AI in Maximo 2026: From Condition Insight to Agentic Workflows

The AI Inflection Point in Asset Management

Maximo Application Suite 9.2, released in June 2026, represents the most consequential step in IBM's AI strategy for asset management. Rather than treating AI as a separate layer that sits on top of existing workflows, MAS 9.2 embeds AI directly into the daily work of managing assets, teams, safety, and operations. The release expands AI across reliability insights, field execution, safety and compliance workflows, document-based information extraction, and orchestration across systems. It also introduces agentic workflows designed to help guide decisions and move work forward in operationally grounded ways.

This is an inflection point. Organizations already have access to large volumes of operational data from sensors, inspections, work orders, and meter readings. The challenge has never been data collection. The challenge is using that data consistently across teams and processes. Reliability teams need to spot changes before they become failures. Technicians need access to information at the point of work. Safety leaders need better ways to connect reporting and corrective action. Operations and IT teams need systems that work together without constant manual coordination. MAS 9.2 addresses each of these needs with AI capabilities that are integrated into existing Maximo workflows rather than requiring users to switch to separate AI tools.

The foundation for all AI capabilities in MAS is IBM watsonx, which provides the machine learning and large language model infrastructure that powers everything from predictive models to conversational interfaces. The integration with watsonx means that Maximo's AI capabilities benefit from IBM's ongoing investment in foundation models while remaining grounded in the trusted asset data model that Maximo provides. This combination of foundation model capabilities with domain-specific asset data is what distinguishes Maximo's AI from generic AI tools. The AI understands asset hierarchies, failure modes, maintenance histories, and operational context because it is connected to the system of record.

The progression from MAS 9.1 to 9.2 is also noteworthy. MAS 9.1, released in June 2025, introduced the Maximo AI Service as a foundation, along with Maximo Assistant for conversational queries, a similarity tracker for work order matching, and an FMEA content builder. These were foundational capabilities that established the AI infrastructure. MAS 9.2 builds on that foundation with Condition Insight, on-device visual inspection, conversational scheduling, AI-assisted incident classification, and RAG-based document abstraction. The two releases together represent a comprehensive AI layer that spans the full asset management workflow.

Maximo Condition Insight: Agentic Diagnostics

The single most important AI capability in MAS 9.x is Maximo Condition Insight, which IBM introduced in late 2025 and which has matured into a flagship feature by mid-2026. Condition Insight is an agentic AI capability within Maximo Asset Performance Management (APM) that interprets asset data to explain asset condition, highlight emerging trends, and recommend corrective actions. It works in concert with the other MAS applications to deliver a unified, condition-based maintenance approach across the Maximo ecosystem.

Condition Insight evaluates work orders, metrics, time-series data, meter readings, Failure Mode and Effects Analysis (FMEA) data, and alerts to assess an asset's condition. It then provides a clear, explainable summary of what is happening, why it matters, and what to do next. The output is communicated in plain language, making it accessible to maintenance teams without requiring data science expertise.

The agentic nature of Condition Insight is what sets it apart from traditional predictive analytics. Rather than simply flagging that an asset has a high probability of failure, Condition Insight investigates the asset's history, examines related work orders, checks inspection results, evaluates meter reading trends, and synthesizes all of this information into a narrative explanation of the asset's current state. It then recommends specific corrective actions, prioritized by impact and urgency.

For reliability engineers, this capability addresses a persistent challenge. Condition-based maintenance has traditionally required significant data analysis by experienced specialists. A reliability engineer might spend hours piecing together information from work order history, inspection reports, sensor trends, and FMEA documentation to understand why an asset is behaving abnormally. Condition Insight performs this synthesis in seconds, producing an explainable summary that the engineer can validate and act on. The explainability is critical. The AI does not just provide an answer. It shows its work, citing the specific data points and patterns that led to its conclusion.

The business impact is measurable. By reducing the time required for condition assessment, Condition Insight enables reliability teams to cover more assets with the same headcount. By standardizing the assessment approach, it reduces variability in decision quality between experienced and junior engineers. And by recommending specific actions, it shortens the path from insight to execution, reducing the window between identifying an issue and completing the corrective work order.

An important design decision in Condition Insight is that it does not automate the decision. It recommends, and the reliability engineer decides. This human-in-the-loop approach is appropriate for asset management decisions that involve safety, cost, and operational tradeoffs. The AI provides the analysis and the recommendation, but the human takes responsibility for the action. This is the right balance for industrial environments where the cost of a wrong decision can be measured in safety incidents and production losses. As trust in the AI's recommendations builds over time, organizations may choose to automate certain low-risk recommendations, but the starting point should always be human review.

