AI in Maximo: How Watsonx, Maximo Assistant, and Agentic Workflows Are Transforming Asset Management

Introduction

Artificial intelligence in asset management has reached an inflection point. For years, the conversation around AI in Maximo was about potential -- what might be possible with enough data, enough compute, and enough time. That conversation has shifted decisively to the present tense. With the release of Maximo Application Suite 9.2 in June 2026, IBM has embedded AI directly into the workflows that reliability, maintenance, field service, safety, and operations teams use every day.

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, 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 provides a comprehensive overview of AI in Maximo as of mid-2026. We will cover the core AI capabilities, how they work under the hood, practical deployment considerations, and what the roadmap looks like for the next 12 to 24 months. Whether you are just starting your AI journey or looking to expand existing deployments, the patterns and practices in this article will help you navigate the rapidly evolving landscape of AI-powered asset management.

Maximo Assistant: Natural Language Access to Asset Intelligence

The most visible AI capability in MAS is Maximo Assistant, a generative AI assistant powered by IBM watsonx that enables users to interact with asset data using natural language. Instead of navigating through application menus and running reports, users can simply ask questions and get answers.

Maximo Assistant was introduced in MAS 9.1 with support for work orders, assets, and service requests. In MAS 9.2, the assistant has been significantly expanded. It can now access any object in the system, including custom objects, and supports a much wider range of queries. The assistant uses the nl2oslc model template, which translates natural language questions into OSLC REST API queries against the Maximo data model.

Here is how the assistant works in practice. A maintenance planner types: "Which work orders are missing job plans?" The assistant translates this into an OSLC query, executes it against the Maximo API, and returns the results in a readable format. A reliability engineer asks: "Show me assets with declining performance trends." The assistant analyzes meter readings and work order history to identify assets whose performance metrics are trending downward. A plant manager asks: "What is the total maintenance cost per site this quarter?" The assistant aggregates cost data across work orders and returns a summary.

The assistant is configurable at multiple levels. Administrators can control which object structures the assistant can access, set default query templates, and define quick starters -- common questions that users can select with a single click. Security is inherited from the existing Maximo security model: users can only get answers about data they have permission to access.

The model behind the assistant has evolved rapidly. As of November 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. Organizations running Maximo AI Service on-premises need adequate hardware to support the larger model, including GPU resources for inference.

Configuration Best Practices

Getting the most out of Maximo Assistant requires thoughtful configuration. The default configuration provides access to work orders, assets, service requests, meters, preventive maintenance, job plans, work logs, assignments, and persons. For most organizations, this is a good starting point. But the real value comes from extending access to custom objects and domain-specific data.

When configuring the assistant, focus on the questions your users actually ask. Review the quick starters and customize them for your organization's terminology and priorities. A utility might configure quick starters about transformer loading and dissolved gas analysis results. A manufacturer might focus on OEE trends and line stop causes. The assistant is most valuable when it speaks your organization's language.

One important limitation to understand: the assistant does not currently include a feedback mechanism for answers, and conversation history cannot be saved. This means users cannot refine a question based on a previous answer in the same session. IBM has indicated that these capabilities are on the roadmap, but for now, the assistant is best used for straightforward queries where the answer is clear from the data.

Predictive AI: From Condition Monitoring to Failure Prediction

Beyond the conversational assistant, MAS includes a suite of predictive AI capabilities that have been maturing over multiple releases. Maximo Predict, Health, and Monitor work together to provide a complete predictive maintenance platform.

Maximo Predict uses machine learning models to predict asset failures before they occur. The models are trained on historical data -- work orders with failure codes, meter readings, sensor data from 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.

The types of models available have expanded significantly. Failure probability models predict the likelihood of failure within a configurable time window, typically 30 days. Remaining useful life (RUL) models estimate how much longer an asset can operate before it needs replacement. Multi-failure-mode models can predict different types of failure for the same asset, each with its own probability score.

Here is a practical example of how Predict models are deployed in production. A chemical plant's compressor has sensors monitoring vibration, discharge temperature, suction pressure, and seal system condition. The Predict model ingests this data daily and calculates a failure probability for each of the compressor's known failure modes: bearing degradation, seal failure, and valve damage. When the bearing degradation probability exceeds 65%, the system automatically creates a work order for inspection. The maintenance team reviews the recommendation, schedules the inspection during the next planned outage, and replaces the bearing before it fails.

The key to successful Predict deployments is data quality and volume. Models need at least 30 historical failure examples to produce reliable predictions. The failure codes must be applied consistently -- if the same type of failure is sometimes coded as "BEARING" and sometimes as "MECHANICAL," the model cannot learn the pattern. Organizations that invest in failure coding discipline before deploying Predict see significantly better results.

