Something significant is already happening in the operations of manufacturers you compete with. It isn’t a pilot program. It isn’t a roadmap slide. It’s running in production, making decisions, and doing it faster than any human team could.
At Microsoft’s Convergence 2025 event, the company demonstrated AI agents embedded directly into Dynamics 365 — handling supplier outreach in supply chain management, scheduling field service technicians, reconciling financial transactions, and managing product change approvals that previously took weeks. Coca-Cola Beverages Africa is already using these agents in live manufacturing operations. This isn’t a preview of 2030. This is December 2025.
Infor describes the shift plainly in its 2026 manufacturing outlook: the conversation has moved from AI experimentation to deployment at scale, from systems that analyze and recommend to systems that “pursue defined outcomes by coordinating decisions, taking actions, and orchestrating processes across planning, production, and execution.” The word they use is agentic. It means the system acts — not waits to be asked.
Gartner projects that 40% of enterprise applications will embed AI agents by the end of 2026, up from less than 5% in 2025. That is not a gradual adoption curve. That is a step change — the kind that separates the operations that were ready from the ones that weren’t.
What Agents Actually Do
An AI agent is not a chatbot with better answers. It is a software system that can perceive a situation, reason about it, connect to external systems through APIs, and take autonomous action — without waiting for a human to initiate each step.
In a manufacturing context, a procurement agent doesn’t generate a report suggesting a supplier is at risk. It detects the risk signal, contacts the supplier autonomously to request an updated delivery estimate, cross-references that estimate against the production schedule, identifies which jobs are affected, adjusts the schedule, and notifies the relevant people — all before the morning shift starts. The procurement manager doesn’t manage the process. They audit the decisions the agent already made.
A maintenance agent doesn’t alert a technician that a machine is showing anomalous vibration patterns. It identifies the pattern, accesses the maintenance history through the ERP, cross-references parts availability, schedules the intervention during a planned low-utilization window, stages the required parts, and generates the work order — autonomously. The technician arrives at a prepared job, not an emergency.
This is not science fiction. These architectures are in production today at early-adopting manufacturers. The AI agent market is projected to grow from $7.8 billion in 2025 to over $52 billion by 2030 — a trajectory that reflects actual deployment, not speculative interest. (Gartner / Machine Learning Mastery, Agentic AI Trends 2026)
Why APIs Are the Infrastructure Nobody Is Talking About
Agents don’t operate in isolation. They operate across systems — pulling data from one, triggering actions in another, coordinating outputs across several simultaneously. The technical mechanism that makes this possible is the API: the documented connection point through which two software systems exchange information and instructions in real time.
This is why the architectural decisions your organization has already made — or hasn’t made — matter so much more than most leadership teams realize.
An AI agent tasked with autonomous production scheduling needs to read real-time machine status data, query inventory availability, check labor schedules, access customer order commitments, and write updates back to the production plan. Each of those operations requires an API connection. In a modern, cloud-native ERP environment with open architecture, those connections are available, documented, and accessible. In a legacy ERP environment with closed architecture and years of custom code, they are not — or they require expensive, fragile middleware that breaks when either system updates.
The manufacturers who will deploy agents effectively in 2028 and 2030 are the ones whose core systems are already API-connected today. The ones whose systems aren’t will spend the next several years rebuilding infrastructure before they can start on the capability that runs on top of it.
Deloitte predicts a fourfold increase in agentic AI adoption in manufacturing between 2025 and 2026 alone — from 6% to 24%. The manufacturers in that 24% are not more technologically sophisticated in some abstract sense. They have cleaner data, more open system architecture, and better API connectivity than their peers. They built the foundation. The agents are the next layer. (Dataiku / Deloitte, Manufacturing’s 2026 Mandate)
What 2030 Actually Looks Like
By 2030, the manufacturers operating at competitive performance levels will run operations that look fundamentally different from what most mid-market plants look like today.
Production planning will be continuously optimized by agents that balance customer commitments, machine availability, material supply, and labor capacity in real time — not in the weekly planning meeting. Supply chain disruptions will be detected and responded to autonomously, with agents rerouting procurement, adjusting schedules, and notifying customers before the disruption becomes a service failure. Financial close processes will be largely automated, with agents reconciling transactions, flagging anomalies, and generating reports without a team spending three days at month end.
The humans in these operations won’t be eliminated. They’ll be repositioned. Plant managers will supervise agent performance rather than manage production directly. Finance teams will review agent decisions rather than produce the underlying analysis. Procurement managers will set policy and handle exceptions rather than execute routine transactions.
SAP describes this trajectory as the shift “from systems of record to systems of action” — a phrase worth sitting with. (SAP, AI in 2026: Five Defining Themes) A system of record stores what happened. A system of action decides what happens next.
The question every manufacturing executive should be asking right now is not whether this future is coming. The data is unambiguous — it’s already arriving. The question is whether the systems your organization runs today are systems of action or systems of record. And if they’re systems of record, what it would take to change that — and how much time that change requires.
Because the manufacturers who are already in the 24% didn’t start building last month. They started building two or three years ago. The ones starting today are two or three years behind them.
That gap doesn’t close by watching it.
Sources:
– Gartner / Machine Learning Mastery, 7 Agentic AI Trends to Watch in 2026 — machinelearningmastery.com
– Dataiku / Deloitte, Manufacturing’s 2026 Mandate: From AI Pilot to Agentic Profit — dataiku.com
– SAP,