Maximo Visual Inspection: On-Device AI

Visual inspection is the second major AI capability in MAS 9.x, and it is the one that has improved the most in the last 12 months. Maximo Visual Inspection uses computer vision models to detect defects, corrosion, hotspots, and other visual indicators of asset condition from photos or video. In MAS 9.2, the capability takes a significant step forward with local inference directly on the mobile device.

On-device inference is a meaningful architectural choice. Previously, visual inspection required sending photos from the field to a server for analysis, which introduced latency and dependency on network connectivity. In remote environments such as offshore platforms, rural substations, or underground mines, network connectivity is unreliable or nonexistent. By running the inference model locally on the technician's device, MAS 9.2 enables visual inspection in any environment, with results available immediately at the point of work.

The computer vision models are trained on asset-specific defect data, which means they can be customized for the particular equipment types in an organization's asset base. A utility might train models to detect insulator damage, transformer oil leaks, and conductor wear. A manufacturer might train models to detect conveyor belt wear, bearing housing cracks, and hydraulic fluid seepage. The training process uses labeled images of known defects, and the resulting models are deployed to mobile devices through the Maximo Mobile application.

The workflow integration is where Visual Inspection delivers practical value. A technician performing an inspection opens the Maximo Mobile app, selects the asset, and takes a photo. The on-device model analyzes the photo in seconds and flags any detected defects. If a defect is found, the technician can create a work order directly from the inspection screen, with the photo, defect classification, and recommended action pre-populated. This closes the loop between inspection and action without requiring the technician to navigate between applications or rekey information.

For organizations with large asset populations that require periodic visual inspection, such as pipelines, transmission lines, or facility roofs, the combination of Visual Inspection with drone-captured imagery creates a powerful inspection workflow. Drones capture photos at scale, the computer vision models screen them for defects, and inspectors focus their expertise on the flagged images rather than reviewing every photo manually. This approach can reduce inspection time by orders of magnitude for linear asset populations. A pipeline inspection that previously required weeks of manual photo review can be completed in days, with the AI pre-screening thousands of images and surfacing only the ones that need human attention.

Model management is an ongoing operational consideration. As assets age and new defect types emerge, the computer vision models need to be retrained with new labeled images. Organizations should establish a model governance process that includes periodic accuracy reviews, false positive and false negative analysis, and controlled model updates. The Maximo Visual Inspection platform supports model versioning, allowing teams to compare model performance and roll back to previous versions if a new model performs worse on specific asset types.

Maximo Assistant and Conversational Scheduling

MAS 9.1 introduced the Maximo AI Service, which provides the foundation for AI-driven experiences across the suite. The first visible capability powered by this service was Maximo Assistant, which allows users to query Maximo work orders, assets, and service requests conversationally. Rather than navigating through the Maximo UI to find asset information, a technician can ask, "What is the maintenance history of pump P-1001?" and receive a structured answer with relevant work orders, inspections, and known issues.

In MAS 9.2, Maximo Assistant on Mobile extends this capability to the field. Technicians can use natural language to find asset information, review work order history, and complete work efficiently. The mobile assistant understands context, so a technician can say, "Show me the last three work orders for this asset" while standing in front of the equipment, and the assistant uses the asset QR code or NFC tag to identify the asset and retrieve the history.

The conversational scheduling and what-if analysis capabilities introduced in MAS 9.2 extend AI to planners, schedulers, and field service managers. These users can explore changes such as increasing capacity or prioritizing critical work using plain language. The AI processes the request, evaluates the current schedule against constraints (resource availability, skill requirements, SLA deadlines, asset criticality), and presents the optimized assignments with explanations of why the changes were recommended.

This is not autopilot. The AI does not execute schedule changes without human approval. It proposes changes and explains the reasoning, allowing the planner to accept, modify, or reject the recommendation. This human-in-the-loop approach is appropriate for scheduling decisions that involve tradeoffs between competing priorities. The AI's value is in processing the volume of variables (dozens of technicians, hundreds of work orders, multiple skill types, varying SLA requirements) faster than a human could, not in replacing human judgment about which tradeoffs are acceptable.

The similarity tracker, another AI Service capability, accelerates the progression and closure of work orders and tickets by identifying similar historical records. When a technician enters a problem description for a new work order, the similarity tracker surfaces similar work orders that were previously resolved, along with their solutions. This can dramatically reduce diagnosis time for recurring issues, particularly in organizations with high technician turnover where institutional knowledge is not fully captured in documentation. The FMEA content builder, also part of the AI Service, helps reliability engineers construct failure mode libraries by analyzing historical failure data and suggesting failure modes, causes, and effects based on patterns in the work order history.

