Maximo Condition Insight, Maximo AI Service, and the watsonx Orchestrate Pattern: A 2026 Practitioner Map

The AI story in Maximo in 2026 is no longer a single feature. It is a layered architecture with three distinct surfaces, each with a different purpose, a different deployment model, and a different cost. If you are trying to figure out "what does IBM mean by AI in Maximo this quarter," the answer is that they mean three different things, and you need to know which is which.

This article maps the three layers as they actually exist in the May 2026 release train, with a practitioner read on when to use each.

The three layers

1. Maximo AI Service — the inference layer. This is the watsonx.ai integration that exposes named templates for specific tasks: recommended problem codes, recommended field values, similar record lookup, AI-assisted FMEA generation, and the Maximo Assistant configuration.
2. Maximo AI Assistant — the conversational surface. The chat-style interface in the Manage UI, backed by NL2OSLC, that lets a user ask "show me all open work orders on the north feeder" in plain language.
3. Agentic workflows — the orchestration layer. This is what Maximo Condition Insight is built on, and what the Maximo IT + watsonx Orchestrate integration is built on. Agents that read, reason, and act across multiple tools and data sources.

The naming and packaging have shifted in the last 12 months. Notably, the AI broker that was introduced in MAS 9.0 was replaced by the Maximo AI Service as of August 1, 2025. Any customer who deployed the AI broker needs to be on the AI Service by now. The AI Service 9.1.x is the current line; AI Service 9.1.15 shipped on May 27, 2026.

A second packaging note: Maximo AI Service 9.1 includes a limited-use license to watsonx.ai and incurs an additional AppPoint cost. This is the most common point of confusion in licensing conversations. AI Service is not "included with Manage" in MAS 9.1. It is an add-on.

Layer 1: Maximo AI Service — the inference layer

The AI Service exposes a small set of named templates. Each template is a pre-configured inference pipeline: input contract, prompt, model selection, output contract. You don't have to write a prompt or select a model. You enable the template, point it at the data it needs, and the AI Service does the inference.

The current templates, as documented in the watsonx.ai integration guide:

To enable any of these, the basic steps are:

1. Deploy Maximo AI Service 9.1.x in your MAS environment. (Customer-managed: install from the IBM Operator Catalog. SaaS: it is provisioned.)
2. Configure a watsonx.ai project and connect the AI Service to it.
3. Enable the template(s) you want.
4. For field-value recommendations, set up the launch point (typically an attribute launch point on a work order field).
5. Verify on a few test records before going live.

The honest read: the templates are well-engineered and they work, but the data they consume is the data you have. If your failure codes are inconsistent, the PCC template's recommendations will be inconsistent. If your asset hierarchy is messy, similarity returns messy results. The AI Service amplifies the quality of your data, good or bad.

Layer 2: Maximo AI Assistant — the conversational surface

The Maximo AI Assistant is the chat interface embedded in the Manage UI. The user types a question in natural language. The Assistant interprets it, generates an OSLC query, runs it, and returns the results in a workspace that the user can filter, sort, and refine.

The capability set as of Manage 9.1:

• Database Retrieval — "show me all work orders on asset X" returns a list of work orders, with the columns the user actually needs.
• Analytical Insights — "how many corrective work orders did we have on the north feeder in Q1 2026?" returns a number (or a small chart).
• Interactive Results — the returned workspace is interactive. The user can filter, group, save, or pivot, just like a normal Maximo result set.

The current architecture (as of Manage 9.1) is that the Assistant is NL2OSLC — the natural-language input is converted to an OSLC query, which is then executed against the Manage business objects. The result is constrained by the user's security group, so the Assistant cannot be used to exfiltrate data the user wouldn't otherwise see.

A 2026 reality check: the Assistant is genuinely useful for ad-hoc data access. It removes the "I need to ask a power user to write a query for me" friction. It is not yet a replacement for the structured reports and dashboards that the business already relies on, and it isn't positioned to be. The Assistant is the quick, conversational layer. The reports and dashboards are the curated, governed layer. They complement each other.

Layer 3: agentic workflows

This is the layer that is new in the last six months and that the IBM announcements are loudest about. Two production patterns exist today:

Maximo Condition Insight

Launched December 5, 2025, Maximo Condition Insight is an agentic AI capability within Maximo APM. It is positioned as the AI for condition-based maintenance (CBM). The four advantages IBM publishes:

1. Instant insights. Condition Insight evaluates work orders, metrics, time-series data, meter readings, FMEA records, and alerts, and produces a plain-language summary of the asset's condition. No data modeling, no integration project.
2. Strategic alignment. The condition summary is mapped back to the failure modes in the FMEA. The output is not just "the asset is in poor condition" — it is "the asset is showing early signs of failure mode X, and the recommended action is the condition task Y from the FMEA."
3. Conversational experience. Condition Insight plugs into the Maximo AI Assistant. A user can ask "what's the current condition of asset Z?" and get a plain-language answer that references the FMEA, the work orders, and the metrics.
4. Automated execution. Coming soon, but already in design partner deployments: the agent will create or update work orders automatically, following the prescribed maintenance strategy. A user can review and approve, or set it to run unattended.

