‘We’ve created an approach to implementing agentic AI in an environment which is secure and enterprise grade. It can be rolled out just like we roll out our infrastructure for customers, banks, and governments and airlines and automotive companies, where we can’t afford to be non-secure, we can’t afford to go down, can’t afford to be performance affected,’ says Ismail Amla, senior vice president of Kyndryl Consult.
Global enterprise technology services provider Kyndryl has introduced its Kyndryl Agentic AI Framework, which aims to ease the deployment of agentic AI to work along human teams.
Kyndryl’s, and the IT industry’s, investment in agentic AI comes on the heels of the widespread adoption of GenAI, said Ismail Amla, senior vice president of Kyndryl Consult.
“GenAI was great for getting content,” Amla told CRN. “Suddenly we’ve got not just content, but incredible content, reasoning, learning, and acting all in an agent and all done in a way where it’s gone from a little bit of technology to an influx of technology capability and with it lots of opportunities.”
[Related: Kyndryl CEO: Reselling ‘Other People’s Equipment’ Was ‘Empty Calorie Revenue’]
Investment in AI by multiple vendors in technologies including large language models, inference engines, and frameworks is generating those opportunities, Alma said.
The big question in the industry is what that all means for things like customer service or human resources, he said. That in turn leads to the question of how to take advantage of agentic AI, not in a “show me a demo” world, but in enterprise-grade workloads, he said.
“Kyndryl is helping make sense for our customers of all of the technologies now available, all the opportunities people talk about, and all of hype had around GenAI, where a lot of the feedback was not a lot of it got implemented,” he said. “There was a lot of hype and those demos, but not a lot got implemented.”
Kyndryl’s response was the launch of Kyndryl Agentic AI Framework, which aims to ease the deployment of agentic AI to human teams, Alma said.
“We’ve created an approach to implementing agentic AI in an environment which is secure and enterprise grade,” he said. “It can be rolled out just like we roll out our infrastructure for customers, banks, and governments and airlines and automotive companies, where we can’t afford to be non-secure, we can’t afford to go down, can’t afford to be performance affected.”
The Kyndryl Agentic AI Framework takes advantage of thousands of infrastructure deployments and over 12 million AI-driven insights via Kyndryl Bridge, the company’s platform for integrating and orchestrating complex IT infrastructures to better meet mission-critical requirements.
Combined, the Kyndryl Agentic AI Framework and Kyndryl Bridge use advanced algorithms, self-learning, optimization, and secure-by-design AI agents to turn complex data into clear insights, Alma said.
“This agentic AI framework allows you to connect all of these incredible technologies in a business environment, but doing it in a way where you are addressing issues around bias, security, enterprise grade, technology lock-in, and so on,” he said.
Kyndryl is already using the Kyndryl Agentic AI Framework with a couple of large clients, Alma said.
In one example, it is working with a government to create a native-AI government. “We define the policies from a human perspective, and the agents with little or no need for people goes and performs the task and comes back to humans when there’s something that might create risk that requires human judgment,” he said.
In another example, Kyndryl is also working with a financial services company.
“Financial services are usually leading-edge in terms of deployment of some of these technologies, but are also very, very concerned about security and risk, speed, and so on,” he said. “We’re working with this financial services company to take a particular process, it could be something like fraud management or making faster payments, and use things like customer experience and actuarial support to introduce agentic AI to sit alongside a human to get more value out of what they’re doing for their customers.”
Alma declined to name either of those two customers.
Transparency within the Kyndryl Agentic AI Framework is essential to ensure bias does not enter the agentic AI processes, Alma said.
“By making sure that you are doing the learning in a way which is fair and transparent, biases are identified for addressing rather than reinforced for learning,” he said. “There’s a lot of extra steps built in, and it’s a little bit different depending on the technology and the models you’re using and the maturity of the models. It also varies according to how you configure your large language models and your small language models as to what you allow in terms of transparency, in terms of learning, in terms of inference, once you’ve got past the learning stage.”
How AI agents work in general also depends on how they are treated, Alma said.
“Think of the workforce being a combination of agents and people,” he said. “In a way, agents need to be treated like humans in the sense that you need to select them, to onboard them, to continually train them, to coach them. And then, maybe by the ‘sell-by’ date, because something else has come up, you might want to exit them. So the IT department becomes a little bit of the HR department for agents. And so it’s a huge cultural shift right in terms of how you go about implementing some of this stuff.”
Going forward, Kyndryl plans to develop agentic AI capabilities focused on specific industries and on how businesses can use their customers’ specific data, Alma said.
“Our focus is now on not just using generic technology for generic issues, but to make it very industry-aligned, and also keeping the ecosystem open so that for our customers can use the framework as new technologies get plugged in,” he said.