AI-Enabled Safety and Compliance

MAS 9.2 expands AI into safety and compliance workflows, connecting these responsibilities more directly to daily operations. The release introduces AI-assisted incident classification that suggests categories and identifies similar events when an incident is reported. This improves the consistency and completeness of reporting by reducing the variability that occurs when different people classify incidents differently.

The practical impact is on data quality. Incident classification is the foundation for safety analytics. If the same type of incident is classified differently by different reporters, trend analysis becomes unreliable and patterns are missed. AI-assisted classification applies consistent criteria across all incidents, and the identification of similar events helps safety teams detect patterns that might otherwise remain hidden in the incident database. For organizations with regulatory reporting requirements, consistent classification also simplifies compliance reporting by ensuring that incidents are categorized according to the reporting framework rather than individual interpretation.

Expanded capabilities across asset-centric waste management, contractor safety oversight, and mobile-first safety workflows make it easier to manage compliance, track risk, and act in real time. Field teams can capture incidents, complete inspections, and initiate permit-to-work processes directly on a mobile device, improving data accuracy and responsiveness while reducing reliance on manual or disconnected processes. The integration of safety workflows with asset and work data means that safety assessments are visible in the context of the work being performed, not in a separate safety system that requires context switching.

Document Intelligence with RAG

MAS 9.2 introduces retrieval-augmented generation (RAG) capabilities for document processing, starting with Lease Abstraction in Maximo Real Estate and Facilities. This capability extracts key information from lease documents and makes it usable in the system without manual data entry. The RAG approach combines a large language model's ability to understand unstructured text with a retrieval mechanism that finds relevant sections of the document, producing structured output that can be validated and imported.

While lease abstraction is the starting point, IBM has indicated that AI-enabled document abstraction will extend to other document types including OEM maintenance manuals, inspection procedures, software licensing agreements, and technical documentation. For maintenance organizations, the ability to extract structured data from OEM manuals is particularly valuable. Maintenance procedures, torque specifications, inspection intervals, and safety warnings are typically buried in hundreds of pages of OEM documentation. RAG can extract this information and link it to the relevant asset records, making it accessible at the point of work.

The document intelligence capability represents a shift from AI that analyzes operational data (sensor readings, work order history) to AI that operationalizes unstructured information. Organizations have vast amounts of unstructured asset-related documentation that is technically accessible but practically unusable because finding the relevant information requires reading through lengthy documents. RAG changes this equation by making the content searchable and extractable on demand. The potential time savings are substantial. A maintenance engineer who currently spends 30 minutes finding the correct torque specification in a 400-page OEM manual could retrieve the same information in seconds through a RAG-powered query.

Practical Implications

For organizations on MAS 9.x, the AI capabilities are not a future roadmap. They are available now and should be evaluated against specific operational challenges. The implementation approach should be incremental rather than transformative. Start with Condition Insight on a pilot group of critical assets where the maintenance team can validate the AI's recommendations against their own expertise. This builds trust in the technology and identifies any data quality gaps that need to be addressed before broader rollout.

For Visual Inspection, the key prerequisites are a library of labeled defect images for training and mobile devices that support the on-device inference models. Organizations without labeled image libraries should start by capturing and labeling images during routine inspections, building the training dataset incrementally. The model accuracy improves with more training data, so this is an investment that compounds over time. Even a few hundred labeled images per defect type can produce a useful initial model.

For Maximo Assistant, the main consideration is data quality. The assistant's answers are only as good as the underlying asset and work order data. Organizations with incomplete asset records or inconsistent work order coding will get limited value from the assistant until they address the data quality issues. This is not a technology problem. It is a data governance problem that requires sustained attention to data entry standards and validation rules.

The conversational scheduling and what-if analysis capabilities require well-maintained resource and skill data. If the system does not know which technicians have which skills, or if craft availability is not accurately reflected in the schedule, the AI's recommendations will be based on incomplete information. Before enabling these capabilities, ensure that the resource, skill, and availability data in Maximo is current and reliable. The effort to clean up this data pays dividends beyond the AI features, improving scheduling accuracy across the board.

Bottom Line

MAS 9.2 makes AI in asset management operationally real. Condition Insight, Visual Inspection, Maximo Assistant, AI-assisted safety classification, and RAG-based document intelligence are not experiments. They are production capabilities integrated into the workflows that maintenance, reliability, safety, and operations teams use every day. The organizations that will see the most value are those that approach AI adoption with the same discipline they apply to any operational change. Start with a specific problem, pilot on a manageable scope, validate the results, and expand based on evidence. The AI is ready. The question is whether the data, the processes, and the team readiness are in place to support it.