A new Kyndryl article declares that the age of AI experimentation is over. The true test of enterprise AI, according to Gonzalo Escajadillo, senior vice president for the Microsoft Alliance at Kyndryl, lies in operationalizing these systems on platforms like Microsoft Azure. Published on May 12, 2026, the piece argues that while many organizations have successfully piloted agentic AI, the real business value will only emerge when these capabilities are woven into the fabric of daily operations—running reliably, securely, and at scale on Azure.

Escajadillo’s message is as clear as it is urgent: enterprises must move beyond proofs of concept and toward full production. That shift demands a robust operational backbone, and for Kyndryl, Azure is the platform best positioned to provide it. The article signals a maturation of the enterprise AI conversation, away from the feverish rush to build the smartest agent and toward the disciplined engineering required to make that agent work in the real world.

The Rise of Agentic AI

Agentic AI refers to autonomous or semi-autonomous software agents that can plan, reason, and execute multi-step tasks with minimal human intervention. Unlike traditional chatbots or generative AI assistants that respond to prompts, agentic systems proactively orchestrate business processes—think an AI that not only drafts an email but also schedules the follow-up meeting, updates the CRM, and triggers the next step in a sales workflow.

Microsoft has been at the forefront of this movement. The Copilot stack, Azure AI Foundry, and tools like Semantic Kernel and AutoGen provide the building blocks for creating agentic applications. In 2025, announcements around Copilot Agent and the integration of autonomous agents into Microsoft 365 and Dynamics 365 made it clear that Redmond sees agents as the next evolution of digital work. By 2026, early adopters have already moved beyond simple retrieval-augmented generation to full-fledged autonomous workflows that connect data across systems, enforce business rules, and adapt to changing conditions.

Yet for all the promise, the journey from a successful pilot to a production-grade system remains fraught with challenges. Escajadillo’s article underscores a critical gap: operational excellence. It’s one thing to demo an agent in a controlled setting; it’s another to have it handle thousands of concurrent users, comply with regulatory requirements, and deliver consistent results over time. That, he argues, is where Azure’s managed services and Kyndryl’s expertise become indispensable.

Why Azure is the Operational Nexus

Azure offers a comprehensive suite of services purpose-built for running AI workloads at scale. Key components include Azure AI Foundry for model serving and fine-tuning, Azure Machine Learning for MLOps, and Azure Kubernetes Service (AKS) for containerized agent orchestration. But beyond compute, Azure provides the governance, monitoring, and security frameworks that enterprises need. Azure Policy, Microsoft Purview, and Azure Monitor combine to create a unified compliance and observability layer across all AI agents.

Moreover, Azure’s hybrid and multi-cloud capabilities allow organizations to deploy agents close to their data—whether on-premises, at the edge, or in a sovereign cloud. This is especially critical for industries with strict data residency requirements, such as finance and healthcare. Microsoft’s investment in confidential computing and its responsible AI tooling further addresses the trust gap that often slows enterprise adoption.

For Kyndryl, Azure is the natural choice because of its deep integration with the Microsoft ecosystem that most large enterprises already rely on—Windows, Office 365, Dynamics, Power Platform, and Entra ID. By building agentic AI on Azure, companies can leverage existing identity and access management, data governance, and low-code automation assets, dramatically reducing the time-to-value. Escajadillo’s article emphasizes that operational AI is not just about model accuracy; it’s about weaving agents into the existing IT landscape without creating new silos or security vulnerabilities.

Kyndryl’s Managed Services Perspective

Kyndryl, spun off from IBM’s managed infrastructure business in 2021, has positioned itself as a leader in multi-cloud management and modernization. Its Microsoft Alliance, led by Escajadillo, is a strategic partnership that has grown to encompass thousands of joint engagements. The new article reflects Kyndryl’s real-world experience: countless enterprises have moved AI pilots into production, but many stumble on the operational complexities.

Escajadillo points to three common failure patterns. First, organizations underestimate the data integration effort—agents need clean, well-governed data from a variety of source systems, which is rarely ready on day one. Second, they overlook the importance of lifecycle management; models and prompts drift, and agents must be continuously monitored, retrained, and updated. Third, they neglect the human-in-the-loop frameworks required for trust and accountability, especially in regulated environments.

