An AI assistant tells you what’s in Maximo. An AI agent changes what’s in Maximo. That’s the real shift behind all the agentic AI news this year, including the agentic workflows IBM introduced in Maximo Application Suite 9.2.

When AI only answers questions, a bad answer costs you a few minutes. When AI can create work orders, update statuses, or recommend who gets assigned, bad data turns into bad work. The question isn’t whether agents are coming to Maximo. It’s whether your Maximo is ready to be acted on.

Assistant vs. agent at a glance

  AI assistant AI agent
What it does Finds, summarizes, and explains Recommends and takes action
Example “Show me the work history on Pump 204” Creates a follow-up work order for Pump 204 and routes it for approval
Touches your data Reads it Reads it and writes to it
If the data is bad You get a wrong answer You get wrong work, at scale
Needs from you Searchable, reasonably clean records Clean records, clear rules, defined approvals, and the right security roles

What an agent leans on in your Maximo

An agent follows the same data and rules your people do. It just follows them faster and doesn’t stop to ask when something looks off.

  • Asset hierarchy: If assets sit in the wrong place, the agent picks the wrong location, the wrong parent, or the wrong crew.
  • Failure codes: If they’re missing or inconsistent, the agent has no reliable pattern to learn from or act on.
  • Work types and statuses: If they aren’t used consistently, the agent can’t tell urgent from routine.
  • Job plans: If a job plan is outdated, the agent spreads that outdated plan into every work order it builds.
  • Security groups and approvals: These define what the agent is allowed to do. If a human wouldn’t get that access, the agent shouldn’t either.

Good first jobs for an agent

Start where the volume is high, the risk is low, and a person still signs off.

  1. Summarizing long work order histories before a planner reviews them
  2. Triaging incoming service requests and suggesting a work type and priority
  3. Drafting follow-up work orders from completed inspections, held for approval
  4. Flagging aging or duplicate items in the backlog

Leave safety-critical work, permits, and anything that spends real money out of scope until you trust the results.

Five questions to ask before you turn it on

  • Which records will the agent read, and which will it change?
  • Who approves its actions, and how quickly?
  • How will you see what it did, and undo it if you have to?
  • Is the underlying data clean enough that you’d trust a new hire to act on it?
  • Who owns the agent’s results when something goes wrong?

If you can’t answer the fourth question with a confident yes, fix the data before you add the agent.

Data Quality · Data Governance · Work Order Data Quality · Asset Hierarchy · Failure Code

Want to know where you stand? Take Maven’s free AI Readiness Assessment. No sign-up required.