Rocket Software Doubles Down On Agentic AI For The Mainframe

Key Highlights

• Rocket Software expands its EVA platform to bring agentic AI capabilities to mission-critical mainframe environments.
• The introduction of Rocket PlanGuard aims to deliver a stringent policy checkpoint between AI reasoning and system execution.
• High-value operational use cases span diagnostics, security, batch processing, application management, and enterprise data access.
• The platform addresses the widening mainframe skills gap by preserving institutional expertise and reducing reliance on specialized talent.
• Governed access and contextual authorization are architected to ensure AI-driven actions remain strictly within approved enterprise boundaries.

The News

Rocket Software announced the expansion of Rocket EVA, its agentic AI platform designed for mission-critical mainframe systems. The update introduces governed, auditable AI agents that can reason across operational data and automate tasks within strict enterprise policies. A key addition is Rocket PlanGuard, a security layer that places a policy checkpoint between AI reasoning and system execution. Find out more by clicking here to read the press release.

Analyst Take

We have been tracking the intersection of AI and legacy infrastructure for several years. Historically, the mainframe community views external automation and AI with deep suspicion. Although this dynamic is changing, HyperFRAME Research’s own datastates that 64% either agree or strongly agree with the following statement: “Our existing mainframe environment provides all the necessary infrastructure and tooling to support our organization’s AI strategy”.

The systems running the global financial backbone do not tolerate downtime, and introducing autonomous agents into these environments raises immediate alarm bells for many legacy operators and system admins. Yet many enterprises still face a significant mainframe skills gap. Experienced programmers are retiring. We see a structural talent deficit that traditional hiring cannot fix.

This reality forces a difficult compromise. IT leaders must choose between maintaining slow, manual operations or risking stability by automating complex tasks. The Rocket EVA expansion attempts to navigate this exact tension. By introducing agentic AI to the mainframe, Rocket Software aims to deliver operational intelligence that can autonomously diagnose and resolve issues. The approach relies heavily on preserving veteran operators' institutional knowledge while shielding the underlying systems from rogue AI actions. It is a bold strategy. Risk is inherent in autonomous execution.

What was Announced

Rocket Software expanded its Rocket EVA platform, adding agentic AI capabilities designed to operate within mission-critical mainframe environments. The core of this expansion is the launch of Rocket PlanGuard. This new security layer establishes a strict policy checkpoint between the AI's reasoning engine and actual system execution. PlanGuard integrates a policy decision point alongside identity controls. These mechanisms ensure any AI-driven actions remain confined to enterprise-approved boundaries. The platform leverages natural language processing to interact with operational data, aiming to reduce the manual burden of system diagnostics.

Specific operational use cases targeted by the expanded EVA platform include advanced diagnostics, security monitoring, batch processing management, and configuration oversight. The system connects directly with z/OS data sources, such as SMF records and RMF performance data, using a lightweight, standards-based framework. It targets granular issues like IBM CICS queue backlogs, high CPU utilization anomalies, and Db2 buffer pool degradation. Furthermore, the platform aims to deliver just-in-time authorization for AI agents, keeping their actions strictly governed. All activities are logged for full auditability, and human oversight is supported where the enterprise requires it.

Our analysis suggests that Rocket Software is acutely aware of the security fears surrounding agentic AI. Introducing PlanGuard is a direct response to those concerns. Enterprises want the efficiency of automation, but they demand absolute control. Placing a hard boundary between the AI's recommendations and the execution of those commands is essential. However, this architecture's effectiveness will depend entirely on the quality of the enterprise policies defining those boundaries. A poorly configured policy engine will either hamstring the AI or expose the system to unintended consequences.

Pilots currently underway across financial services, government, insurance, retail, and telecommunications will serve as a crucial proving ground. Moving from installation to actionable insights in days sounds promising, but mainframe environments are notoriously complex and highly customized. Each deployment will inevitably uncover unique operational quirks. We expect early adopters to lean heavily on the human-in-the-loop features before trusting the AI with autonomous execution rights.

