Social impact
The cyber hack that hooked a classroom
Article
May 13, 2026
By Xerxes Cooper, Global Leader, Kyndryl Delivery, Global Integrity Champion
In IT delivery, traditional automation has been, and continues to be, an essential component of how technology runs smoothly. At Kyndryl, we execute 200 million automations each month, saving our customers billions of dollars by making automatic, preprogrammed decisions. It’s our first line of response, and it’s based on deep operational observability in Kyndryl Bridge, our AI-powered open integration business platform. And as important as automation is, it also has real limits.
Traditional automation follows strict, preprogrammed rules. But there are clear limits of what preprogrammed rules can do in the reality of today’s messy, multivendor, sprawling — and business-critical — IT environments.
Enter agentic AI for IT delivery. It’s the next game-changer, and early applications are already showing dramatic results. It’s also enhancing how engineers work, with agents taking on parts of the workflow so people can focus on judgement, exceptions and customer outcomes.
The end result? These informed agents can take on many more anomalies and help improve delivery. This means more uptime, greater automation coverage and higher productivity. And with shared, trusted operational data, customers can feel confident that agentic AI can do this without compromising stability, security or trust.
Agentic AI becomes a scalable delivery capability, not a collection of disconnected experiments.
A new model is emerging in which software doesn’t just execute preprogrammed steps. Now, AI agents can take in the full context of complex IT environments, interpret signals, suggest actions, look up documentation, and coordinate work across different tools and services. Agents can also drastically speed up modernization projects, from mapping out dependencies to converting code and generating documentation. Critically, people still set boundaries and hold responsibility.
This is the promise of agentic AI, and it’s why IT operations and managed services are becoming one of the best places to prove its value, moving toward autonomous IT operations.
Importantly, just about no organization is starting from scratch. Many AI predictions overlook three hard truths: enterprise IT is already live, enterprise IT is already complex, and enterprise IT is already business‑critical. Today’s global companies use dozens of tools running on unique combinations of hardware and software, and they can’t rebuild their systems overnight. Kyndryl’s model was built to operate AI inside that complexity without breaking trust, compliance, or continuity.
One example of how this can work in action: for an international airport, Kyndryl deployed our agentic AI capabilities, with more than 50 use cases across service management and core IT operations. The result: stronger service quality and reliability at the airport, including everyday systems, such as flight information displays and fewer Wi‑Fi incidents.
That kind of impact didn’t come from theory or a lab experiment – it came from deploying agentic AI in real delivery environments first. As we’ve moved from experimentation to execution, a few clear lessons have emerged:
Agentic AI has a real upfront investment.
Success takes new skills, integrating platforms and tools, and strong governance. This isn’t optional, it’s foundational. The mistake is assuming that effort alone guarantees value.
Not every use case is worth it, but the right ones are game-changing.
Lower-value use cases tend to be isolated, hard to scale, and difficult to reuse. They slow progress and make it harder to standardize and replicate across the enterprise. In contrast, the best use cases create reusable patterns, accelerators, and reference architectures. That’s what enables scale, speeds up delivery, and turns initial investment into long‑term momentum.
Scaling is the real challenge – not pilots.
The hardest part isn’t building one agent. It’s enabling teams across the business to build many agents in a consistent way, without losing speed, quality, governance or reuse.
Training alone isn’t enough.
Teams need hands‑on experience. Developers must be able to design, build, test, and deploy standard agents in real environments. Learning has to be practical, repeatable, and aligned to clear standards.
Governance must be designed first, not added later.
Agentic systems can behave in non‑deterministic ways. Traditional monitoring isn’t enough, so you need built-in operational controls: an “agent operations” layer, clear audit trails and logging, security operations alerting, and protection against prompt injection – designed into the platform from the start. Without that foundation, scaling from pilots to enterprise rollout stalls.
Deploying agentic AI for the first time is as much a psychological journey as a technical one.
Customers are both eager to deploy this new technology and understandably cautious about how it’s actually implemented in their core systems. Better visibility into what agents are doing, strong security, and clear registration all help. And new tools are improving this every day.
When we prioritize high-value use cases with clear outcomes, and enable teams to work independently within clear standards, we don’t slow down, we speed up. That’s how agentic AI becomes a scalable delivery capability, not a collection of disconnected experiments.
Agents can’t act responsibly without a shared, trusted view of the environment they’re operating in. For our customers, Kyndryl Bridge can provide that operational truth — standardizing signals across automations, health, incidents, and environments so both people and AI agents share common understanding and are working toward the same results.
Because Kyndryl Bridge is already embedded in how we run IT delivery, agentic AI can move beyond pilots into live operations that improve uptime, consistency, and speed. Kyndryl Bridge is what turns agentic AI from experimentation into enterprise delivery. The platform can also draw from external, up‑to‑date sources, giving agents both historical perspective and real‑time awareness. And there’s the component of scale: agents that learn a pattern in one environment can help resolve similar issues faster in another.
But agentic AI isn’t a plug-in for existing ways of working. It requires a mindset shift – embedding agentic use cases into day-to-day delivery – and rethinking processes so workflows are designed for real-world context, exceptions, and governed autonomy from the start. The Agentic AI Framework provides a structured, secure, and repeatable way for organizations to embed AI into their operational workflows. It helps enterprises evolve from manual, reactive processes to more intelligent, automated, and context‑aware operations powered by AI agents working alongside upskilled teams of people. For our customers, this means more predictable and resilient IT delivery – powered by agentic AI and governed by design.