Maximo Health and Condition Insight

MAS 9.2 introduced Maximo Condition Insight, a new capability that brings together work orders, inspections, meter readings, and reliability strategies to identify patterns in asset behavior and recommend what to do next. Condition Insight is different from Predict in an important way: where Predict focuses on statistical failure probability, Condition Insight focuses on operational condition assessment.

Condition Insight analyzes the full range of data available for each asset -- not just sensor readings, but also inspection findings, work order history, meter trends, and reliability strategy inputs. It identifies assets that are showing signs of degradation even if those signs do not yet meet the threshold for a Predict alert. This gives reliability engineers a broader view of asset health and helps them prioritize attention across their asset portfolio.

Work Order Intelligence

Another AI capability that has flown somewhat under the radar is Work Order Intelligence, which automatically suggests problem codes based on work order descriptions. When a technician enters a work order description like "Pump making grinding noise on startup, vibration high," the AI analyzes the text and suggests the most likely failure code. Users report 91% accuracy in these suggestions, which dramatically improves the consistency of failure coding across the organization.

Consistent failure coding is the foundation for everything else in predictive maintenance. Without it, you cannot identify which failure modes are most common, which assets are bad actors, or whether your maintenance strategies are effective. Work Order Intelligence solves this problem at the source -- when the work order is created -- rather than relying on after-the-fact data cleanup.

Visual Inspection: AI at the Edge

Maximo Visual Inspection brings computer vision capabilities directly to the field. Technicians can use their mobile devices to capture images of assets during inspections, and the AI analyzes those images in real-time to detect anomalies, read gauges, identify corrosion, and verify equipment condition.

The key technical achievement of Visual Inspection is that inference runs locally on the device. This means inspections work even in areas with no network connectivity -- a critical requirement for field operations in remote locations. The models are trained on labeled images and can be customized for specific asset types and inspection requirements.

A typical use case is visual inspection of pressure vessel nameplates. A technician photographs the nameplate, and the AI extracts the serial number, manufacturer, date of manufacture, and pressure rating. This data is automatically populated into the asset record, eliminating manual data entry and the associated errors. Another common use case is corrosion assessment: the AI analyzes images of pipe surfaces and quantifies the extent of corrosion, providing objective measurements that support condition-based maintenance decisions.

Visual Inspection is particularly valuable for organizations that conduct large-scale inspection campaigns. Instead of sending engineers to review thousands of images manually, the AI pre-screens the images and flags only those that show anomalies requiring human review. This can reduce inspection analysis time by 80% or more while improving consistency and accuracy.

Agentic Workflows: AI That Acts

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 work like this: the system detects that a critical pump's vibration levels have been trending upward for three consecutive days. An agent evaluates the situation against predefined rules: is the vibration within acceptable limits, is there a planned maintenance window in the next 7 days, are replacement parts available in inventory. If all conditions are met, the agent creates a work order, assigns it to the appropriate craft, schedules it for the next available maintenance window, and notifies the maintenance supervisor. The human reviews and approves the work order, but the agent has done all the preparation work.

The agentic framework is built on the MCP Server (Model Context Protocol Server) introduced in MAS 9.2. The MCP Server enables organizations to bring their own agents and integrate them directly with Maximo Manage APIs. This means you are not limited to IBM's predefined agents -- you can build custom agents that implement your specific business rules and decision logic.

The practical implications are significant. Agentic workflows can handle routine decisions that currently consume planner and scheduler time: creating work orders from Predict alerts, adjusting PM schedules based on asset condition, optimizing technician assignments based on skill and location, and managing inventory reservations for planned work. Each of these is a small task, but collectively they represent a substantial portion of the administrative overhead in maintenance operations.

Building Your First Agent

Organizations looking to start with agentic workflows should begin with a simple, well-defined use case. The best candidates are decisions that are currently made manually but follow clear, consistent rules. For example: "When a Predict alert exceeds 70% probability and the asset is not already under a work order, create a corrective work order with priority 2, assign it to the rotating equipment crew, and notify the reliability engineer."

