5 conversations to have before deploying agentic AI for the mainframe

Can agentic AI for the mainframe create value for our business?

Agentic AI for the mainframe captures decades of institutional knowledge and makes it instantly accessible to IT teams. This capability helps address one of the biggest mainframe challenges: the growing mainframe skills gap.

What’s more, agentic AI for the mainframe standardizes processes and learns from every action, creating more resilient and adaptable systems. Agents also make it easier to update application code and integrate with cloud and microservices.

How to address

Determine if — and how — using agentic AI on the mainframe can help advance your business strategy, support innovation and future-proof your organization instead of focusing solely on IT optimization.

Think in terms like faster time-to-market, improved customer experience, reduced operational risk or new revenue streams. Less downtime, improved compliance and more productivity are just a few measurable outcomes that demonstrate the technology’s true value.

How do we ensure security and governance in an autonomous environment?

Guardrails and governance help ensure employees use AI responsibly, reducing exposure to regulatory, legal and security liabilities that might otherwise lead to fines, lawsuits, cyberattacks, revenue loss and reputational damage.

Surprisingly, Kyndryl’s 2025 Cloud Innovation Study reveals that only 39% of respondents have a documented position for responsible AI, which is essential for balancing innovation and risk. Another 25% have yet to start discussing the topic.

Meanwhile, results from Kyndryl’s 2025 AI Readiness Report show that 86% of IT leaders are confident in their AI implementation and believe it’s best-in-class, but just 29% of respondents say that their implementation is ready to manage future risks.

Agentic AI’s ability to autonomously plan and perform multi-step tasks will amplify risks. Without policy enforcement, monitoring and other safeguards, agents can operate free from human oversight, potentially leading to data breaches, regulatory non-compliance, ethical concerns and other unintended consequences.  

How to address

Organizations must build security and governance into the entire implementation process to minimize risks when deploying agentic AI on the mainframe. This approach requires embedding compliance policies directly into workflows and ensuring every agent action is logged and auditable.

Continuous monitoring — powered by AI and managed on open integration platforms — should underpin operations, providing real-time oversight and early anomaly detection. Most importantly, always keep a “human in the loop” to address critical actions and to regularly review agent decisions for bias, fairness and ethical standards.

Are we operationally and culturally ready for agentic AI?

Integrating AI and agentic AI in mainframe environments before your people and processes are prepared can create or accentuate numerous challenges. 

Case in point: Nearly all (95%) respondents to the Cloud Innovation Study are confident or somewhat confident that their company’s IT infrastructure can support enterprise-wide use of AI. However, only 43% of organizations are using AI for mission-critical business processes.

Kyndryl’s People Readiness Report 2025 shows that employees play a particularly important role in transformation success. Although 95% of respondents have invested in AI, 71% of IT leaders say their workforces aren’t ready to use the technology successfully.

In addition, 45% of CEOs think most employees are resistant or even openly hostile to AI, and 37% admit integrating AI and automation with existing systems is proving to be a significant challenge.

How to address

Deploying agentic AI for the mainframe isn’t just a technology shift — it’s a change in how work gets done. So, you’ll want to prioritize organizational change management as much as technology transformation when developing plans. Best practices include:

• Maintaining open communication with all employees and providing regular updates
• Launching pilot projects to build confidence and demonstrate value
• Offering ongoing skills training, knowledge-sharing and support
• Addressing job concerns and illustrating how the technology helps employees
• Appointing change champions across departments and at all seniority levels

Which use cases should we pursue first—and how do we avoid the wrong ones?

Identifying high-value use cases for deploying agentic AI on the mainframe determines whether the technology becomes a true growth driver. Look for opportunities that can reduce compliance gaps, enable faster onboarding, and improve customer experience. The best applications deliver visible business impact that cuts across industries, from healthcare and manufacturing to banking and government services.

How to address

Assemble a cross-functional team to identify and prioritize use cases. Possible high-value applications include:

• Proactive fraud detection and risk management for secure banking and payment transactions
• Predictive maintenance and self-healing for uninterrupted service in healthcare and government
• Accelerated application modernization for supply chain and inventory systems in manufacturing and retail
• Customer experience personalization to drive loyalty and revenue in retail, financial services and telecom

How do we implement agentic AI for the mainframe without disrupting our business?

Mainframes are designed for stability and reliability, so introducing agentic AI — which requires blending new technologies and workflows with older tools and platforms — can disrupt established processes, create compliance concerns or overwhelm teams with change. Without careful planning and strategic implementation, organizations may face additional challenges like unintended downtime, data integrity issues or business interruptions.

How to address

A pragmatic and practical approach is the ideal way to deploy agentic AI on the mainframe.

First, ensure your data structure and information architecture are clean and adhere to the latest data privacy and cybersecurity standards. Next, develop a roadmap to roll out the technology in phases. Start with non-critical processes like batch processing, application testing and data analysis­­.

Use lessons learned from early phases to refine your deployment approach and ensure agentic AI works properly with existing mainframe and cloud systems. Create continuity plans to minimize the impact of any unexpected issues and maintain human oversight throughout the transition.