At SAP Sapphire 2026, SAP moved the conversation from assistive AI to agentic AI -  systems that do not just recommend an action but trigger it. For maintenance teams, that means work orders created automatically, parts reserved automatically, technicians scheduled automatically.

Here is the uncomfortable question that announcement raises for aerospace and defense asset owners: can your asset data actually support that? For most A&D maintenance programs, the honest answer is not yet. This piece lays out what actually changed, where the real gap sits, and what to fix first.

Why Reactive Maintenance Is a Program Risk in A&D, Not Just a Cost Problem

BLUF: In aerospace and defense, unplanned downtime is not only a budget line. It is a compliance exposure, a safety exposure, and a program-schedule exposure -  and those three combine in ways most cost-per-outage models never capture.

A commercial manufacturer that loses a production line for a shift absorbs a cost and moves on. An A&D operator that loses an asset mid-program can trigger a contractual delivery slip, a regulatory reporting obligation, or an airworthiness question that outlives the repair itself.

Preventive maintenance schedules were built for a world where "safe" meant "serviced on time, according to plan." They were never built to catch the failure mode that happens between scheduled intervals -  which is exactly the failure mode predictive maintenance exists to catch. That gap is why reactive and calendar-based preventive maintenance keep showing up on A&D risk registers even in mature EAM environments.

The cost conversation matters. But for this audience, it is not the opening argument. The opening argument is exposure.

What Changed at SAP Sapphire 2026: From Assistive AI to Agentic AI

BLUF: SAP's Sapphire 2026 announcements drew a clear line between AI that assists a human decision and AI that acts on its own, inside defined guardrails. That distinction is the one worth understanding before any budget conversation happens.

Assistive AI is the version most SAP customers already run: a chatbot-style assistant that summarizes an asset's failure history, drafts a work order for a human to approve, or answers a natural-language question about inventory. Useful, but a person is still in the loop for every action.

Agentic AI, as SAP framed it under its "Autonomous Enterprise" and Joule direction, is a step further: the system observes a condition, evaluates it against a policy, and initiates the work order, the parts reservation, or the technician dispatch without waiting for a human to click approve. The human still sets the policy and reviews the outcome -  but the trigger is automatic.

That shift changes the requirements on the data underneath it. An assistant that drafts a suggestion for a person to check can tolerate some noise in the source data, because a human is the final filter. An agent that acts on its own cannot. It needs asset data that is accurate, current, and structured consistently enough to trust without a human catching the error after the fact.

This is the part of the Sapphire 2026 announcement that got less coverage than the demo. The AI license is the easy purchase. The data foundation underneath it is not.

The Data Readiness Gap Nobody's Talking About

BLUF: Most A&D EAM programs already hold the SAP AI license they need. Very few of them hold asset data clean enough for that AI to act on it without supervision.

This is the gap worth naming directly: SAP's new agentic AI cannot act on asset data that is still sitting in a spreadsheet, split across shadow systems, or entered inconsistently across sites. Data readiness, not AI licensing, is the real 2026 bottleneck for A&D maintenance teams.

The pattern shows up the same way across programs. Equipment master data was built for transactional reporting, not for a model deciding whether to trigger a work order. Sensor and IoT data lives in a separate historian that was never mapped cleanly to the SAP asset record. Maintenance history includes years of free-text technician notes that a model cannot parse with confidence. None of this is a failure of ambition. It is the accumulated residue of years of point fixes layered onto a core system.

SAP research backs up how far along the underlying S/4HANA migration already is: SAPinsider's most recent ERP Migration and Transformation research found that 55% of organizations have deployed SAP S/4HANA or SAP S/4HANA Cloud, which puts the technical foundation for AI-driven predictive maintenance in place for a majority of the market already. The same research found that SAP's AI announcements ranked as the top external factor shaping ERP strategy for 54% of respondents -  in other words, the platform is there and the pressure to act on AI is there. What is inconsistently there is the asset-data discipline required to let the AI act safely.

How SAP AI Actually Works for Asset Lifecycle Intelligence Today

BLUF: Set the roadmap talk aside for a moment. Here is what is actually in production today, in plain terms, without the T-codes.

SAP's asset intelligence stack today is built from a few components working together:

  • SAP Predictive Asset Insights (PAI) uses machine learning against historical failure and sensor data to estimate the likelihood and timing of an asset failure, rather than relying on a fixed calendar interval.
  • SAP Asset Performance Management (APM) brings that risk scoring into the maintenance planning process, so a technician or planner sees a prioritized list of what actually needs attention rather than a flat schedule.
  • IoT integration feeds live condition data -  vibration, temperature, cycle counts -  into the asset record, so the risk score reflects current state rather than a snapshot from the last inspection.
  • SAP Business AI and Joule sit on top as the interface layer: summarizing what the data shows, answering natural-language questions about an asset, and -  under the newer agentic model -  initiating defined actions within policy.

