Moving AI in Industry from Chatbots to Operational Action

[Erkan Teskancan]

## Moving AI from Chatbots to Operational Action in Industry

The manufacturing sector is moving artificial intelligence beyond the chatbot stage and towards more operational capabilities: AI is becoming a solution that can explain what is happening in production facilities, suggest what needs to be done, and safely translate it into action. In this transformation, RAG (retrieval augmented generation) and MCP (model context protocol) technologies play a key role.

### What are RAG and MCP?

RAG enables AI to access up-to-date, facility-specific information, allowing it to generate responses from accurate documents and databases. MCP, on the other hand, allows AI to connect with systems in a standardized way, call functions, query data, and trigger workflows.

Since downtime costs are high and cybersecurity is critical in production environments, both technologies should be used together.

### RAG: Feeding the Language Model with Facility Realities

RAG supports large language models (LLMs) with external knowledge sources, enabling them to generate more accurate and up-to-date responses by using information such as standard operating procedures, quality records, maintenance notes, MES/SCADA data, and safety standards, rather than relying solely on what is within the model.

While RAG can summarize shift notes, alarm statuses, and relevant procedures, MCP can pull live production status and maintenance records from systems with precise data.

### MCP: The Connecting Protocol for Industrial AI

MCP is an open standard that enables AI applications to establish secure, bidirectional connections with external data sources and tools. This protocol provides consistent and reusable integrations between complex systems in factories.

For example, using MCP, system queries for MES, CMMS, QMS, creation of work orders, quality control, and equipment status information can be accessed securely and manageably.

### How are RAG and MCP Used Together in Workflows?

• In the analysis of quality issues, RAG retrieves relevant procedures and reports, while MCP performs real-time data queries to create work orders regarding equipment operating status.
• During shift handover, RAG provides document and alarm summaries, while MCP extracts live production data, ensuring a secure handover with real and documented information.
• In recipe and setpoint changes, RAG supports processes with approved information, while MCP records change requests in the system, manages approval processes, and keeps records for auditing.

### Risk Management and Security Layer

Industrial AI is not just software; it is a decision support system where security and compliance are critical. Therefore:

• Standard approaches such as the NIST AI Risk Management Framework should be adopted.
• Compliance with OT cybersecurity practices (such as ISA/IEC 62443) must be ensured.
• Reliable data sources should be used with industrial data standards like OPC UA.

### Implementation Roadmap

For high-value information workflows, start with RAG to provide verified responses from internal sources. MCP implementation for controlled tool access should initially be limited to read-only operations (production status, alarms, historical data). Write operations should be carefully added to existing workflows (maintenance requests, ticket creation, etc.).

Throughout these processes, traceability and security must be kept at a high level by adhering to standards such as NIST and ISA/IEC.

AI for industry is becoming a technology that can not only provide information but also generate actions safely and integrally on the shop floor. This enables operators to make effective and secure decisions in processes without having to manually manage many different systems.