The ERP agent rerouted the build. Nobody filed the engineering change.

Gartner's 2026 hype cycle for agentic automation projects a $40 billion agentic ERP core for discrete manufacturing. Vendor samplings, Farnborough demos, governance columns — autonomous agents resequencing work orders, rerouting builds, reassigning suppliers in real time. I have seen what happens when a routing changes without an engineering change notice. At Airbus, the 97% reduction in internal lead time I drove came from Routing Verification KPIs — obsessive, granular control of routing accuracy, because routing drift is one of the fastest paths to nonconforming product I have ever seen. That discipline was built for human-driven changes that moved at ECN speed. Now we are handing software agents the authority to make the same changes in 200 milliseconds, with no paper trail, no PFMEA review, and no awareness that PPAP exists.

What the agent actually does on the floor

The vendor demos look compelling. Supply disruption detected, component reassigned to a qualified alternate, work orders resequenced to keep the line fed. Machine trending toward failure, production queue rerouted to a backup cell. In logistics terms, these are optimisation decisions. In quality terms, they are manufacturing process changes — and that is the only term that matters when the auditor walks in.

When you reroute a work order from Cell A to Cell B, you may be changing the tooling, the work instructions, the in-process inspection points, and the operator qualifications required for that operation. Under AS9100 §8.1, each of those variables carries validation requirements. When you reassign a supplier mid-build, you are introducing new material into a production process that was validated — through PPAP, through first article inspection, through supplier qualification audits — against a specific supplier's process capability. The agent sees a constraint and solves it. The quality system sees a process change and has no record it occurred.

Under IATF 16949 §8.5.6, any change to a manufacturing process requires controlled conditions: change identification, documentation, risk assessment, and — depending on severity — customer notification or PPAP revalidation. The agent does not know this clause exists. Your QMS does not see the transaction. The finding writes itself.

The traceability gap

The real problem starts after the commit. The PFMEA still references the old path. The control plan still lists the old inspection points. The standard work instructions still describe the old sequence. You are now building product to two specifications at once: the one your quality system documents and the one your ERP agent just created. Every part produced after the agent's decision carries a traceability gap that widens with each work order.

I spent years building the routing-verification discipline that catches this drift — daily KPI checks, node-by-node accuracy tracking, deviation flags that fire before product moves to the next operation. That system works because the rate of human-driven change is manageable. An engineer files an ECN, the change board reviews it, PFMEA and control plans are updated, work instructions are revised, and only then does the routing change in the system. The whole cycle might take a week. An agent makes 400 routing decisions in an hour.

The agent doesn't need to understand PPAP. Your quality system needs to understand the agent.

This is not speculative. Enterprise reporting shows only 26% of organisations believe their AI governance keeps pace with deployment. Major industrial vendors are demonstrating autonomous agents for aerospace operations right now, while governance frameworks trail behind by quarters if not years. The gap between what the technology can do and what the quality system can control is not closing.

What real control looks like

Banning agents from the ERP is not the answer. That ship has sailed, and the optimisation potential is genuine. The answer is to hard-wire change-management discipline into the integration layer before the agent ever touches a routing.

Agent decision boundaries must be explicit and enforceable at the system level — not in the agent's prompt. Prompts are advisory. APIs are enforceable. If an agent can reroute a work order, the integration layer must verify whether that routing exists in the approved engineering baseline. If it does not, the commit is blocked and routed through an expedited ECN workflow. The agent proposes. The QMS validates. The routing commits. In that order.

Every agent action that constitutes a process change — rerouting across cells with different capability profiles, supplier reassignment, resequencing that alters inspection timing — must trigger an engineering-change event in your QMS before the transaction posts. Not after. Before. And the Routing Verification KPI structure that caught human-driven drift needs to run at machine speed, because a daily audit cycle cannot absorb hundreds of autonomous decisions per hour.

Gartner's hype cycle will tell you when agentic ERP peaks. It will not tell you whether your change-management system can absorb a decision made in 200 milliseconds by software that has never opened IATF 16949. The $40 billion opportunity is real. So is the audit finding that comes with it. The organisations that win in agentic manufacturing will be the ones that extend their quality discipline into the integration layer before the first agent goes live — not the ones scrambling to explain a routing change to a customer after a nonconformance nobody authorised and nobody caught.