AI in Maximo 9.2: From Condition Insight to Agentic Workflows, the 2026 Reality

AI in Maximo 9.2: From Condition Insight to Agentic Workflows, the 2026 Reality

The conversation about AI in asset management has shifted decisively. In 2024, AI in Maximo was a promising set of features: Health provided scoring, Predict provided forecasting, and Visual Inspection provided computer vision. Each was useful, but they operated as separate capabilities that required specialized expertise to deploy and interpret. In 2026, with the release of Maximo Application Suite 9.2 in June 2026, IBM has embedded AI more deeply 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 not a marketing claim. It is a description of what shipped. The AI capabilities in MAS 9.2 are integrated into the daily workflow, proven in production at scale, and operationally grounded in the data that Maximo already holds. This article walks through the AI capabilities in Maximo 9.2: Condition Insight, Visual Inspection, Maximo Assistant, the watsonx foundation, and the agentic shift that defines the 9.2 release.

Maximo Condition Insight: The Flagship AI Capability

The single most important AI capability in MAS 9.x is Maximo Condition Insight. IBM introduced it in late 2025, and it 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.

The architectural design of Condition Insight is critical to understanding why it works. It does not replace the existing Health and Predict applications. It sits on top of them, adding a layer of pattern recognition and recommendation generation that the individual applications cannot produce on their own. The data flow has four stages:

Stage 1: Maximo Health calculates health scores. Health scores are numeric values (0-100) that represent the current condition of an asset relative to its expected baseline. The default health score is a weighted combination of several factors: the number of open corrective work orders, the number of open service requests, the remaining useful life (if a Predict model exists), the chronological age as a proportion of expected life, and the asset's criticality.

Stage 2: Maximo Predict applies machine learning models. Predict takes historical failure data from Manage and time-series data from Monitor, builds a model for each asset or asset class, and forecasts the probability of failure, the days to failure, and the most likely failure mode. The output is a numeric score consumed by Health, by a work queue, or by a custom dashboard.

Stage 3: Maximo Monitor collects IoT sensor data. Monitor takes data from PLCs, SCADA systems, IoT sensors, and other operational data sources, normalizes it, and exposes it through time-series storage and analytics. The output feeds into Predict for model training and into Health for real-time condition visibility.

Stage 4: Condition Insight synthesizes everything. Condition Insight brings together the health scores from Health, the failure predictions from Predict, the real-time condition data from Monitor, inspection results, and reliability strategy data to identify patterns that no single data source would reveal on its own. It then generates recommendations for action.

The practical value of this synthesis is significant. Before Condition Insight, a reliability engineer who wanted to understand why a particular asset was degrading would need to manually pull data from Health, Predict, Monitor, and the work order system, correlate the findings, and form a judgment. That process could take hours per asset. Condition Insight performs the synthesis automatically and presents the engineer with a prioritized list of assets that need attention, the likely cause of degradation, and a recommended action.

Condition-based maintenance has traditionally required significant data analysis by experienced specialists. Maximo now simplifies and scales that analysis, reducing complexity and the effort required to interpret large volumes of asset data. Instead of spending time piecing together disconnected information, reliability teams can focus more quickly on the actions most likely to protect uptime and asset performance.

Maximo Visual Inspection: Computer Vision for Asset Health

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 (MVI) uses computer vision models to detect defects, corrosion, hotspots, and other visual indicators of asset condition from photos or video.

The original Visual Inspection capability required custom model training for each use case, which made it expensive to deploy. The 9.x release introduces a low-code Visual Prompting capability that lets subject matter experts train models by highlighting examples in a few images, without writing code or working with labeled datasets. An engineer highlights examples of corrosion on a transformer in five images, the model trains in the background, and it is available for inference within hours. The same engineer can then use the model from a mobile device, pointing the camera at the asset and getting a real-time assessment of whether the defect is present.

The 9.2 release adds local inference directly on the mobile device, which is a major improvement for field use. The deployment options for the inference engine include:

MVI on the cluster. The model runs in the MAS cluster. The technician's device sends the image to the cluster, the cluster returns the classification. Lowest latency for a connected device, requires connectivity at inspection time.

MVI Edge on the device. A smaller model runs on a laptop, a ruggedized tablet, or an industrial gateway. The device sends the image to the local edge, the edge returns the classification, and the result is cached for later sync. This is the right pattern for sites with intermittent connectivity (a remote substation, a mine pit, a rail corridor).

