Editor’s note: The following is a guest post from Jamie Rutledge, president of Kyndryl US.
While early AI pilots and proof of concepts spark excitement, turning experimentation into production that results in ROI is proving far more difficult for infrastructure and technology leaders.
So far, much of the conversation centers on new model development and performance, but leaders are failing to focus on something equally as important: production readiness. AI initiatives tend to stall when they encounter technical hurdles such as hybrid cloud constraints, fragmented data environments, latency requirements or review processes that were not designed with agents in mind.
A shift from experimentation to enterprise deployments exposes new architectural realities that cannot be solved with new models.
In an AI-driven world, business success hinges on how companies orchestrate, secure, scale and use AI agents to drive business impact. AI adoption should start with mission-critical technology foundations – AI-ready infrastructure, governed data and embedded security – that is integrated within modern platforms and applications.
This effort also means using AI itself, particularly agentic AI, to modernize legacy systems and workflows, not just layer intelligence on top of them.
Rethinking IT strategy
Companies looking to make the most of their investments must integrate AI into live systems that operate continuously, securely and in compliance with regulatory standards.
For enterprise environments that rely on a mix of on-premises infrastructure, private cloud or public cloud providers, data is sprawled across multiple systems. Without a cohesive strategy, AI becomes another hurdle rather than a beneficial capability.
Leading organizations are using agentic AI to accelerate this modernization, automating integration, improving data flow and helping rearchitect legacy environments for scale. But to scale AI successfully, they must take on a process leaders cannot overlook: operating model reinvention.
AI must be embedded into enterprise architecture with the same discipline applied to any other mission-critical workload. Agentic AI must dynamically manage, optimize and modernize systems as they operate.
Systems must be “always on,” withstand cyber threats, integrate with existing IT service management processes, meet evolving compliance requirements and scale efficiently to maintain operational and cost competitiveness. Governance must then also be operationalized; it fails when it lives inside policy documents.
Governance must be approached systematically, with clear approval workflows before production deployment, role-based access controls for agents and runtime logging to monitor AI outputs. Approaches such as policy-as-code can help build governance directly into these processes and ensure guardrails are enforced consistently.
Once AI begins influencing decisions, workflows and customer interactions, organizations must define ownership of its output, validation processes and intervention protocols in the event something goes wrong.
Time for change management
Across industries, a consistent pattern is emerging: Initial automation gains quickly fade when organizations try to layer AI onto existing processes without rethinking how work is structured. Efficiency improves, but structural bottlenecks remain. Ownership over decisions remains unclear, leading to new risks and structural uncertainty.
In conversations with enterprise leaders across different industries, it’s evident that the companies pulling ahead in the race to scale AI are redesigning workflows rather than just optimizing them. They are moving beyond task automation to reconsider how functions collaborate, how accountability is assigned and how value is measured when humans and AI operate side by side.
This is where organizational change management becomes a core technical discipline.
For AI to operate reliably in production environments, teams must redefine roles. Infrastructure teams must understand new AI workload requirements. Security teams must adapt to new threats like prompt injection and model manipulation.
Business leaders must also work with technical and non-technical teams to clarify where human judgment overrides automated outputs — and how AI is governed for continuous enforcement across the enterprise.
AI readiness ultimately comes down to workforce readiness – in fact only 29% of leaders feel their workforce is ready to leverage AI successfully, according to Kyndryl data. When organizations design new AI-enabled ways of working from the ground up, they equip employees with the skills required to operate effectively alongside intelligent systems.
As those capabilities grow, employees increasingly expect visibility into how AI systems influence their work and decisions. For leaders, this means AI adoption cannot be treated purely as a top-down technology rollout. Organizations that co-create new ways of working with AI will see stronger adoption, better accountability and longer lasting impact across the business.
For technical leaders, this means expanding the scope of AI transformation beyond infrastructure and controls. It requires orchestrating change across people, process and technology simultaneously. In practice, that means working “East-West” across the organization – connecting business functions to redesign how work gets done.
Industrializing AI cannot happen in vertical silos because AI itself cuts across functions. Agentic AI-enabled modernization further accelerates this shift, requiring organizations to rethink how systems and teams evolve together. Success will depend on whether enterprises can redesign themselves to operate alongside it.
