Abstract
Organizations are increasingly able to buy or build AI agents that retrieve maintenance data, recommend work, create records, and coordinate workflows. Technical capability does not mean organizational readiness. This paper argues that industrial AI autonomy should expand only as data quality, work processes, governance, and verification mature around it.
The adoption trap
A plant can add an advanced AI assistant to a weak maintenance system and obtain a faster weak maintenance system. Poor asset hierarchy, unreliable bills of material, inconsistent work-order descriptions, weak closeout, undefined planning standards, and unclear approval authority do not disappear when a model is connected to the CMMS. In some cases AI makes these deficiencies more consequential by propagating them faster.
Research is converging on staged adoption
A 2026 process-centric review of agentic AI in maintenance mapped evidence from 42 studies and found greater maturity in technical areas such as diagnostics, prognostics, scheduling, and execution control than in strategic areas such as criticality, lifecycle costing, and KPI alignment. The authors propose a six-stage workflow for introducing agentic AI, meant to keep projects out of pilot purgatory and to let autonomy grow in order, from diagnostic assistance toward strategic, lifecycle-level decision support.
This is the right direction. Organizations should not advance because the model can perform a task in a demonstration. They should advance because the maintenance system can verify, govern, and learn from that task in production.
A practical maturity ladder
Stage 1, Search. AI retrieves controlled documents and records.
Stage 2, Review. AI identifies missing information, contradictions, stale evidence, and quality problems.
Stage 3, Recommend. AI proposes diagnostic steps, priority, scope, or maintenance actions while humans retain decision authority.
Stage 4, Prepare. AI drafts work orders, job plans, schedules, and closeout structures.
Stage 5, Coordinate. AI orchestrates tools and workflows across CMMS, ERP, documents, and condition systems under explicit permissions.
Stage 6, Execute bounded actions. AI performs transactional or physical actions only where engineered limits, auditability, verification, and recovery are mature.
The best first projects are narrow bottlenecks
The most practical AI projects often reduce expert review load without replacing expert authority: backlog readiness screening, work-order quality control, historical evidence retrieval, inspection triage, closeout review, duplicate detection, or shutdown work-package assurance. These problems have clear inputs, bounded outputs, and measurable failure modes.
They also improve the maintenance system itself. A readiness review forces better work definition. A closeout review creates better history. An evidence layer exposes stale master data. Those improvements make later automation safer.
Conclusion
Maintenance maturity and AI maturity are coupled. The plant that attempts to skip governance and process discipline will eventually automate its own ambiguity. The strongest adoption strategy is therefore not “AI everywhere.” It is a sequence of narrow capabilities, each earning greater authority only after the surrounding system can verify what it does.
References
- Crespo-Márquez & Gómez Fernández, “Agentic AI for maintenance management: a process-centric review and a staged framework for industrial adoption,” Computers & Industrial Engineering, 2026. https://www.sciencedirect.com/science/article/pii/S0360835226004869
- IBM, “Introducing Maximo Application Suite 9.2,” June 25, 2026. https://www.ibm.com/new/announcements/introducing-maximo-application-suite-9-2
- IBM Docs, “Connecting external AI agents to the Maximo MCP server.” https://www.ibm.com/docs/en/masv-and-l/cd?topic=developing-connecting-external-agents
- Prometheus Group, “From Hours to Minutes: How AI Is Transforming Maintenance Planning and Scheduling,” June 26, 2026. https://www.prometheusgroup.com/resources/posts/from-hours-to-minutes-how-ai-is-transforming-maintenance-planning-and-scheduling