Moreover, the promise of reducing Mean Time to Resolution is attractive, yet it shifts the burden of work. Instead of manually parsing logs, operators will need to verify an AI agent's reasoning. This requires a different type of technical literacy. While Rocket EVA is designed to capture the expertise of seasoned professionals, junior staff will still need sufficient knowledge to determine if the AI is making a sound recommendation. This dynamic introduces a subtle intellectual ambiguity. Does AI close the skills gap, or does it simply change the skills required?

We must also consider the broader operational context. Mainframe modernization is rarely a standalone project. It integrates deeply with hybrid cloud strategies, data analytics pipelines, and stringent compliance frameworks. Rocket Software’s choice to make EVA model-agnostic is a smart architectural decision. It prevents vendor lock-in regarding the underlying large language models. Enterprise IT environments are heterogeneous; forcing a single AI model upon a customer would severely limit adoption. By focusing on the connective tissue and the governance layer, Rocket positions itself as an enabler rather than an AI dictator.

Yet a model-agnostic approach has trade-offs. Abstracting the AI layer can sometimes lead to lowest-common-denominator performance, also what is to stop SysAdmins always picking the most expensive models, leading to increased token usage and cost implications. We will watch how effectively EVA translates generalized model intelligence into the highly specific, esoteric language of z/OS diagnostics. The system must understand the precise difference between a transient CICS slowdown and a systemic Db2 deadlock. If the AI hallucinates or misinterprets these signals, the resulting automated actions could be disastrous. That is why the PlanGuard security layer isn't just a feature; it is the fundamental premise that makes the entire product viable.

The competitive dynamics in this space are also shifting. Traditional hardware vendors and emerging software upstarts are all vying for a piece of the mainframe modernization budget. Rocket Software has a distinct advantage because of its deep historical footprint in enterprise systems. They understand the conservative nature of their customer base. They know that a bank will prioritize a transparent audit log over a flashy natural language interface every single time. By leading with governance and auditability, Rocket is speaking the language of mainframe administrators.

Ultimately, Rocket Software is attempting to bridge a massive divide. They are bringing modern, probabilistic AI to deterministic, legacy systems. The success of this endeavor will rely on balancing aggressive automation with paranoid security controls. We see this announcement as a necessary step for the mainframe ecosystem. Doing nothing is no longer a viable strategy for enterprises bleeding institutional knowledge. We anticipate that initial deployments will be heavily scrutinized, but if the promised efficiencies materialize, agentic AI could become standard issue for core system operations.

Looking Ahead

Rocket Software's announcement highlights a systemic tension within enterprise IT. Organizations are desperate to modernize, yet they remain tethered to the mainframe's uncompromising stability and security. We believe governed AI offers a viable bridge between these competing realities. However, this transition will be bumpy.

Based on what weare seeing, the adoption of generative AI in core systems is rapidly shifting from basic code generation to operational autonomy. The key trend we'll be watching is the evolution of enterprise trust architectures. Solutions like Rocket PlanGuard serve as early indicators of how the industry will govern non-deterministic systems operating inside highly deterministic environments. Competitors will likely follow suit; you could argue its a two-horse race for leadership with BMC, although we expect a race to develop the most robust policy and authorization engines for AI agents. The market demands verifiable proof that AI cannot break critical systems.

Going forward, we are going to be closely monitoring how the company performs on its promise of rapid time-to-value. Mainframe integrations traditionally take months, rarely days. If Rocket Software can consistently demonstrate successful implementations across its pilot programs, it will establish a formidable competitive moat against hardware incumbents. We will be tracking how the company does with its diverse edge cases and highly customized z/OS environments in future quarters. The market will undoubtedly be watching these adoption metrics closely as they assess the viability of agentic AI in legacy enterprise technology. We certainly will.