Rolls-Royce Hands Jet Engine Maintenance Planning to a Team of AI Agents
Rolls-Royce has put a multi-agent AI system in charge of one of the thorniest logistics problems in aviation: deciding, continuously, which jet engines go into which maintenance shop, in what order, and with which parts and technicians on hand. The system, called Optimiser, was built with Microsoft and disclosed this week in a joint account published by Microsoft UK Stories on October 21, with independent writeups following from Windowsforum and the AI-news newsletter RioRundown the same day and the day after.
Four agents, one plan
Optimiser coordinates four specialized agents: an Engine Agent that tracks the condition of individual engines, a Workscope Agent that determines what maintenance is actually needed, Facility Agents that model workshop capacity and capability, and a Supplier Agent that manages parts and logistics. Rolls-Royce says the four were built to replace a manual planning process that used to tie up four or five staff for roughly a week each time conditions changed — a regulatory hold, a parts delay, a workshop running over capacity. Executives Jamie Barkley, who leads aftermarket optimisation, and Matt Ley, director of engineering, technology and safety for civil aerospace, both described the goal as reducing 'aircraft-on-ground' situations, where airlines have a plane grounded waiting on an engine.
The scale of the underlying problem explains why Rolls-Royce reached for agents rather than a conventional scheduling tool. The company managed close to 1,500 engine shop visits in 2025. Each engine carries roughly 30,000 parts, and sensors sample around 100 operating parameters every 1.6 seconds, generating millions of variables across a single year's maintenance programme. More than 40 Rolls-Royce business systems now run on Microsoft Azure, which is what makes it possible for the agents to pull engine health data, workshop schedules and supplier status into one continuously updating plan rather than a static document that goes stale the moment something changes.
What's actually proven, and what isn't
What's notable is what Rolls-Royce has not published alongside the launch. Neither the Microsoft account nor the independent coverage of it includes a measured reduction in planning hours, a percentage improvement in turnaround time, or a count of shop visits the system has actually replanned to date. Microsoft's own language calls Optimiser 'still developing,' and the 'materially reduced' planning effort is an executive characterization rather than an audited figure. That gap is worth flagging precisely because the pitch — replacing a week of senior staff time with an always-current agentic plan — is the kind of claim enterprise AI vendors increasingly make before the operational data exists to back it up.
Both companies are explicit that humans retain sign-off: regulators still have to clear an engine's safety criteria before it returns to service, and Rolls-Royce frames the agents as recommending plans for staff to approve rather than executing them autonomously. That human-in-the-loop structure mirrors a pattern showing up across other agentic rollouts this month, as enterprises pair new agent deployments with explicit oversight layers rather than letting agents act unsupervised in safety- or compliance-sensitive workflows.
For Microsoft, the deal is another data point in its push to sell Azure as the substrate for industrial multi-agent systems, following a string of enterprise AI wins this year. For Rolls-Royce, which has spent the past decade digitizing engine health monitoring, Optimiser is a bet that the next efficiency gain in its services business comes not from better sensors but from software that can replan faster than its own supply chain can change. Whether that bet pays off in measurable turnaround-time or cost terms is something outside observers won't be able to check until Rolls-Royce — or an independent customer airline — publishes numbers of its own.