None of these components is new in isolation. What changed at Sapphire 2026 is how directly SAP is positioning Joule to act rather than just advise, and that positioning is what raises the bar on the data feeding it.

For an executive audience, the useful frame is this: SAP AI does not replace the maintenance planner's judgment. It replaces the static calendar with a live risk picture, and it changes how much of the routine decision work still needs a human hand on it.

What This Looks Like in A&D: MRO, Compliance, and Long-Program Asset Management

BLUF: A&D asset management runs on longer program cycles, tighter compliance obligations, and more export-control-sensitive data than most industries SAP's general EAM messaging is written for -  and that changes how this technology should actually be deployed.

Aerospace and defense MRO operations do not run on the same rhythm as discrete manufacturing. An asset in a long-cycle program might need documented maintenance history that spans a decade or more, tied to specific regulatory and contractual reporting requirements. Predictive maintenance data has to feed into that documentation trail, not sit alongside it in a separate system.

Compliance-sensitive data also changes how the underlying architecture gets built. Export-control classifications, program-specific access restrictions, and government customer requirements mean asset data governance in A&D is not just a data-quality exercise -  it is a data-segregation and access-control exercise that has to be designed in from the start, not retrofitted after an AI pilot goes well.

None of this is a reason to avoid the technology. It is a reason to sequence the rollout deliberately: start with the assets and the data domains where governance is already clean, prove the model there, and expand from a position of evidence rather than a company-wide mandate on day one.

Where Clean Core Discipline Fits In

BLUF: Technical debt in the SAP core is the single biggest drag on how fast an A&D organization can move from assistive to agentic AI -  which is why Clean Core discipline belongs in this conversation, not as a separate initiative.

Years of custom code, workarounds, and point-to-point integrations accumulate in most long-tenured SAP environments. Each one is defensible on its own. Together, they are exactly what slows an S/4HANA upgrade, complicates a move to RISE with SAP, and -  most relevant here -  makes asset data harder to trust at the level an agentic AI model requires.

Clean Core is not a rebuild-everything mandate. It is a discipline: keep custom logic in extension layers where it belongs, keep the core close to standard, and treat every workaround as a decision with a cost attached. Programs that have already made progress on Clean Core tend to find the AI readiness conversation shorter, because the data hygiene work overlaps directly with the AI data-quality work.

[Internal link: ITChamps Clean Core pillar page]

A Practical First Step: Assessing Asset Data Readiness Before Scaling AI

BLUF: Before scaling any agentic AI capability across an A&D asset base, the lower-risk move is a focused assessment of what the data actually looks like today -  not a full AI rollout and not a wait-and-see posture either.

An asset data readiness assessment is a scoped, low-commitment starting point. It answers three questions directly: which asset domains have data clean enough to support predictive and agentic AI today, which need remediation first, and what sequencing gets the organization to a defensible pilot without a company-wide commitment upfront.

ITChamps is an SAP Gold Partner with SAP Application Managed Services (AMS) and S/4HANA migration experience across manufacturing environments, including asset-intensive operations. [NEEDS INTERNAL INPUT: confirm current service-line description and A&D-specific project references against the ITChamps Approved Claims Registry before publishing.] The starting point is not a new AI license. It is an honest look at whether the data underneath the license is ready to be trusted.

Frequently Asked Questions

What is the difference between assistive AI and agentic AI in SAP's 2026 announcements? 

Assistive AI drafts a recommendation -  a summary, a suggested work order -  and waits for a person to approve it. Agentic AI, as SAP described it at Sapphire 2026, evaluates a condition against a defined policy and initiates the action itself, such as creating a work order or reserving parts, within guardrails a human has set in advance.

Do we need to buy new SAP AI licenses to start predictive maintenance? 

Often not as a first step. Many A&D organizations already run S/4HANA or S/4HANA Cloud and hold access to SAP Predictive Asset Insights or Asset Performance Management. The more common gap is asset data quality, not license coverage -  which is why a data readiness assessment is usually the more useful starting point than a licensing conversation.

How is predictive maintenance different for aerospace and defense compared to general manufacturing? 

A&D asset management runs on longer program cycles, tighter regulatory and export-control requirements, and documentation obligations that a predictive maintenance rollout has to integrate with, not bypass. Data governance and access segregation matter as much as the predictive model itself.

What does "Clean Core" have to do with AI readiness? 

Custom code and workarounds accumulated in the SAP core over time make asset data harder to trust and harder to move quickly. Clean Core discipline -  keeping customization in extension layers and the core close to standard -  overlaps directly with the data hygiene work AI models need, which is why the two conversations tend to move together.

What is a reasonable first step for an A&D maintenance team evaluating this? 

A scoped asset data readiness assessment rather than a full AI rollout. It identifies which asset domains are ready for predictive or agentic AI today, which need remediation, and how to sequence a pilot without committing the whole program upfront.