The 2026 practical read: Condition Insight is the feature that turns the "AI in Maximo" conversation from "AI helps me find records" to "AI helps me decide what to do." For reliability engineers, this is the most consequential new AI capability in Maximo since the original Predict.

Maximo IT + watsonx Orchestrate

The Maximo IT + watsonx Orchestrate integration is the production reference for the agentic pattern in Maximo. The integration uses automation scripts delivered with Maximo IT (WXOAISUMMARY, GENAISOLRECOMMENDATIONS) to call watsonx Orchestrate agents. Three use cases ship today:

1. AI Incident Summarization. The agent continuously summarizes open Maximo IT incidents. Every time the ticket is updated, the agent regenerates the summary. The summary appears in the Service Delivery application, so an agent or manager gets a structured overview (impact, symptom, root cause, resolution, preventative actions) without reading the work log.
2. AI Knowledge Creation. When an incident is resolved, the agent drafts a structured solution record (cause, symptom, resolution) from the ticket history. The agent or knowledge manager reviews and publishes. This converts operational work into reusable knowledge.
3. AI Agent Chat. Users can chat with the agent to understand incident status, do root cause analysis, or get the latest summary.

The Maximo IT + Orchestrate pattern is a template for the next wave of agentic workflows in Maximo. The shape is:

1. Identify a knowledge-work task that humans do repeatedly.
2. Build an agent in watsonx Orchestrate that has the right tools and the right prompt.
3. Connect Maximo to the agent via a delivered automation script.
4. The agent runs continuously, producing structured outputs that land back in Maximo as records.
5. A human reviews and approves, except in cases where the business has explicitly accepted the agent's authority.

This pattern is replicable in any Maximo application. The same architecture that summarizes IT incidents can summarize work orders, draft PMs from FMEA recommendations, or pre-populate service requests from emails.

The MAS 9.2 horizon

IBM has telegraphed three big shifts for MAS 9.2, all on the agentic side:

• The Maximo Assistant evolves from one-off Q&A to a persistent, context-aware digital teammate that can reason across the user's data and documents.
• Extensible agentic workflows become a first-class pattern. IBM, customers, and the partner ecosystem can build and ship agents that coordinate across tools and systems.
• An explosion of use cases across every part of the asset lifecycle. IBM publishes a partial list: asset condition insights, intelligent CBM, lease abstraction, technician work assistance, WO generation, natural language formula, risk assessment advisor, admin assistance, asset inspection analysis.

The 9.2 agentic layer is the largest expansion of the AI surface since the original Maximo AI broker was introduced. The MAS 9.2 beta is in customer hands in early 2026, with general availability expected in the second half of the year.

A 2026 deployment pattern

For a team getting started with AI in Maximo in 2026, the recommended sequence is:

1. Deploy the Maximo AI Service 9.1.x. This is foundational. Without it, none of the other layers work.
2. Enable the PCC and MCC templates first. These are the highest-leverage, lowest-risk templates. They improve technician productivity on every work order they touch.
3. Roll out the Maximo AI Assistant to a pilot group. Start with the power users (planners, supervisors). Get their feedback on what works and what doesn't.
4. Build the Reliability Strategies FMEA library with the sfmea template. This is the most data-intensive template, but it pays back the fastest if you have reliability engineers who can curate the output.
5. Pilot Condition Insight on one asset class. Prove the pattern. Expand.
6. Build your first Orchestrate-pattern agent for a specific knowledge-work task. The Maximo IT templates are the model.

The key discipline: each layer is a real piece of work, with real data dependencies, and real change management. The temptation to skip steps and go straight to "we want agentic workflows" is the same temptation that leads to a stalled AI program. The layers compose. The data and the trust have to be built up from the bottom.

The honest limits

Three things to be clear about:

• The AI is not a replacement for the reliability engineer, the planner, or the technician. It is an accelerator. The reliability engineer still owns the FMEA. The planner still owns the schedule. The technician still owns the work.
• The data quality story is the same as it has always been. Bad data in, bad AI out. The AI Service exposes the data quality issues that were always there.
• The AppPoint cost is real. AI Service is an add-on. The model selection (Granite, third-party, or your own fine-tune) affects both performance and cost. Plan for it.

The Maximo AI story in 2026 is a serious one. The tools are real, the integration is direct, and the roadmap is visible. The work is the work: data quality, organizational alignment, and the discipline to use the AI well.