AI in Maximo 2026: Condition Insight, Predict, Visual Inspection, and the Agentic Shift

As of the Maximo Application Suite 9.2 release in June 2026, every reliability engineer, planner, technician, and safety lead in a Maximo shop is working alongside AI capabilities that have shipped to production in hundreds of organizations. This is not a future state. It is the current state. IBM has embedded AI directly into the workflows that maintenance, field service, safety, and operations teams use every day, and the shift is not subtle.

MAS 9.2 introduces AI capabilities across four distinct dimensions: natural language interaction through Maximo Assistant, predictive analytics through Maximo Predict and Health, visual inspection through computer vision models, and a new category called agentic workflows that can autonomously guide decisions and move work forward. These are not standalone AI tools bolted onto the side of the platform. They are integrated capabilities that operate within the same data model, security framework, and user interface as every other Maximo function.

This article focuses on the AI capabilities that ship with Maximo or with the watsonx platform that IBM bundles with Maximo deployments. The goal is to provide a practical, technically grounded guide to what each capability does, how it works, and how to sequence adoption for maximum value.

The watsonx Foundation: What Maximo AI Runs On

The architectural change that enables the current generation of Maximo AI is the deeper integration between Maximo and the IBM watsonx platform. watsonx provides the model serving, the data ingestion, and the agentic orchestration that the Maximo AI capabilities call into. From the Maximo user's perspective, the AI is just there. From the architect's perspective, the AI is a set of services on the watsonx platform that Maximo calls through well-defined APIs.

watsonx serves as the enterprise AI platform that hosts the large language models, the machine learning models, and the orchestration layer that ties them together. For Maximo deployments, this means the AI models run on the same infrastructure as Maximo, with the same security, governance, and compliance controls. Data does not leave the Maximo environment to reach an external AI service. The models are trained, deployed, and monitored within the watsonx platform, and Maximo calls them through authenticated, audited API calls.

The model serving in watsonx supports both IBM-developed models and third-party models. As of late 2025, IBM deprecated the Granite 3.2 8B Instruct model in favor of the gpt-oss-120b model, which provides significantly better accuracy on complex queries. The model swap was transparent to Maximo users because the watsonx abstraction layer insulates Maximo from the underlying model implementation. When a better model becomes available, IBM can swap it in without requiring Maximo customers to reconfigure anything.

The watsonx integration also provides the data pipeline that feeds the AI models. For Maximo Predict, this means the training data (work order history, failure records, meter readings, sensor data) flows from Maximo through watsonx's data ingestion pipeline into the model training environment. The trained models are deployed back to watsonx's serving layer, where Maximo Predict calls them for inference. This closed-loop architecture means that model retraining can happen automatically as new data accumulates, without manual intervention.

Maximo Condition Insight: Agentic AI for Asset 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 evaluates work orders, metrics, time-series data, meter readings, FMEA (Failure Mode and Effects Analysis) records, and alerts to produce a clear, plain-language assessment of asset condition, emerging trends, and recommended corrective actions.

The key word here is "agentic." Condition Insight does not just provide a score or a threshold alert. It synthesizes multiple data sources, reasons about what they mean together, and produces a recommendation that a reliability engineer can act on. This is the capability that distinguishes MAS 9.2 from previous releases, where AI was limited to individual predictive models that produced failure probabilities without context or explanation.

The technical workflow is as follows. Condition Insight continuously monitors the assets that have been configured for condition-based maintenance. When it detects a change in asset behavior, such as a rising vibration trend, an increase in work order frequency, or a meter reading outside normal range, it triggers an evaluation. The evaluation aggregates all available data for the asset, including recent work orders, inspection results, meter history, FMEA records, and real-time sensor data from Maximo Monitor. The AI agent reasons about this data, identifies the pattern, and produces a plain-language summary of the asset's condition, the emerging trend, and a recommended action.

# Example: Condition Insight evaluation output (simplified)
condition_insight_result = {
    "asset": "PUMP-204",
    "site": "BR-PLANT-01",
    "evaluation_date": "2026-08-05T10:30:00Z",
    "condition": "DEGRADED",
    "confidence": 0.87,
    "summary": "PUMP-204 shows a rising vibration trend on the "
               "outboard bearing over the past 14 days. Vibration "
               "readings have increased from 2.1 mm/s to 4.8 mm/s, "
               "approaching the alarm threshold of 5.0 mm/s. "
               "Work order history shows two bearing-related "
               "repairs in the past 6 months. FMEA record FMEA-0042 "
               "identifies bearing degradation as a high-risk "
               "failure mode for this pump class.",
    "trends": [
        {
            "metric": "vibration_outboard_bearing",
            "trend": "INCREASING",
            "rate_of_change": "0.19 mm/s per day",
            "current_value": "4.8 mm/s",
            "threshold": "5.0 mm/s",
            "days_to_threshold": 1
        }
    ],
    "recommendation": "Schedule bearing inspection within 48 hours. "
                      "If vibration exceeds 5.0 mm/s, initiate "
                      "corrective work order for bearing replacement. "
                      "Check lubrication system for adequate grease "
                      "flow to the outboard bearing.",
    "recommended_action_type": "INSPECTION",
    "priority": "HIGH"
}

