Why You Need A Framework for Agentic AI in Manufacturing Operations?

Today, anyone can build an AI agent, and many are. But that leads to a much more important question:

What agents should we build, and how do we organize them into something meaningful?

For decades, manufacturing thinkers have dreamed of systems that are adaptive, modular, and self-optimizing.

We had the theory, but not the tools. Now, we do.

Thanks to IIoT, cloud-native platforms, and AI, that vision is starting to take shape.

Manufacturing applications are already behaving like agents:

‍
→ They have goals (e.g., guide operators, track units).
→ They act autonomously in their context.
→ They collaborate through shared data and events.

Add Agentic AI; We go beyond automation, toward true operational intelligence.

The real breakthrough happens when these agentic applications are composed into systems, dynamic, adaptive, and reflective of real-world operations.

Digital twins, in the truest sense.

But here’s the challenge: Without structure, we risk a mess of disconnected bots that confuse more than they help.

That’s why Gilad Langer, Head of Digital Manufacturing Practice at Tulip Interfaces, proposes a Composable Agentic Framework for manufacturing operations.

At its core is the Artifact Model, a shared context that defines products, orders, machines, deviations, and people.

Agents operate within this model, enriching it in real time with state, decisions, and traceability.

The Framework defines a Taxonomy of Agents:

‍
1. 𝐏𝐡𝐲𝐬𝐢𝐜𝐚𝐥 𝐀𝐠𝐞𝐧𝐭𝐬
Represent physical entities like machines, products, and totes.

2. 𝐎𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐀𝐠𝐞𝐧𝐭𝐬
Coordinate workflows such as scheduling, orders, and handling deviations.

3. 𝐒𝐲𝐬𝐭𝐞𝐦 𝐀𝐠𝐞𝐧𝐭𝐬
Bridge and synchronize platforms like ERP systems, UNS, and data lakes.

4. 𝐂𝐨𝐦𝐩𝐚𝐧𝐢𝐨𝐧 𝐀𝐠𝐞𝐧𝐭𝐬
Support humans, operators, engineers, and citizen developers.

‍