Kyndryl’s approach combines Azure’s native services with its own operational frameworks, such as AIOps for predictive incident management and financial operations (FinOps) for cost control. The company offers a “landing zone” accelerator for Azure AI that pre-configures governance, networking, and security for agentic workloads. By pairing this with managed AI operations, Kyndryl aims to collapse the timeline from pilot to production from months to weeks. The article suggests that enterprises should think of agentic AI not as a standalone project but as an extension of their existing IT service management—with the same rigor applied to availability, capacity, and change management.

Moving from Pilots to Operations: The New Playbook

The shift from pilot to operations is less a technology problem and more a discipline problem. Escajadillo outlines a playbook that starts with defining clear business metrics for AI agents—moving beyond “coolness” to measurable outcomes like cost reduction, revenue growth, or customer satisfaction improvement. Without such metrics, production agents become expensive curiosities.

Next, he advises designating an AI operations owner—someone responsible for the end-to-end reliability, security, and performance of agentic systems. This role often sits at the intersection of IT operations, data engineering, and application development. The owner must champion the adoption of Azure’s operational toolchain: Azure DevOps for versioning agent configurations, Azure Monitor for telemetry, and Azure Security Center for threat detection.

A critical step is the implementation of a robust testing and evaluation pipeline. Pilot agents are typically vetted with a handful of scenarios; production agents require continuous red-teaming, automated response quality checks, and performance benchmarking under load. Azure AI Evaluation Metrics and the Azure AI Safety System provide native tools for this, but Kyndryl’s experience shows that most enterprises need help configuring them correctly. Escajadillo’s article insists that “evaluation as code” must become a standard part of the CI/CD pipeline for AI, just as unit testing is for traditional software.

Finally, cost management cannot be an afterthought. Agentic AI can consume significant compute resources, especially when using large language models for reasoning. Azure Cost Management and Kyndryl’s FinOps dashboards help enterprises set budgets, allocate costs to business units, and optimize model selection. Escajadillo warns that without proactive financial governance, a successful pilot can quickly balloon into an unsustainable operational expense.

Overcoming the Trust Barrier

Even with technical excellence, agentic AI fails if users don’t trust it. Escajadillo highlights the human dimension: employees need to understand what agents are doing, why decisions are made, and when to override them. The article advocates for a transparency-first architecture, using Azure AI Content Safety and Azure AI Prompt Shields to surface reasoning traces and provide clear audit logs.

Kyndryl often implements a “confidence threshold” system, where agents escalate to humans when their certainty falls below a predefined level. This human-in-the-loop pattern is essential in fields like legal, accounting, and medical services. Azure’s built-in support for asynchronous human feedback—via tools like Microsoft Teams and Power Automate—makes such workflows almost natively available. But Escajadillo notes that many enterprises fail to design these loops from the start, leading to costly rework later.

The article also touches on the evolving regulatory landscape. With the EU AI Act fully enforced and similar regulations emerging globally, operational AI must demonstrate compliance by design. Azure’s responsible AI dashboards and compliance certifications provide a foundation, but Kyndryl’s role is to help clients map regulatory requirements to specific technical controls—a task that requires deep knowledge of both Azure and industry-specific mandates.

The 2026 Outlook: Operational AI Becomes the Standard

Escajadillo’s central thesis is that 2026 marks a turning point. Early mover advantage in agentic AI will go not to those with the most clever demos but to those who master operations. He predicts that within two years, operational AI maturity will be as critical to business success as IT service management maturity is today. Companies that treat agents like another application—subject to the same change management, performance SLOs, and incident response processes—will outcompete those that continue to treat AI as an experimental sandbox.

For Microsoft, this narrative aligns perfectly with its “AI-platform” strategy. By making Azure the operational home for agents, Microsoft ensures that enterprises deepen their Azure consumption and stickiness. Kyndryl, as a managed services giant, benefits from the growing complexity: Azure AI is powerful but sprawling, and many organizations lack the in-house skills to operationalize it effectively. The partnership between the two companies is thus symbiotic, with Kyndryl providing the services layer on top of Azure’s product layer.

The article closes with a call to action: enterprises should audit their current AI pilots and assess whether they have a clear path to production on Azure. For those that don’t, the time to engage a partner like Kyndryl is now, before the operational debt becomes overwhelming. The message is practical and grounded—refreshingly free of the hype that often surrounds AI. In Escajadillo’s view, the next frontier of AI is not larger models or more autonomous agents; it’s boring but essential operational excellence.