IT & AI Glossary for Manufacturing

Plain-language definitions of ERP, MES, RPA, AI agents, RAG and the other IT and AI terms that matter for Swiss manufacturing.

Agent skill: A packaged, reusable instruction set telling an AI agent how to carry out one defined task, including which tools to call. Holding a procedure as a skill keeps it consistent between runs, instead of leaving the model to work it out again each time.
AI agent: A software system that uses a language model to pursue a goal across several steps (calling tools, reading data and taking actions) rather than answering a single prompt. In manufacturing it can triage documents, query systems or orchestrate a workflow.
API (Application Programming Interface): A defined contract that lets two software systems exchange data automatically. APIs are how an ERP, MES and AI tools integrate instead of relying on manual re-keying.
CAD / CAM: Computer-Aided Design (3D models of parts) and Computer-Aided Manufacturing (turning those models into machine/NC programs). Together they connect engineering to production.
CAQ (Computer-Aided Quality): Software for managing inspection plans, measurement data and quality records across production. It is the digital backbone of quality assurance.
Computer Vision (Bildverarbeitung): AI that interprets images: detecting, inspecting and measuring. On the shop floor it automates visual quality control and part recognition.
Cpk (Process Capability): A process-capability index measuring how reliably a process stays within tolerance. A higher Cpk means fewer defects and less rework.
Data Lake: A central store for structured and unstructured data from many sources, kept in raw form until needed. It is the foundation for analytics and AI across the plant.
Deep Learning: Machine learning with deep neural networks, the foundation of modern image and language models. It handles complex, unstructured inputs such as photos, scans and free text.
Determinism: A process is deterministic when the same input always produces the same output. Plain code is deterministic; a language model is not, so the same document can be read slightly differently on a second run. This is why anything that can be stated as a rule belongs in code rather than in the model.
Deterministic solver: A program that calculates the best answer to a bounded planning problem, such as a production sequence or a machine assignment. Given the same inputs it always returns the same result, and it can show how it got there. That is what separates it from a language model, which produces a plausible answer rather than a provably best one.
Digital twin: A live virtual replica of a machine, product or process, kept in sync with real sensor data. It is used to monitor, simulate and optimise operations without touching the physical asset.
ERP (Enterprise Resource Planning): Central software that plans and records a manufacturer’s core resources (orders, inventory, purchasing, production and finance) in one shared data model. It is usually the system of record that other shop-floor and AI tools integrate with.
Hyperscaler: One of the very large global cloud providers that rent out computing power and AI services, such as AWS, Microsoft Azure or Google Cloud. The term matters when data location or vendor dependence is a constraint, because the alternative is running the same workload on your own hardware.
IIoT (Industrial Internet of Things): The network of connected sensors, machines and controllers on the shop floor that stream operational data. It is the foundation for monitoring, analytics and AI on production assets.
LLM (Large Language Model): An AI model trained on large text corpora that understands and generates natural language. It is the engine behind document extraction, plain-language querying and assistant features.
Machine Learning (ML): A branch of AI where systems learn patterns from data instead of being explicitly programmed. In manufacturing it powers quality prediction, demand forecasting and anomaly detection.
MCP (Model Context Protocol): An open standard that lets AI agents connect securely to systems and data such as an ERP or MES, so they act on live information instead of guessing.
MES (Manufacturing Execution System): The layer between the ERP and the machines. It dispatches production orders to the shop floor, tracks work-in-progress in real time and feeds actual output, quality and downtime data back up to the ERP.
MRP (Material Requirements Planning): The planning logic that turns demand and bills of material into what to buy or produce and when. It sizes purchase and production orders so material arrives just in time for manufacturing.
OCR (Optical Character Recognition): Technology that turns scanned or photographed documents into machine-readable text. Combined with AI it extracts structured data from unstructured PDFs such as orders and invoices.
OEE (Overall Equipment Effectiveness): A single figure for how productively a machine runs, combining availability, performance and quality. It is the standard KPI for spotting hidden capacity losses on the shop floor.
On-premises: Software that runs on hardware you own and control, in your own building or your own data centre, rather than in a provider’s cloud. For AI it became practical once models small enough to run on a single server became available.
OPC UA: An open, vendor-neutral industrial standard for exchanging machine and process data between equipment, MES and IT systems. It is the common language of the connected shop floor.
Open-weights model: A model whose trained parameters are published, so it can be downloaded and run on your own hardware. Parameters are the learned values the model is made of. Their count, often written as 7B or 70B, is a rough proxy for size and hardware appetite.
PLC / SPS (Programmable Logic Controller): The industrial computer that controls a machine or line in real time. It reads sensors and drives actuators, and is the primary data source for machine states, cycle times and faults.
Predictive maintenance: Using machine and sensor data to predict failures before they happen, so maintenance is scheduled by actual condition instead of a fixed calendar, which cuts unplanned downtime.
Process Mining: Analysing event data from your systems to reconstruct how processes actually run, revealing bottlenecks, deviations and drift over time.
RAG (Retrieval-Augmented Generation): A technique that grounds a language model in your own documents and data by retrieving relevant passages before generating an answer. It reduces hallucination and lets an assistant cite company-specific facts.
RPA (Robotic Process Automation): Software “robots” that automate repetitive, rule-based digital tasks such as copying data between systems, filling forms and moving files, without changing the underlying applications.
Script: A short program that runs a fixed sequence of steps the same way every time. In an AI system it handles the steps that have one correct outcome, so the model never re-derives them.
Straight-through processing (STP): The share of incoming documents that run through a process from end to end without anyone touching them. In order intake it is the measure that describes the relief a team actually feels, because an extraction can be correct and still leave every order sitting in a review queue.
Token: The unit AI providers bill by, roughly a fragment of a word. Both the text you send and the text the model returns are counted. Usage-based pricing therefore grows with how much a system reads and writes.
Vision-Language Model (VLM): A model that processes an image and text together, so it can answer questions about a document page instead of only transcribing it. In document processing it competes with classic OCR pipelines, and the two have different strengths depending on the document type.