Rocket Software Brings Governed Agentic AI to the Enterprise Mainframe

WALTHAM, Mass. – September 23, 2026 – Mainframes quietly process roughly 70 percent of the world's production transactional value, acting as the invisible backbone for core banking ledgers, airline reservations, and global supply chains. Yet, the enterprise data center is facing a demographic cliff. As the veteran engineers who built and maintained these legacy systems reach retirement age, a severe operational knowledge vacuum is threatening mission-critical infrastructure.

In a strategic move to bridge this widening divide, Rocket Software, a global technology leader in modernization software, today announced a major expansion of Rocket EVA, its agentic artificial intelligence platform for mission-critical systems. Crucially, the update introduces Rocket PlanGuard, a patent-pending security layer designed to audit, govern, and control AI agent reasoning before any system execution occurs on core IBM z/OS environments.

The announcement marks a significant shift in enterprise IT strategy. Rather than attempting risky, wholesale migrations of legacy workloads to the public cloud—a process often fraught with latency issues and regulatory hurdles—organizations are increasingly leaning into "modernization-in-place." By injecting autonomous, agentic AI directly into the mainframe environment, Rocket Software is attempting to capture decades of institutional knowledge and automate complex operational diagnostics.

Solving the COBOL Crisis: Autonomous Agents as Digital Teammates

For decades, conservative enterprise IT departments have relied on highly specialized human expertise to manage mainframe environments. Today, that expertise is vanishing. Universities rarely teach z/OS systems programming, Assembler, or Job Control Language (JCL), favoring modern cloud-native software engineering languages.

According to a June 2026 Hanover Research study cited by Rocket Software, 81 percent of financial services IT leaders report a "very or extremely significant" mainframe skills gap. More tellingly, 87 percent believe AI will be the primary mechanism to offset this shortage within the next two years, with 94 percent ranking AI-enhanced IT operations as a top priority.

"Enterprises have relied on the mainframe to run their mission-critical workloads for decades," said Milan Shetti, president and CEO of Rocket Software. "AI promises to unlock even more value from these environments, but only if enterprises can deploy it securely and without disruption. We’re helping customers apply agentic AI to mission-critical systems with speed, confidence, and control, closing the skills gap and putting that expertise within reach of every enterprise team."

The expanded EVA platform transitions the system from a passive conversational assistant into a proactive digital teammate. Built on the industry-standard Model Context Protocol (MCP), EVA can ingest and correlate vast amounts of disparate z/OS telemetry—including System Management Facilities (SMF) records, Resource Measurement Facility (RMF) data, and job logs. When an end-of-month financial batch run stalls, for example, EVA can autonomously trace the bottleneck across IBM Db2 databases and CICS transaction systems, reducing the Mean Time to Resolution (MTTR) from hours to minutes.

Governing the Machine: The PlanGuard Checkpoint

Running autonomous AI agents in environments that process trillions of dollars in daily transactions introduces distinct, catastrophic failure modes. An unchecked AI agent misinterpreting a performance bottleneck could theoretically cancel a long-running batch job or flush memory buffers inappropriately. Furthermore, granting an AI elevated mainframe privileges risks violating the strict segregation of duties mandated by regulations like the Sarbanes-Oxley Act (SOX), the Payment Card Industry Data Security Standard (PCI-DSS), and the Digital Operational Resilience Act (DORA).

To address these existential risks, Rocket Software introduced PlanGuard. Functioning as an explicit Policy Decision Point (PDP) and Policy Enforcement Point (PEP), PlanGuard establishes an architectural firewall between an AI's reasoning and its system execution.

When an EVA agent formulates an action plan, PlanGuard intercepts the request and evaluates multiple variables, including the caller's identity, the agent's session context, the targeted subsystem, and predefined enterprise safety policies. Crucially, PlanGuard does not replace established native z/OS security frameworks. Instead, it interfaces directly with external security managers like IBM RACF, Broadcom CA-ACF2, and CA-Top Secret.

If an automated action is approved, PlanGuard issues an ephemeral, strictly scoped execution token tied to existing security profiles. Once the task completes, the temporary identity is revoked, preserving an immutable audit trail. If a requested action violates policy—such as attempting a destructive update without proper authorization—PlanGuard blocks the execution and flags the event for human intervention.

The Next Battleground: The Enterprise Data Center Core

Rocket Software's latest release intensifies the competitive race to bring generative AI to the enterprise core. The mainframe modernization landscape is currently divided between vendors advocating for in-situ enhancement and public cloud hyperscalers pushing for workload migration.

Within the modernization-in-place camp, distinct strategies are emerging. IBM's watsonx Code Assistant for Z focuses heavily on the developer lifecycle, utilizing proprietary Granite foundation models to accelerate COBOL-to-Java transformations and code refactoring. BMC Software has integrated agentic capabilities into its Automated Mainframe Intelligence (AMI) portfolio, leveraging its dominance in workload scheduling via its Control-M Agent Gateway.

Rocket Software, backed by Bain Capital Private Equity and bolstered by recent acquisitions of OpenText’s Application Modernization business and Vertica, is carving out a niche in Systems Operations (SysOps) and telemetry correlation. By utilizing a model-agnostic architecture and adopting the open MCP standard, Rocket EVA allows organizations to query disparate mainframe silos without requiring monolithic direct access. Industry analysts project that by 2028, the vast majority of enterprise AI agent frameworks will utilize an MCP-style service model, validating Rocket's architectural direction.

Navigating the Regulatory Tightrope

Global organizations across banking, government, insurance, retail, and telecommunications are currently participating in EVA pilot programs. Early customer implementations have primarily focused on read-only and diagnostic workflows, such as identifying sluggish Db2 queries, tracing hung CICS threads, and discovering legacy post-quantum cryptography (PQC) risks ahead of federal compliance deadlines.

While vendor models suggest substantial annual returns on investment through accelerated diagnostics, the transition to fully autonomous execution remains a measured process. Enterprise risk committees and federal banking examiners are scrutinizing these deployments closely, ensuring that policy checkpoints like PlanGuard satisfy stringent audit and compliance requirements.

As the mainframe workforce continues to age out of the industry, the reliance on agentic AI is no longer a speculative luxury; it is becoming an operational necessity. The success of platforms like Rocket EVA will ultimately depend not just on the intelligence of their underlying language models, but on the robustness of the guardrails keeping those models securely in check.