MVI Edge on a gateway. A site-level gateway (NVIDIA Jetson, Intel NUC, or a rugged industrial PC) runs the MVI Edge model, and technicians on the site connect to it. This is the right pattern for a plant with multiple technicians and limited network reach to the central MAS cluster.

The inference is fast enough, typically under 2 seconds per image on MVI Edge, that the technician does not experience it as a delay. The integration between the Maximo Mobile camera pipeline and the MVI inference engine in 9.2 follows a clean pattern:

1. The technician opens the inspection form in Mobile, takes a photo of the asset, and the image is sent to MVI for inference.
2. MVI returns a defect class and a confidence score. The defect class is matched to the failure hierarchy (the MVI model is trained on the customer's failure taxonomy).
3. The technician confirms or overrides the AI's classification. The override is recorded as a feedback signal used to retrain the model.
4. The classification lands in the inspection record and propagates to the work order failure code.

The 9.2 release also introduces geofencing as a first-class feature for inspection workflows. A geofence is defined around an asset (a substation perimeter, a transformer yard, a tank farm) or a linear asset (a pipeline segment, a rail line, a conveyor route). The Mobile app, with the technician's consent, monitors the device's location against the geofence. When the device enters the geofence, the Mobile app surfaces the relevant inspection forms. The technician completes the inspections, the data is recorded locally, and the results sync to Manage when the device exits the geofence or regains connectivity.

Maximo Assistant on Mobile: Natural Language for Field Technicians

Maximo Assistant on Mobile brings AI-powered natural language interaction to field technicians. Technicians can use natural language to find asset information, review work order history, and complete work efficiently in the field. Instead of navigating through menus and search interfaces to find the information they need, they ask questions in plain language.

The practical scenarios are direct. A technician arrives at a pump that has failed. They ask Maximo Assistant: "What is the maintenance history of this pump?" The assistant retrieves the work order history, the asset's health score, the last inspection results, and any open failure codes. The technician asks: "What are the common failure modes for this asset type?" The assistant pulls the failure hierarchy and the reliability strategy for the asset class. The technician asks: "Are there any safety permits required for this work?" The assistant checks the permit-to-work system and the safety plan.

This is not a chatbot. It is an AI assistant that operates within Maximo's security model and data structure. It can only access the data that the technician's security groups permit. It uses the same REST APIs that the Maximo Mobile application uses, which means it has access to the full range of Maximo business objects. The natural language processing is powered by IBM watsonx, which provides the language understanding and generation capabilities.

At the same time, AI-enabled conversational scheduling and what-if analysis empower planners, schedulers, and field service managers to explore changes such as increasing capacity or prioritizing critical work using plain language. A planner can ask: "If I add two technicians to the afternoon shift, how many more work orders can we complete?" The assistant runs the analysis against the current schedule and the available resources and returns an estimate. This kind of what-if analysis, previously requiring custom reporting or spreadsheet work, becomes a conversation.

The watsonx Foundation

The AI capabilities in MAS 9.2 are built on IBM watsonx, IBM's AI and data platform. watsonx provides the machine learning, natural language processing, and generative AI capabilities that power Condition Insight, Visual Inspection, Maximo Assistant, and the new agentic workflows.

For Maximo teams, the watsonx foundation matters for three reasons:

Model governance. watsonx provides model governance capabilities that ensure AI models are traceable, versioned, and auditable. This is critical in regulated industries where AI-driven maintenance decisions must be explainable. A utility regulator who asks why a maintenance decision was made can be shown the model version, the training data, the model's confidence score, and the recommendation it generated.

Integration with the broader IBM AI ecosystem. watsonx is not specific to Maximo. It is IBM's enterprise AI platform. Models built for Maximo can be shared with, and informed by, models built for other IBM applications. An organization that uses watsonx for supply chain optimization can leverage the same platform for asset performance management.

Custom model development. While the out-of-the-box AI capabilities in MAS 9.2 are designed to work without custom model development, data scientists can build custom models using the Jupyter notebooks provided with Maximo Predict. These notebooks run in the watsonx environment and can be customized for specific asset types, failure modes, or operational contexts.

Agentic Workflows: The 9.2 Shift

The most architecturally significant change in MAS 9.2 is the introduction of agentic workflows. The shift from AI as a recommendation engine to AI as an agent that can guide decisions and move work forward is the defining characteristic of the 9.2 release.

