Agentic AI in Maximo: How MAS 9.2 Transforms the Assistant from Chatbot to Coworker

The Maximo Assistant has been around since MAS 8.x, but if you used it back then, you probably remember a capable but limited conversational tool. It could answer questions about work orders, look up asset records, and help you find service requests. Useful, sure, but it was fundamentally a query interface wrapped in natural language. You asked, it retrieved, and that was the end of the interaction. MAS 9.2 changes that relationship entirely. The Assistant is no longer a chatbot that fetches data. It is an agentic system that can plan, coordinate, and execute multistep tasks on your behalf. IBM calls this the most significant evolution of the Maximo Assistant since its launch, and after looking at what is actually shipping in 9.2, that claim holds up.

The shift matters because the daily reality of working with Maximo involves far more than looking up records. Technicians review work orders, update statuses, report issues, and create follow-up tasks. Supervisors prioritize requests, manage backlog, and handle escalations. Planners explore scheduling scenarios and adjust assignments. Each of these workflows involves multiple steps, decisions, and system interactions. The agentic Assistant in MAS 9.2 is designed to take on that orchestration burden. Instead of manually navigating each step, you give it a high-level request and it handles the smaller, repetitive decisions and actions required to achieve the goal. This article takes a detailed look at what agentic workflows mean in practice, how the architecture works, what tools the Assistant has access to, and what your organization should be thinking about as you plan adoption.

The Supervisor Agent: How Agentic Orchestration Works

At the core of the MAS 9.2 Assistant is what IBM calls the Supervisor ALM agent. Think of it as the brain behind the operation. When you send a message to the Assistant, the Supervisor agent receives the request, understands your intent, breaks it down into logical steps, and then coordinates the tools available to the Assistant to complete those steps. This is a fundamentally different architecture from the previous retrieval-only model, where the Assistant translated your question into an OSLC query and returned the result.

The Supervisor agent can reason about what you are asking. If you say "What assets need my attention?" it does not just run a query for assets with high failure probability. It combines data from Condition Insight, work order history, alert status, and asset criticality to produce a prioritized answer. If you ask "How is asset 1010 doing?" it pulls health scores, recent work orders, meter readings, inspection results, and any active alerts, then synthesizes them into a plain-language summary. If you ask "How do I create a PM schedule in Maximo Manage?" it pulls step-by-step instructions from IBM Documentation and presents them in context.

The orchestration layer is what makes this agentic rather than conversational. The Assistant can coordinate multiple tools in a single response. For example, if you ask it to create a follow-up work order for a specific asset, it needs to retrieve the asset record, check the work order history, identify the appropriate job plan, populate the work order fields, and create the record. The Supervisor agent handles all of those steps. You do not need to specify each one. You give the intent, and the agent figures out the execution path.

This architecture also means the Assistant can handle multistep workflows that span different parts of the system. A supervisor might ask the Assistant to review all high-priority work orders, identify which ones are at risk of missing their deadline, and create escalation service requests for the at-risk items. In the previous model, that would require manual work order review, filtering, analysis, and then individual service request creation. With agentic workflows, the Assistant can handle the entire sequence. It is worth noting that the Assistant operates within configured boundaries. Administrators control which tools and object structures the Assistant can access, and the system maintains audit trails of actions taken. This is not an AI running unchecked through your Maximo environment.

The Tool Ecosystem: What the Assistant Can Actually Do

The agentic capabilities of the Assistant depend on the tools it has access to. In MAS 9.2, IBM ships a growing set of enterprise-ready tools designed specifically for asset and work management workflows. Understanding what these tools do is essential for planning how to roll out the Assistant in your organization.

General knowledge search gives users instant answers to asset lifecycle management questions without leaving Maximo. If a technician needs to understand the difference between a corrective work order and a breakdown work order, they can ask the Assistant directly. If a reliability engineer wants to know what FMEA means in the context of Maximo APM, the Assistant provides the answer. This reduces context switching and reliance on external search tools.

IBM product guidance delivers official Maximo best practices and troubleshooting advice taken directly from IBM Documentation. Instead of navigating to the documentation portal and searching through pages of content, users get step-by-step instructions in the Assistant interface. This is particularly valuable for new users who are still learning the platform, but it also helps experienced users find configuration steps they might not perform frequently.