The agent logic can be implemented as an automation script that runs on a schedule or in response to events. Here is a simplified example:

# Agentic workflow: Auto-create work order from Predict alert
from psdi.mbo import MboConstants
from psdi.server import MXServer
from java.util import Date

def evaluate_predict_alerts():
    # Get all Predict alerts above threshold
    alert_set = MXServer.getMXServer().getMboSet("PREDICTALERT", user)
    alert_set.setWhere("probability > 70 and status = 'NEW'")
    
    for i in range(alert_set.count()):
        alert = alert_set.getMbo(i)
        asset = alert.getMbo("ASSET")
        
        # Check if work order already exists for this asset
        wo_set = MXServer.getMXServer().getMboSet("WORKORDER", user)
        wo_set.setWhere(
            f"assetnum='{asset.getString('ASSETNUM')}' and "
            f"status in ('WAPPR','APPR','INPRG')"
        )
        
        if wo_set.count() == 0:
            # Create new work order
            wo = wo_set.add()
            wo.setValue("DESCRIPTION", 
                f"Predict Alert: {alert.getString('FAILURECODE')} - {asset.getString('DESCRIPTION')}")
            wo.setValue("ASSETNUM", asset.getString("ASSETNUM"))
            wo.setValue("WORKTYPE", "CM")
            wo.setValue("STATUS", "WAPPR")
            wo.setValue("WOPRIORITY", 2)
            wo.setValue("TARGETSTARTDATE", Date())
            wo.save()
            
            # Update alert status
            alert.setValue("STATUS", "WO_CREATED")
            alert.setValue("REFERENCENUM", wo.getString("WONUM"))
        
        wo_set.close()
    
    alert_set.close()

This is a simple example, but it illustrates the pattern. The agent evaluates conditions, makes a decision, and takes action. As you gain confidence, you can add more sophisticated logic: checking parts availability, considering technician schedules, evaluating cost-benefit tradeoffs, and coordinating with production schedules.

The AI Roadmap: What Comes Next

Looking ahead, the trajectory of AI in Maximo is clear. The capabilities introduced in MAS 9.2 are foundation layers for a more autonomous future. Based on IBM's announced direction and the patterns emerging from early adopters, several developments are likely in the next 12 to 24 months.

Maximo Assistant will evolve from a query tool to a proactive advisor. Instead of waiting for users to ask questions, the assistant will surface insights automatically: "Three critical work orders need attention," "Asset MTBF has declined 15% this quarter," "Recommended PM frequency change for pump group P-100 based on condition data." This shift from reactive to proactive AI will change how users interact with the system.

Agentic workflows will expand from simple rule-based automation to AI-driven decision making. Instead of following predefined rules, agents will use machine learning to optimize decisions based on outcomes. An agent might learn that certain work order types are more cost-effective when scheduled on specific days, or that certain technician assignments produce better quality results.

Mobile AI will become more capable. Maximo Assistant on Mobile, introduced in MAS 9.2, will expand to support voice-based queries and hands-free operation. Technicians in the field will be able to ask questions, capture inspection data, and receive guidance without stopping their work to type.

Multi-modal AI will combine text, image, and sensor data for richer insights. A technician inspecting a pump could photograph a leaking seal, and the AI would analyze the image, check the pump's vibration history, review recent work orders, and suggest the most likely root cause and recommended repair procedure.

Practical Implications

Deploying AI in Maximo requires more than just enabling features. Organizations need to prepare their data, train their teams, and establish governance for AI-driven decisions.

Data readiness is the most common barrier to AI success. Predictive models need clean, consistent historical data. Maximo Assistant needs well-structured object structures and clear field descriptions. Visual Inspection needs labeled training images. Before deploying any AI capability, audit your data quality and address the gaps. A practical starting point is to review your failure code usage: are codes applied consistently, are there too many codes, are there gaps in the code hierarchy?

Team readiness is equally important. AI changes the role of maintenance planners, reliability engineers, and technicians. Planners shift from creating work orders to reviewing and approving AI-generated recommendations. Reliability engineers shift from data analysis to model validation and exception handling. Technicians shift from manual data entry to AI-assisted inspection. These role changes require training and change management support.

Governance is essential for agentic workflows. Define clear boundaries for autonomous action: what decisions can agents make without human approval, what requires review, and what is always human-decided. Establish monitoring and audit trails so you can review agent decisions and improve the logic over time. Start with conservative boundaries and expand as you build confidence in the system.

The Bottom Line

AI in Maximo has moved from potential to production. Maximo Assistant provides natural language access to asset data. Predict and Health deliver failure predictions that reduce unplanned downtime. Visual Inspection brings computer vision to the field. And agentic workflows automate routine decisions, freeing skilled workers for higher-value activities.

The organizations that will benefit most are those that start now, start small, and build capability over time. Pick one use case -- a single asset class, a single failure mode, a single agent workflow -- prove the value, learn from the experience, and expand. The technology is ready. The question is whether your organization is ready to embrace it.