AI in Maximo 9.2: Agentic Workflows, MCP Servers, and the End of Isolated AI Use Cases

The AI story in Maximo has been building for years. Maximo Assistant arrived as a natural language interface. Maximo Visual Inspection brought computer vision to asset inspection. Maximo Predict added failure forecasting. Health introduced scoring and risk indicators. Each was valuable. Each was also, to some degree, isolated. A technician could ask the Assistant a question. An inspector could use MVI to check for defects. A reliability engineer could review Predict forecasts. But these capabilities did not talk to each other.

MAS 9.2 changes that. The introduction of agentic workflows, the MCP Server, Condition Insight, and conversational scheduling connects AI capabilities that were previously separate. An anomaly detected by Monitor can trigger Condition Insight analysis, which can generate a Smart Alert, which can create a work order, which can be scheduled by conversational Optimizer, which can be executed by a technician using Maximo Assistant on Mobile. This is not a collection of AI features. It is an AI fabric that spans the entire asset management lifecycle.

This article examines the AI architecture in MAS 9.2, the patterns that connect AI capabilities into workflows, and what practitioners need to know to deploy AI effectively in their Maximo environments. It is written for technical leads and architects who need to understand not just what the AI can do, but how it works, how to configure it, and where the boundaries are.

The AI Architecture: From Features to Fabric

The MAS 9.2 AI architecture has four layers:

Layer 1: AI Service. This is the foundation. The AI Service provides the model hosting, inference, and management capabilities that all AI features use. In 9.2, the AI Service ships with the June Feature Channel (Reference #7276603 for 9.2, #7276604 for 9.1.16, #7270126 for April Feature Channel). It supports multiple model types: classification, regression, anomaly detection, natural language, and computer vision.

Layer 2: AI Capabilities. These are the specific AI features built on the AI Service: Condition Insight, Smart Alerts, RCM Advisor, Maximo Assistant, MVI, Predict, and the conversational interfaces for scheduling and what-if analysis. Each capability uses one or more AI models from the AI Service.

Layer 3: Agentic Workflows. This is the new layer in 9.2. Agentic workflows connect AI capabilities into sequences that can guide decisions and move work forward. An agentic workflow might: detect an anomaly (Monitor), analyze it against asset history (Condition Insight), generate a prioritized alert (Smart Alerts), create a work order (Work Order Automation), and schedule it (conversational Optimizer). The workflow can operate with varying degrees of autonomy, from fully advisory (AI recommends, human decides) to semi-autonomous (AI acts within defined guardrails).

Layer 4: MCP Server. This is the external interface. The MCP Server exposes Maximo Manage APIs to external AI agents through the Model Context Protocol. An AI agent running in Claude, IBM Bob, or a custom framework can interact with Maximo through a standardized protocol. This means AI capabilities are not limited to what IBM ships. Organizations can build their own AI agents that interact with Maximo through MCP.

Agentic Workflows: What They Are and How They Work

The term "agentic" is everywhere in 2026, and it is worth defining precisely in the Maximo context. An agentic workflow in MAS 9.2 is a sequence of AI-driven steps that:

1. Observes: Monitors data sources (sensor data, work orders, inspections, alerts) for conditions that require action
2. Analyzes: Applies AI models to understand the condition in context (asset history, failure modes, reliability strategy)
3. Recommends: Proposes actions based on the analysis (create work order, adjust schedule, escalate alert)
4. Acts: Executes the recommended action, either automatically (within defined guardrails) or after human approval
5. Learns: Incorporates the outcome of the action into future recommendations

The key distinction from traditional automation: agentic workflows adapt. A traditional automation rule says "if vibration exceeds 7.0 mm/s, create a work order." An agentic workflow says "vibration is at 8.2 mm/s, the pattern matches bearing degradation, similar pattern preceded failure on P-102 by 14 days, spare bearing is available in Storeroom A-3, recommended action is bearing replacement within 7 days, and I have created a draft work order for your review."

The agentic workflow does not just follow rules. It reasons about the situation using multiple AI models and multiple data sources. It presents a recommendation with supporting evidence. It learns from whether the recommendation was accepted or rejected.

The IBM announcement describes it as "agentic workflows designed to help guide decisions and move work forward across these processes in practical, operationally-grounded ways." The emphasis on "practical" and "operationally-grounded" is important. These are not experimental AI capabilities. They are designed for production operations where the cost of a bad recommendation is measured in downtime, safety incidents, and lost revenue.

The MCP Server: Maximo as an AI-Native Platform

The MCP Server in MAS 9.2 is the most strategically significant AI capability because it changes who can build AI for Maximo. Before MCP, building an AI integration with Maximo required understanding the Maximo REST API, handling authentication, managing error responses, and writing custom integration code. With MCP, an AI agent can discover Maximo capabilities, understand parameters, and execute operations through a standardized protocol.

The MCP ecosystem has grown explosively. By March 2026, MCP had 97 million monthly SDK downloads, 81,000 GitHub stars, and support from every major AI vendor: Anthropic, OpenAI, Google, Microsoft, and AWS. Over 18,000 MCP servers exist, covering databases, CRMs, Slack, GitHub, Snowflake, and now Maximo.

The practical implications for Maximo AI:

Build your own AI agents. Organizations can build AI agents that interact with Maximo through MCP. A maintenance planner agent could review the work backlog, identify scheduling conflicts, and propose optimized schedules. A reliability agent could monitor asset health scores, identify deteriorating assets, and recommend interventions. These agents are not built by IBM. They are built by organizations using their own AI models and their own domain expertise.