Data retrieval is the tool that democratizes access to Maximo data. Users no longer need to understand the schema, know the relationships between objects, or build OSLC queries from scratch. They can ask conversational questions and the Assistant translates them into the appropriate queries. In MAS 9.2, the default configuration provides access to work orders, assets, service requests, meters, preventive maintenance, job plans, work logs, assignments, and persons. Administrators can extend this to any object structure, including custom objects, by selecting the "Use for assistant" checkbox in the Object Structures application.

Condition insights deliver clear, contextual summaries of asset health in plain language. This tool goes beyond reporting what happened to explain why it happened and what to do next. When a user asks about an asset, the Assistant can surface underlying causes and trends, drawing on the same Condition Insight engine that powers the APM module. This means a technician in the field can get the same quality of asset intelligence that a reliability engineer sees on the desktop dashboard.

AI-assisted actions are where the Assistant moves from answering questions to taking action. Users can create, update, and modify records through natural language commands. The Assistant can change work order status, create follow-up work orders, and create service requests. These are the everyday operational actions that consume significant time when performed manually, especially across multiple records. The Assistant handles the data entry and multistep workflow, letting the user focus on the decision rather than the mechanics.

Custom tools via automation scripts extend the Assistant capabilities to organization-specific workflows. Any automation script in Maximo can be registered as a custom tool for the Assistant. For example, you might have a script that looks up all breakdown codes supported by a particular asset type. By registering that script as a tool, a technician can ask the Assistant to run it when reporting a downtime event, without navigating to the script execution interface.

Maximo Assistant on Mobile: Field-First AI

The mobile experience of the Maximo Assistant in MAS 9.2 deserves specific attention because it is where agentic workflows have the most immediate operational impact. Technicians in the field often work in challenging conditions: gloved hands, poor lighting, time pressure, and limited access to desktop tools. The ability to interact with Maximo using natural language on a mobile device transforms how they access information and complete work.

In MAS 9.2, Maximo Assistant is available on Maximo Mobile when operating in connected mode. The mobile Assistant is best suited for finding and exploring asset and work order data, including asset insights, and searching IBM Documentation. While the desktop Assistant can be configured to retrieve data for all objects, the mobile experience is more focused, reflecting the different needs of field workers versus office users.

A technician arriving at a job site can ask the Assistant for a summary of the asset they are about to work on. The Assistant pulls the asset record, recent work order history, inspection results, any active alerts, and condition insights, then presents them in a format designed for mobile consumption. The technician does not need to navigate through the mobile application menus, apply filters, or switch between asset and work order views. They ask, and the information is there.

The mobile Assistant also integrates with Maximo Visual Inspection (MVI), which runs AI models locally on the device for photo-based defect detection. A technician can take a photo of an asset component, and the MVI model identifies potential defects. The Assistant can then help the technician create a work order or inspection note based on the detected issue. This integration between visual inspection and the agentic Assistant creates a workflow that would require multiple manual steps without AI.

For organizations rolling out Maximo Mobile, the Assistant should be a key part of the training and adoption strategy. Field technicians who are new to Maximo often struggle with the complexity of the mobile interface. The Assistant reduces that complexity by letting them interact conversationally. Instead of learning where every field and button is, they can ask for what they need. This does not eliminate the need for training, but it does lower the barrier to productive use of the system, especially for new or infrequent users.

The MCP Server: Extending AI Beyond Out-of-the-Box

One of the most architecturally significant additions in MAS 9.2 is the Model Context Protocol (MCP) Server. IBM describes it as the "USB-C standard for AI communication across systems," which is a useful analogy. The MCP Server enables external AI agents to connect to Maximo Manage APIs in a governed, trusted way. This means organizations are not limited to the tools that IBM ships with the Assistant. They can bring their own AI agents and integrate them directly with Maximo.

The MCP Server opens possibilities that go well beyond what the out-of-the-box Assistant can do. An organization might have a custom AI agent that monitors supply chain data and predicts material shortages. Through the MCP Server, that agent can query Maximo for inventory levels, check job plans for upcoming material requirements, and create purchase requests when shortages are predicted. Another organization might have an AI agent that analyzes energy consumption patterns and correlates them with asset performance. Through MCP, that agent can pull asset performance data from Maximo and push maintenance recommendations back.

The MCP Server also enables the use of automation scripts as custom tools for the Assistant. This is a significant capability for organizations that have invested in custom automation. Instead of running scripts manually or through scheduled jobs, users can invoke them through the Assistant using natural language. The Assistant handles the interaction, and the script handles the execution. This bridges the gap between conversational AI and programmatic automation.

