge. At same time, we use our agentic ingestion unit to make sure we understand how we’re operating today. We reverse-engineer and extract and analyze our own infrastructure so we can again build the policies and make sure they’re all being enforced, build the security rules, all the things that you would think about from a guardian perspective. So yes, we use this today to build our own agents. We then deploy our agents and manager agents and coordinate our agents internally, just like we help our customers.

For the agents Kyndryl builds for customers, are they expected to be custom agents? In other words, are they built specifically for a customer? Or are there agents that can be applied to multiple customers?

It’s a bit of both. What I expect with our customers using our framework, our ingestion tools, and our agent builder, these are going to be different. So engaging with our consult experts and with Kyndryl Vital, I think they’re going to look a lot different by customer. Maybe they’ll be a little bit similar by industry. A ‘know your customer’ or KYC process at one bank could look like another bank’s KYC, but how it was implemented could be completely different. So I think there’s going to be high variability for the agent builder after they’ve gone through the ingestion tool to understand what happens.

You mentioned ‘Kyndryl Vital.’ What is that?

Kyndryl Vital is a process focused on the end user experience that our customers are trying to achieve. We build back from the user experience, and that allows our agents to be built in a way that they make sure customers are achieving the kinds of user experiences that they’re shooting for. So it’s a set of methodologies by which we help customers build the right things.

Is Kyndryl Vital something new that comes with agentic AI?

It’s been part of how we’ve been running for a number of years. We launched Kyndryl Vital over three years ago, and it’s already very well received by customers. So it’s been around. Think of it as sort of our co-creation and design approach.

As customers deploy agents and agentic AI, do you see them using agentic AI as a way to reduce head count?

I would be a little surprised, because what we find from the way we run, and I think most customers are going to find in this space, is the deep know-how, the understanding, the judgment that it takes to run complex infrastructure. AI is going to be about supplementing what people are doing. It’s going to be about working people and machine. So in our space and what we’re doing for customers, I don’t think it’s going to be about replacing people. You just can’t. If you’ve ever used, and I’ve done this, by the way, but if you ask ChatGPT or Grok, ‘I have this kind of infrastructure on this size database running across this network with this applicator,’ you don’t get a useful answer back. … It’s not useful for what we do. Now, having said all that, if we were in a different space, if we were in more of a coding space, managing applications, creating applications, GenAI is really good for that. You can build agents to build code for you. It’s just not the space we’re in.

I think for our space, managing complex infrastructure, one of our differentiators has always been our people, the engineering we have, the judgment that they bring, the imagination, the ability to collaborate, all that, that is not what these agents are going to do. But these agents are. They are going to get to root cause analysis faster. They’re going to be more data driven. They’re going to be relentless in finding the data needed to get to root cause analysis of what’s going on in the infrastructure, etc. So for us, it is very much about humans working with these agents.

You’ve seen reports from organizations like MIT saying that 95 percent of AI projects fail. What is your view in terms of the potential for failing AI and agentic AI projects?

Even in baseball, you only have to hit one out of three times to go to the Hall of Fame. But failing 95 percent of the time seems like it’s a bit rough. For us, I’d say two things. One, our customers are not quite at 95 percent. Our customers tell us that about 90 percent of them are experimenting with AI. We know that because we’re helping with a lot of it. But out of the 90 percent that are doing those experiments, I’d say maybe just under half are actually getting positive ROIs at this point from the things they’re trying. So lots of trying, and about half are already achieving positive ROIs.

Second thing, I’d say, that’s really what our Agentic AI news is about. Remember: mission critical. Everything has to be secure, has to retain resiliency. For a lot of the stuff we do, it has to be acceptable to a regulator, etc. In that world, I think what we’ve unveiled is going to dramatically improve the ROIs for what we do with our customers. The co-creation element of this the ability to reverse engineer and use our ingestion tools to give customers a full view of how they’re running today, and then to use our agent builder to build agents that can accomplish things in a way that preserves the security and the resiliency with their business rules and their security rules, their regulatory environments, the guardrails they need to operate. In this co-creation world, it’s going to advance their ability to get positive ROI sooner. But again, we sit in a different world compared to what I think MIT was asking about, so it’s not surprising.

What comes next from Kyndryl?

I’m excited about providing more and more customer examples of agentic AI. In fact, in about three weeks, we’ll be describing something we’re working on now, using this as the basis for what we’re doing in the travel and transport industry, an airline and how we’re helping them. This is getting accepted by our customer base, helping them along their journey. This is very exciting. So what’s next for us is using this stuff, getting our customers to use this stuff, and showing the world that when the two things that matter in their world, their infrastructure and data, work and you have the right service partner to help you get there, you are going to change how you operate. You are going to change how you do business. You’re going to improve how you reach your customers in the new agentic AI world. It’s coming.