Agentic AI Needs Guardrails, Not Hype: Kyndryl’s Michael Bradshaw on What’s Really Holding Enterprises Back

Agentic AI is moving fast into the heart of enterprise operations, but most companies are not ready for it. On the surface, it might look like leaders are slowing down because they are tired of the noise. In reality, the hesitation comes from risk. Because the stakes are different when the AI agents take action for mission-critical systems - systems that don't yet have strong governance or security.

Kyndryl has been leaning into that gap. The company recently expanded its Agentic AI Framework with new orchestration and agent-building capabilities so customers can move beyond pilots and actually scale these systems. And now, with the launch of Agentic AI Digital Trust, Kyndryl is adding a layer that allows enterprises to manage agents' behavior, monitor them, certify them, and keep them compliant across hybrid and multi-cloud environments. It applies the kind of structure and control enterprises use for any core system, just adapted for autonomous workflows.

To get a better view of what’s really slowing enterprises down and what it takes to make agentic AI safe and usable at scale, plus what Kyndryl is doing in the space, ChannelE2E spoke with Michael Bradshaw, Kyndryl’s SVP and Global Practice Leader for Applications, Data and AI. And he put it plainly: companies aren’t lacking interest in agentic AI. They’re lacking the readiness to govern it responsibly.

ChannelE2E: A lot of people assume enterprises are hesitating on agentic AI because of hype fatigue. But underneath that, it’s really about risk - and that hesitation seems justified. What kinds of risks do you think are most misunderstood when it comes to systems that can reason and act on their own?

Michael Bradshaw: First, many believe that agentic AI will replace people; however, the future is not AI versus people – its AI agents collaborating with people, who will define and engineer progress.

Second, unintentional, unmanaged, unorganized AI is a real risk. Our Readiness Report found that if given the opportunity, 95% of leaders would change the way their organization implemented their cloud strategy. That can’t happen with AI. Most enterprises underestimate the complexity of securing systems that can reason and act autonomously and integrating them into their existing technology environments. To avoid previous mistakes, enterprises must make sure these agents operate within governed boundaries, they are being used to drive business impact, and that their business and technology teams are aligned.   

ChannelE2E: The core of the Kyndryl Agentic AI Framework is built around embedding governance, security, and transparency from the start. How does that shift the way enterprises think about building and deploying AI, not just as a tool, but as part of their operational fabric?

Michael Bradshaw: Previously, many automations were “set and forget,” so they didn’t grow and evolve with organizations. However, with the Kyndryl Agentic AI Framework being a secure-by-design agentic capability, it enables AI to be a tool that can consistently monitor performance and identify what actions need to be taken and when. Its full transparency, logging, alerting, and compliance reporting enable agentic AI to become part of your operational DNA – one that shifts as your organization does, while keeping compliance and oversight as the foundation.

ChannelE2E: One of the hardest parts of agentic systems is explainability - understanding why an agent made a decision. How is Kyndryl helping customers bring more visibility and traceability into these autonomous workflows so they’re not just powerful, but accountable?

Michael Bradshaw: Explainability and transparency are prerequisites for trust – and you can’t explain what you can’t see. This is why we built transparency into the Kyndryl Agentic AI Framework, so agents act in a form of autonomy and in accordance with business rules and policy. The result? AI you can trust.

The Framework uses people-centric design, composable architecture, seamless integration, policy-driven intelligence and transparent decisioning and more to provide customers visibility into agentic workflows by design. Every agent action is logged, traceable and auditable.

ChannelE2E: Plenty of organizations are still stuck in pilot mode because of compliance or risk barriers. What have you learned about helping them break through that wall and scale safely, without losing control or oversight?

Michael Bradshaw: According to the Readiness Report, 62% of organizations are stuck in the experimentation phase, with two key factors being foundational issues in their technology stack and having more pilots than they can realistically scale.

To combat this, organizations need help aligning their business and technology teams and navigating their complex technology environments and the ever-evolving regulatory and compliance landscape.  

ChannelE2E: The expanded Agentic AI Framework now includes orchestration, ingestion, and an “agent builder” to help customers move beyond pilots. How do those additions change the way risk, compliance, and security are built in from day one?

Michael Bradshaw: Our enhancements to the Framework, including our unique IP, like the core of the Framework, agent builder, and agentic ingestion, combined with Kyndryl’s mission-critical experience and Consult methodology, provide customers with a holistic approach to responsibly scale their agentic AI solutions and progress from one-off AI pilots to an AI-native enterprise.

The core of the Framework is where new operational processes are imagined and designed to deliver customers with efficiency and growth opportunities faster, while also providing full transparency, logging, alerting, and compliance reporting. It secures and governs every workflow. To leverage the full benefits of the core of the Framework, Kyndryl’s agentic ingestion uses reverse engineering agents to extract and analyze an enterprise’s technology estate, so agents understand the business context and compliance rules.

The Kyndryl agent builder uses our industry and domain reference architectures and agent catalog to design, engineer, and deploy - forward-engineer - modern systems and agents that meet security and regulatory standards before they deploy.

ChannelE2E: You recently partnered with Nova Intelligence to bring agentic AI into SAP transformations. What makes integrating domain-specific agents - like those built for SAP - challenging, and how do you maintain consistent governance across them?

Michael Bradshaw: Integrating domain-specific agents like those for SAP requires balancing flexibility with consistency. The Kyndryl’s Agentic AI Framework – and the holistic approach that accompanies it – enables us to be agile and flexible, so we can meet our customers’ needs today, tomorrow, and into the future, and apply the best-in-breed technology – whether that’s from our alliance ecosystem, like Nova Intelligence, bespoke, or co-created with customers.

Our Framework provides customers with managed, end-to-end AI solutions, featuring unified governance and automation to keep mission-critical operations running reliably, so that every agent can operate with the same security, compliance, and observability standards

ChannelE2E: You’ve talked about governments, insurers, and banks already using this framework. Can you share a concrete example of where embedding governance directly into AI workflows has led to measurable gains in efficiency or trust?

Michael Bradshaw: We are furthering the adoption of our Agentic AI Framework across several industries. For example, we are working with insurance industry customers on an agentic AI-enabled actuarial solution that creates and embeds AI agents to deliver an end-to-end intelligent, automated workflow process. The agents dynamically generate regulatory filings, save time and reduce errors, support proactive regulatory compliance checks, and deliver insights to drive real-time analysis and decision-making.

Additionally, we’re collaborating with a banking customer to create a dynamic client onboarding process to streamline and automate an intensive manual process that involves application submission, review, validation, and vetting with external parties. Using intelligent AI agents embedded within all stages of the process, the Framework is enabling the customer to streamline and accelerate onboarding time, while enhancing the overall customer experience.

ChannelE2E: You’ve also written about a future where AI agents act more like digital colleagues than tools. What needs to change within organizations - culturally and structurally - for humans and agentic systems to actually work side by side, rather than in silos?

Michael Bradshaw: Organizations require a culture transformation. The Readiness Report revealed that while 87% of leaders agree that AI is going to completely transform roles and responsibilities at their organization over the next 12 months, only 29% say their organization is currently ready to successfully leverage AI, and 48% see their culture as a key contributing factor.

The integration of AI into existing systems must be more than thoughtful – it must be intentional, governed, and people-centered. That begins with establishing a technology foundation that can support your AI goals and creating trust among your employees, where you reinforce that agentic AI isn’t about replacing people, it’s about augmenting them, so people are collaborating with agents to deliver bigger business impact.

To create this culture of trust, leaders must involve employees in the implementation process, create ethical guidelines for AI use, and ensure transparency around AI goals and implementation.