Connect Maximo to the broader AI ecosystem. An AI agent that interacts with Maximo through MCP can also interact with other MCP-compatible services. A single agent could query Maximo for asset data, query a weather service for forecast data, and recommend maintenance scheduling that accounts for both asset condition and weather constraints.

Governed AI access. The MCP Server operates within Maximo's security model. An AI agent can only access the data and perform the operations that its associated user's security groups permit. This is critical for enterprise adoption.

Sankar Ganesh V.S. demonstrated a practical MCP pattern in June 2026: using the MXProfile object to retrieve user context (default site, org, authorized sites) and pass it to an AI agent. The agent uses this context to scope its queries, returning site-specific or org-specific data without hardcoding parameters. This is the kind of pattern that becomes standard with MCP.

IBM Bob and Maximo: AI-Assisted Development

IBM Bob is not a Maximo-specific tool, but its Maximo code optimization skill (demonstrated in June 2026) is directly relevant to Maximo AI. Bob can analyze existing automation scripts, identify improvement opportunities, and suggest optimizations. The five-stage process (Discover, Analyze, Optimize, Validate, Update) is designed for governed, repeatable code improvement.

Bob's relevance to the broader AI story: it demonstrates the pattern of AI augmenting human expertise rather than replacing it. Bob does not automatically rewrite automation scripts. It analyzes them, suggests improvements, and lets the developer review and approve. The developer remains in control. The AI accelerates the work.

This pattern applies across all of Maximo's AI capabilities. Condition Insight recommends actions. The reliability engineer decides. Smart Alerts prioritize work. The planner adjusts. Work Order Automation creates draft work orders. The supervisor approves. The AI augments. The human decides. The system learns.

watsonx Integration: Enterprise AI for Maximo

The watsonx platform provides the enterprise AI foundation for Maximo's AI capabilities. The watsonx.data integration MCP Server, which reached 1.0.0 GA in June 2026, connects data integration workflows to AI agents. This is relevant for Maximo organizations that need to build data pipelines between Maximo and their AI infrastructure.

The watsonx MCP Server introduces a planning workflow: agents draft an intent document (sources, stages, transformations, targets) before building. This surfaces misunderstandings as reviewable sentences rather than failed jobs. The pattern is the same as Bob's governed optimization: AI proposes, human reviews, system executes.

For Maximo organizations using watsonx, the integration points include:

• Data pipelines: Extract Maximo data (work orders, asset history, sensor readings) into watsonx.data for AI model training
• Model deployment: Deploy trained models to Maximo AI Service for inference
• Agent orchestration: Use watsonx Orchestrate (updated June 14, 2026) to coordinate AI agents that interact with Maximo through MCP

Practical Implications

For organizations deploying AI in Maximo, the 9.2 release changes the conversation from "what AI features are available?" to "how do we connect AI capabilities into workflows that deliver measurable outcomes?"

The recommended approach:

1. Start with a specific workflow, not a platform. Identify one process where AI can make a measurable difference: reducing unplanned downtime on a critical asset, improving first-time fix rates for field technicians, reducing the time planners spend on scheduling. Deploy AI to improve that specific workflow. Measure the outcome. Expand from there.

2. Use the advisory-to-autonomous spectrum. Start with AI in advisory mode: it recommends, humans decide. As confidence grows, move toward semi-autonomous mode for low-risk decisions. Reserve fully autonomous mode for well-understood, low-consequence scenarios.

3. Invest in data quality. AI is only as good as the data it learns from. If your work order history is incomplete, your failure codes are inconsistent, or your asset registry is out of date, the AI will make bad recommendations. Data quality is an AI prerequisite, not an afterthought.

4. Build MCP literacy. The MCP Server makes Maximo accessible to AI agents. Your team needs to understand how to build, deploy, and govern AI agents that interact with Maximo through MCP. This is a new skill for most Maximo teams.

Bottom Line

MAS 9.2 is the release where Maximo's AI capabilities stop being a collection of features and start being a connected fabric. Agentic workflows, the MCP Server, and the integration with the broader AI ecosystem (watsonx, IBM Bob, MCP-compatible agents) mean that AI is no longer something you use in Maximo. It is something Maximo is. The organizations that understand this shift and build the skills to operate in an AI-native Maximo environment will have a compounding advantage over those that treat AI as a feature checklist.

Sources

• IBM MAS 9.2 Announcement (June 25, 2026): https://www.ibm.com/new/announcements/introducing-maximo-application-suite-9-2
• IBM Bob Maximo Code Optimization Skill Demo (June 2026): https://www.youtube.com/watch?v=bNID8QRi7Iw
• Sankar Ganesh V.S.: Maximo API Results with User Profile Data (June 2026): https://www.linkedin.com/posts/sankar-ganesh-v-s-9878aa23_ibm-community-activity-7473383369490264064-jRPn
• IBM watsonx.data Integration MCP Server 1.0.0 GA (June 2026): https://community.ibm.com/community/user/blogs/john-wen/2026/06/15/agentic-data-integration-is-here-with-the-watsonxd
• Jean Brant: Envisioning the Next Generation of Support with AI (June 2026): https://www.linkedin.com/posts/jean-brant_technology-lifecycle-services-envisioning-activity-7474784449956593664-sINE
• Govind HV: IBM Bob Optimizes Inference Economics (May 2026): https://www.linkedin.com/posts/govindhv_bob-for-enterprise-ai-architects-activity-7465737421633290240-8xEt
• IBM Maximo Application Suite Releases Information: https://www.ibm.com/support/pages/maximo-application-suite-releases-information-0