From predictive maintenance to Agentic AI: the next step in intelligent maintenance

Predictive maintenance has become an established part of many digital maintenance strategies. Sensor data, condition monitoring and analytical models can identify patterns that indicate degradation or an increased risk of failure.

However, the output often remains a prediction, an alarm or a health indicator. That is when the work of the maintenance or reliability professional begins: reviewing historical data, investigating possible causes, consulting technical documentation, assessing asset criticality and determining the required intervention.

It is precisely in this follow-up process that the potential of a new generation of AI applications lies.

Generative AI can summarise information, search technical documentation and support users in interpreting knowledge. Agentic AI adds goal-oriented action.

In principle, an AI agent can combine information from multiple sources, navigate between different digital systems and execute a sequence of steps to achieve a predefined objective.

Suppose a condition monitoring system detects an increased failure risk in a bearing. A future AI system could not only explain the anomaly, but also consult maintenance history and technical documentation, structure possible causes, take asset criticality and production impact into account, and prepare relevant information for work preparation or maintenance planning.

That is fundamentally different from a standalone predictive model. The value lies not only in the analysis, but in the connection between diagnosis, context and follow-up action.

For maintenance organisations, an important application area therefore lies in connecting existing systems.

Condition monitoring platforms, EAM or CMMS systems, document management systems, spare parts information and production data often each contain part of the information required to make a maintenance decision.

In theory, Agentic AI can bring this information together more quickly. This may be particularly relevant in processes where engineers or planners currently spend significant time searching for, checking and transferring information.

The potential impact is therefore not limited to increased automation. The quality and speed of decision-making may also change.

For KPIs such as Mean Time to Repair (MTTR), planned versus unplanned maintenance, schedule compliance and asset availability, the time between detection, diagnosis and an executable maintenance decision can become particularly relevant. Whether AI actually improves these KPIs, however, depends strongly on the quality of the underlying maintenance process.

The technical possibilities do not change a fundamental principle of maintenance and asset management: decisions can only be as good as the information on which they are based.

An AI agent that accesses incorrect asset data, outdated procedures or incomplete maintenance history will not automatically solve those shortcomings.

This makes data quality, master data management, asset structures and traceability more important, not less.

Organisations must be able to determine which data can be considered a reliable source, how current that information needs to be and how clearly it remains visible which sources were used to generate a recommendation or action.

This challenge aligns with broader asset management principles. ISO 55000 highlights, among other things, the importance of integrated decision-making, value creation across the asset lifecycle and alignment between assets, information, risk and organisational objectives. AI does not change those principles, but it places new demands on the digital information chain that supports such decisions.

Agentic AI in an industrial environment does not automatically mean full autonomy.

Especially for critical or safety-critical assets, organisations need to define in advance which systems an AI agent may access, which data it may modify and which actions require human approval.

This creates a new governance challenge. A technical system may, for example, propose an intervention or prepare a work order, but that does not mean it should be allowed to shut down an installation autonomously or release a maintenance action.

Authorisations, validation rules, logging and human oversight therefore remain essential.

Responsibility must also remain explicit. When an AI system gathers information or formulates a proposal, the organisation remains responsible for how that proposal is validated and applied.

The development of Agentic AI also changes the order in which investment questions should be considered.

An organisation with incomplete asset registers, inconsistent failure coding or insufficiently standardised work processes will not solve those shortcomings by adding another AI layer.

This theme will be addressed explicitly at Asset Performance 2026. Tor Idhammar, Director of Reliability at Domtar, will open the conference with AI for Reliability at Domtar: Hype, Help, or Hard Truths? His starting point is familiar to many industrial organisations: advanced technology creates little value when the foundations of maintenance and reliability are insufficiently developed.

From a maturity perspective, this means AI should be seen as part of a broader reliability and asset management architecture. Processes, data, roles and technical systems need to be sufficiently stable before more autonomous applications can be deployed responsibly.

In the longer term, the discussion goes even further.

Prof. Diego Galar of Luleå University of Technology and SISTEPLANT will close Asset Performance 2026 with From Agentic AI to Self-Preserving Assets: Redefining Maintenance and Asset Management.

The concept of the ‘self-preserving asset’ combines, among other things, condition-based maintenance, reliability engineering, digital twins and autonomous decision-making.

Today, this is more of a future vision than a description of common industrial practice. It does, however, illustrate the direction in which the discussion is evolving: from assets that report their condition to systems that can interpret more context and support the choice of a follow-up action.

The maintenance professional does not disappear in this evolution. As digital systems provide more decision support, technical expertise, risk assessment and clearly defined responsibilities become even more important.

A distinctive perspective during the conference will come from Dr Stefaan De Mey, Head of the Strategy Team for Human and Robotic Exploration at the European Space Agency.

Under the title Sending Spare Parts to Mars is a Bad Business Case, he will discuss systems that need to continue operating reliably for long periods without direct access to spare parts, specialist technicians or external intervention.

Conditions in space exploration are obviously very different from those in a manufacturing site, power plant or water treatment facility. Yet the underlying asset management question is familiar: how do you design and manage critical assets when failure is difficult, expensive or sometimes impossible to compensate for?

This directly connects autonomy with maintainability, reliability, logistical risks and performance across the full asset lifecycle.

For maintenance and asset management organisations, the key question is therefore not how advanced an AI model is.

More relevant is which specific step in the maintenance process it improves. Can technical information be found faster and more reliably? Is diagnosis better supported? Can work preparation become more efficient? Does it remain clear what information a recommendation is based on? And are responsibilities and authorisations properly defined?

This shifts the assessment of technology from functionality towards process performance and risk control.

Asset Performance 2026, taking place on 18 and 19 November at the Flanders Meeting & Convention Center Antwerp, will explore this evolution from several perspectives. In addition to Agentic AI, the programme will cover AI Voice Agents, predictive and prescriptive maintenance, asset health, asset data management, EAM, reliability strategies and data-driven asset management.

Agentic AI can represent a next step in the digitalisation of maintenance: moving from predicting problems towards supporting the actions that follow.

The technical ability to connect multiple systems and process steps simultaneously creates new requirements for data quality, governance, authorisation and human validation.

For maintenance, reliability and asset management professionals, the core issue is therefore not maximum autonomy. The relevant question is where AI can demonstrably contribute to better-informed decisions, more reliable assets and manageable risks across the full asset lifecycle.