Mainframe operations are moving beyond the era where teams waited for an alert, diagnosed the problem and then acted. AIOps is increasingly helping organizations detect anomalies, correlate events, anticipate capacity issues and identify likely root causes before business services are affected.
Mainframes continue to support mission-critical operations, so greater automation must be introduced carefully. Progress depends on giving AI enough operational context to act reliably while retaining appropriate controls around decisions that could affect resilience, security or business continuity.
From reactive operations to greater autonomy
The progression can be understood in four stages.
- Reactive operations respond after an incident occurs
- Predictive operations use data to anticipate what may happen
- Proactive operations intervene to prevent disruption
- Autonomous operations go further by executing approved actions independently
Organizations are already moving along this path. One longstanding challenge has been making mainframe telemetry, including System Management Facilities (SMF), Resource Measurement Facility (RMF) and log data, accessible to mainstream observability platforms. Advances in data streaming now transform this information into OpenTelemetry-compatible formats, expanding the range of vendor-agnostic platforms able to consume it.
This supports increasingly mature automation across routine health checks, alert correlation, workload balancing, incident prioritization and predefined remediation workflows. Our work on AI-powered mainframe modernization reflects the same progression toward AI-assisted operations.
Risk determines how far that automation should extend. Restarting services, balancing workloads or resolving known issues are suitable for automation when the action is well understood and reversible. But changes affecting production environments, security policies, sensitive data or regulatory compliance should remain subject to human approval.
Trusted automation develops incrementally, with AI assuming more routine operational work while experienced engineers remain accountable for high-impact decisions.
From recommendations to orchestration
AI is also evolving from an advisory layer into an orchestration capability. Instead of only identifying a likely issue, it can increasingly coordinate actions across applications, infrastructure and operational tools.
Technologies such as Ansible extend configuration-management automation into the mainframe environment, while Terraform supports automated infrastructure provisioning. Combined with Agentic AI, these capabilities widen the range of activities that can be coordinated with less manual intervention.
Broader autonomy requires reliable observability, high-quality operational data, explainable AI, policy-driven governance and rigorous testing under real production conditions. Clear confidence thresholds are equally important—teams need to know which actions AI can execute independently, and which require human review.
Trust grows through consistent performance, transparency and measurable operational outcomes.
Making institutional knowledge accessible
Mainframe operations also depend on decades of specialized institutional knowledge, much of it held by experienced professionals.
AI makes that expertise easier to access. A retrieval-augmented generation (RAG) assistant trained on relevant IBM Redbooks and enriched with organization-specific knowledge gives engineers natural-language access to operational guidance. Integrating the assistant with IT service management records also makes previous incidents and resolutions easier to retrieve.
Newer engineers can reach relevant knowledge more quickly, while experienced professionals can use contextual recommendations to reduce time spent searching documentation and historical records.
Human expertise remains important. Engineers still need to validate recommendations, handle exceptional scenarios and refine operational practices as environments change. Used in this way, AI strengthens the transfer of institutional knowledge without weakening professional judgment.
Building governance into autonomy
As AI begins taking corrective action, governance becomes part of the operational architecture.
Enterprises can begin with lower-risk or read-only activities, such as issuing display commands or performing read-only batch operations. These autonomous actions should operate within predefined confidence thresholds, produce detailed audit trails and support rollback where appropriate. Modifications to production environments should also remain attributable to an identified user or role.
High-impact actions should continue to require human approval, while routine activities should move progressively toward autonomous execution as operational evidence supports greater confidence.
This creates a model in which AI takes on a growing share of routine activity while experienced teams continue to govern resilience, security and business continuity.
Mainframe autonomy will develop through that combination of intelligence and control. The systems become more capable as enterprises gain greater confidence in the data, guardrails and governance surrounding their actions.