The value of this agentic approach is that it scales the expertise of senior reliability engineers. A condition-based maintenance program traditionally requires experienced specialists who can interpret multiple data sources, identify patterns, and recommend actions. There are not enough of these specialists to cover every asset in a typical Maximo installation. Condition Insight performs the initial analysis, producing recommendations that the reliability team can review and act on. The specialists focus on the assets that need human judgment, while the AI handles the routine analysis.

Maximo Predict: Machine Learning for Failure Forecasting

Maximo Predict uses machine learning models to predict asset failures before they occur. The models are trained on historical data, including work orders with failure codes, meter readings, sensor data from Maximo Monitor, and asset characteristics. The output is a failure probability score for each asset, typically calculated daily, that indicates the likelihood of failure within a specified window.

Predict supports several prediction types. Failure probability models estimate the probability that an asset will fail within a given time window, such as 7 days, 30 days, or 90 days. Days to failure models estimate the remaining useful life of an asset. Degradation models estimate the rate at which an asset's condition is declining. The models are deployed in Watson Machine Learning and called by Maximo Predict for daily inference.

The model training process is collaborative. Data scientists work with Maximo Predict's notebook environment to develop and train models using the historical data that Maximo has accumulated. The notebooks provide a starting point with pre-built templates for common prediction types, but they can be customized for specific asset classes, failure modes, or data patterns. Once a model is trained and validated, it is deployed to Watson Machine Learning, and Maximo Predict begins using it for daily predictions.

The practical value of Maximo Predict is best illustrated by the work queue concept. Predict populates work queues with assets that have a high probability of failure or assets that are predicted to fail before the next scheduled preventive maintenance work order. Maintenance planners can review these queues daily and prioritize work based on the predicted failure probability, the criticality of the asset, and the cost of failure. This transforms maintenance planning from a calendar-driven activity to a risk-driven activity.

An Asia Pacific oil and gas producer implemented Maximo Predict and achieved 87% predicted failure accuracy, with some models consistently providing 100% accurate results. The proactive approach avoided $10 million in missed revenue by preventing unplanned critical failures. This is a concrete example of how predictive AI translates into business value when the data foundation is solid and the models are well-tuned.

Maximo Visual Inspection: Computer Vision in the Field

Maximo Visual Inspection brings computer vision capabilities directly to the field. The capability uses AI models to detect defects from images and video, with the option to run inference locally on mobile devices. This is particularly valuable for inspection workflows where a technician photographs equipment and the AI identifies potential defects in real time.

The visual inspection models are trained on labeled images of equipment in various conditions. For example, a model trained on images of insulators on power lines can identify cracked insulators, contaminated insulators, and missing hardware from photos taken during a patrol. The model runs inference on the image and highlights areas of concern, allowing the technician to verify the finding and create a work order if needed.

The on-device inference capability is a significant advancement. In previous versions, images had to be uploaded to a server for processing, which required network connectivity and introduced latency. With local inference, the model runs on the mobile device, providing instant results even in areas without network coverage. This is critical for field inspections in remote locations, such as pipeline patrols, transmission line inspections, and offshore platform walkdowns.

# Example: Visual Inspection workflow (conceptual)
import requests

# Submit image for visual inspection
inspection_url = "https://mas-host/maximo/api/vi/inspect"
headers = {"Authorization": f"Bearer {access_token}"}

image_data = {
    "assetnum": "TRANS-1042",
    "inspection_type": "INSULATOR_CONDITION",
    "image": base64_encoded_image_data,
    "model": "insulator_defect_v3"
}

result = requests.post(inspection_url, headers=headers, json=image_data)
findings = result.json()

# Process findings
for finding in findings.get("defects", []):
    print(f"Defect: {finding['type']}, Confidence: {finding['confidence']}")
    if finding["confidence"] > 0.75:
        # Auto-create inspection work order
        create_inspection_wo(finding["assetnum"], finding["type"])

Maximo Assistant on Mobile: Natural Language in the Field

Maximo Assistant on Mobile helps technicians use natural language to find asset information, review history, and complete work efficiently in the field. The assistant uses the nl2oslc model template, which translates natural language questions into OSLC REST API queries against the Maximo data model. A technician can ask "What is the maintenance history of pump 204?" and the assistant translates that into the appropriate OSLC query, retrieves the results, and presents them in a readable format.