The distinction between a recommendation engine and an agentic workflow is important. A recommendation engine says: "Asset PUMP-1024 has a health score of 42 and a 78% probability of failure within 30 days. Recommended action: inspect bearing assembly." A human then decides whether to act on the recommendation. An agentic workflow says: "Asset PUMP-1024 has a health score of 42 and a 78% probability of failure within 30 days. I have checked the work order schedule, and there is a maintenance window on Thursday. I have checked the spare parts inventory, and the bearing assembly is in stock. I have created a draft work order for the inspection, assigned it to the Thursday maintenance window, and reserved the spare parts. The work order is in WAPPR status awaiting your approval."

The agentic workflow does not replace human decision-making. It reduces the time between insight and action by handling the coordination work that previously required manual effort. The human still approves the work order. The agent has simply done the preparatory work that a planner would have needed to do manually.

IBM's announcement of MAS 9.2 describes this as "accelerating work across operational systems with agentic AI." The capability is designed to help guide decisions and move work forward across processes in practical, operationally grounded ways. The key word is "operationally grounded." The agents are not making decisions in the abstract. They are operating within Maximo's existing workflows, security model, and data structure.

Safety and Compliance: AI-Assisted Incident Classification

For organizations focused on safety and compliance, MAS 9.2 connects these responsibilities more directly to daily operations. 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.

AI-assisted incident classification helps improve the consistency and completeness of reporting by suggesting categories and identifying similar events. When a technician records an incident on a mobile device, the AI suggests the most likely incident category based on the description and the asset involved. It also identifies similar incidents that have been recorded, which helps detect patterns that a single incident might not reveal. This makes it easier to detect systemic issues and strengthen overall data quality.

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 safety capabilities are most effective when safety is part of daily operations rather than treated as a separate process.

Document Abstraction: Turning Documents into Operational Intelligence

MAS 9.2 also expands how organizations can use information that has traditionally been difficult to access and analyze. With Lease Abstraction in Maximo Real Estate and Facilities, retrieval-augmented generation (RAG) can help extract key information from leases and make that information usable more quickly. This reduces the time spent reading through long documents and manually entering data into systems.

More broadly, AI-enabled document abstraction will continue to extend beyond leases to other document types such as OEM maintenance manuals, inspection procedures, software licensing agreements, and technical documentation. This represents the starting point for how Maximo can apply RAG across the suite, with the ability to unlock and operationalize information from a wider range of documents over time.

For maintenance teams, the practical application is significant. A technician working on an unfamiliar asset can ask Maximo Assistant to find the relevant section of the OEM maintenance manual. The RAG system retrieves the section, summarizes the relevant procedure, and presents it to the technician. The technician does not need to search through a 500-page PDF. The information is delivered in context, at the point of work.

Practical Implications

The AI capabilities in MAS 9.2 change the way maintenance organizations work, but they do not change the fundamentals. The data still needs to be accurate. The failure hierarchies still need to be defined. The reliability strategies still need to be established. The AI amplifies the value of these fundamentals, but it cannot compensate for their absence.

For organizations deploying MAS 9.2, the recommended approach is:

Start with Condition Insight. It is the flagship capability and the one that delivers the most immediate value. It works with the data you already have in Maximo. You do not need to deploy new sensors or build new data pipelines to get started.

Add Visual Inspection for critical assets. The low-code Visual Prompting capability means you can train a model with a few images. Start with the assets where visual inspection is already part of the maintenance process (corrosion on transformers, wear on conveyor belts, crack detection on rail components).

Deploy Maximo Assistant to your most experienced technicians first. They are the ones who will most quickly identify the gaps in the assistant's knowledge and the scenarios where it adds the most value. Their feedback will improve the system for the broader technician population.

Evaluate agentic workflows for your highest-volume coordination tasks. If your planners spend significant time coordinating work orders, spare parts, and maintenance windows, the agentic workflow capability can reduce that coordination overhead.

Bottom Line

AI in Maximo 9.2 is not experimental. Condition Insight, Visual Inspection, Maximo Assistant, and agentic workflows are production-ready capabilities that are being used by organizations to reduce downtime, improve maintenance efficiency, and accelerate decision-making. The watsonx foundation provides the governance and scalability that enterprise deployments require. The shift from recommendation engine to agentic workflow is the defining architectural change of the 9.2 release, and it represents the direction IBM is taking the platform: AI that does not just suggest actions but helps move work forward within governed, operational workflows.