It is important to understand the deployment options for the AI Service that powers the Assistant and MCP Server. MAS 9.2 offers three deployment patterns. Full SaaS uses existing AppPoints with no new purchase required. Hybrid AI SaaS requires SaaS AppPoints but no new infrastructure. Full Customer Managed includes a watsonx.ai license for use with the AI Service only. The choice of deployment pattern affects how much control the organization has over the AI models, data residency, and integration architecture. Organizations with strict data sovereignty requirements will likely lean toward the Customer Managed option, while those already running MAS in SaaS mode can use the Full SaaS pattern for the simplest path to adoption.

Planning Your AI Adoption: A Phased Approach

Organizations looking to adopt the agentic Assistant in MAS 9.2 should take a phased approach rather than attempting a big-bang rollout. The recommended sequence begins with auditing the data foundation, because the Assistant is only as good as the data it can access. Look at the asset hierarchy, meter history, work order history, and failure records. Identify the asset classes where the data is complete and consistent enough to support AI interactions. 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.

Start by enabling the Assistant for a pilot group of users. Choose a team that is comfortable with new technology and has workflows that benefit from the out-of-the-box tools. A maintenance supervisor who spends significant time reviewing work orders and creating follow-up tasks is a good candidate. A reliability engineer who needs to query asset data across multiple applications is another. Roll out the data retrieval and IBM product guidance tools first, because they provide immediate value without requiring any system configuration changes. Let the pilot group use the Assistant for two to four weeks and gather feedback on what works and what does not.

Next, enable the AI-assisted actions for the pilot group. This is where governance becomes important. Decide which actions the Assistant is allowed to take. Creating a follow-up work order is low risk. Changing the status of a approved work order might require additional controls. Configure the object structures and tools that the Assistant can access based on the specific workflows of the pilot group. Do not enable everything at once. Start with the actions that have the clearest productivity benefit and the lowest risk of unintended consequences.

After the pilot, extend access to the broader user base. The mobile Assistant should be part of this second wave, because field technicians benefit from the conversational interface. Provide role-based training that focuses on the specific tasks each user group will perform with the Assistant. Technicians need to know how to ask for asset information and how to use the Assistant to complete work order updates. Supervisors need to know how to ask for prioritized work order lists and how to create escalation requests. Planners need to know how to use conversational scheduling and what-if analysis.

Finally, explore the MCP Server for custom integrations. This is an advanced capability that requires development resources, but it unlocks the most transformative use cases. Organizations with in-house AI expertise can build custom agents that address specific operational challenges. Organizations without that expertise can work with IBM partners who are developing MCP-based solutions. The MCP Server is the foundation for the next generation of Maximo automation, and organizations that start exploring it now will be better positioned as the ecosystem of MCP-compatible agents grows.

Practical Implications

The agentic Assistant in MAS 9.2 changes the calculation for organizations deciding when to upgrade. If you are on MAS 9.1, the Assistant is useful but limited to retrieval. The agentic workflows in 9.2 represent a step-function improvement in what users can accomplish through natural language. For organizations with large technician populations, the mobile Assistant alone could justify the upgrade by reducing time spent on data entry and record lookup.

Governance is the key consideration. The Assistant can take actions on behalf of users, which means administrators need to think carefully about which actions are enabled and for whom. The audit trail captures all Assistant-initiated actions, but organizations should establish review processes for those actions, especially in the early stages of adoption. Start with read-only tools and gradually enable write actions as confidence in the system grows.

Training requirements are lower than you might expect for the Assistant itself, because the natural language interface is intuitive. The real training need is helping users understand what the Assistant can and cannot do. Users who expect a general-purpose AI will be disappointed when the Assistant cannot answer questions outside its configured scope. Users who understand that the Assistant is a Maximo-specific tool with defined capabilities will get more value from it.

Bottom Line

MAS 9.2 transforms the Maximo Assistant from a conversational query tool into an agentic system that can plan, coordinate, and execute multistep tasks. The Supervisor agent architecture, the growing tool ecosystem, the mobile experience, and the MCP Server for custom integrations collectively represent the most significant AI advancement in the Maximo platform's history. Organizations should approach adoption in phases: audit data quality, pilot with a focused user group, enable tools incrementally, extend to the broader user base, and explore MCP for custom use cases. The agentic Assistant is not a finished product. It is a platform that will grow as IBM and the community add tools, agents, and integrations. Organizations that start building familiarity now will be better positioned to take advantage of that growth.