The assistant can access any object in the system, including custom objects, and supports a wide range of queries. It can answer questions about asset specifications, work order status, inventory availability, and failure history. For technicians working on complex equipment with extensive maintenance histories, the ability to ask questions in natural language rather than navigating through application menus is a significant productivity improvement.

The assistant also supports conversational scheduling and what-if analysis for planners and schedulers. A planner can ask "If I add two technicians to next week's schedule, how many work orders can we complete?" and the assistant will model the scenario and provide an answer. This enables faster, more adaptive planning as priorities shift.

Agentic Workflows: The 9.2 Defining Feature

The most significant AI advancement in MAS 9.2 is the introduction of agentic workflows. While Maximo Assistant answers questions and Predict provides recommendations, agentic workflows can take action autonomously within defined boundaries.

An agentic workflow in MAS 9.2 might monitor asset health indicators, detect a degrading condition, create an inspection work order, check inventory for required parts, notify the assigned technician, and update the work order status as the inspection progresses. All of this happens within the governance controls that Maximo provides, with audit logs for every action the agent takes.

The agentic workflows are not uncontrolled AI making decisions on their own. They operate within boundaries that administrators define: which assets they can act on, which work order types they can create, which statuses they can change, and which notifications they can send. The workflows are designed to augment human teams, not replace them. The agent handles the routine coordination work, while humans handle the decisions that require judgment.

The recommended sequence for organizations starting their Maximo AI journey in 2026 is:

1. Audit the data foundation. Look at the asset hierarchy, the meter history, the work order history, and the failure records. Identify the asset classes where the data is complete and consistent enough to support AI. This audit will reveal that some asset classes have excellent data and others have gaps that need to be filled before AI can add value.
2. Start with Monitor and Predict on a single asset class. These capabilities have the most mature data requirements and produce measurable value within the first six months. Choose an asset class that is critical to operations, has good historical data, and has a clear failure mode that predictive models can detect.
3. Add Visual Inspection for high-value inspection types. Identify inspections that are currently manual, time-consuming, and subject to human variability. Train visual inspection models on labeled images and deploy them to mobile devices for field use.
4. Pilot Condition Insight on the same asset classes where Predict and Monitor are deployed. The data foundation that Predict and Monitor require is the same data that Condition Insight uses for its agentic evaluations. Starting with the same asset class means the data is already clean and the team is already familiar with the AI outputs.
5. Roll out Maximo Assistant on Mobile to the field technician population. The assistant benefits from having a broad user base, so rollout should target the entire technician population rather than a small pilot group.

Practical Implications

For Maximo administrators and architects, the AI capabilities in MAS 9.2 change the calculus of deployment planning. The traditional approach of deploying Manage first, then adding Health, then adding Predict, and eventually considering AI is no longer optimal. The AI capabilities are integrated into the platform, not layered on top. Planning for AI from the start, by ensuring data quality and choosing the right initial asset classes, compresses the time to value.

The data foundation requirement cannot be overstated. Every AI capability in Maximo depends on the quality of the underlying data. Assets must be properly classified in the hierarchy, with correct parent-child relationships. Meter readings must be captured consistently, with the right meters on the right assets and readings taken at appropriate intervals. Work order history must include failure codes, not just completion status. FMEA records must exist for the asset classes where Condition Insight will be deployed. Organizations that invest in data quality before deploying AI see results in weeks. Organizations that try to deploy AI on top of poor data quality spend months troubleshooting and never achieve reliable results.

The agentic shift in MAS 9.2 is not a marketing term. It represents a real architectural change in how AI operates within Maximo. The previous generation of AI in Maximo provided recommendations that humans had to act on. The current generation can take action within defined boundaries. This shifts the role of the human from operator to supervisor, reviewing the actions that the AI has taken and intervening when judgment is needed.

Bottom Line

AI inside Maximo in 2026 is asset-first, workflow-native, and grounded in the watsonx platform. Condition Insight provides agentic diagnostics that scale the expertise of senior reliability engineers. Predict provides failure forecasting that transforms maintenance planning from calendar-driven to risk-driven. Visual Inspection brings computer vision to the field with on-device inference. Maximo Assistant puts natural language interaction in the hands of technicians. And agentic workflows coordinate actions across systems without constant human intervention.

The recommended adoption sequence is Monitor and Predict first, Visual Inspection second, Condition Insight third, and Assistant fourth. But every sequence should start with a data quality audit, because the AI capabilities are only as good as the data they are built on. The organizations that succeed with Maximo AI are the ones that treat data quality as a strategic investment, not a cleanup project. The technology is ready. The question is whether your data is.