Glossary

A

Agentic Coding

Software development in which an AI agent reads files, writes code, runs tests and iterates – e.g. with Cursor or Claude Code.

AI Agent

An AI system that doesn't just respond but independently plans steps, calls tools and completes tasks – the evolution of the chatbot.

AI Literacy

The ability to use AI systems sensibly, critically and in compliance with rules. Article 4 of the EU AI Act addresses it; the rule was amended in 2026.

API

Programming interface through which two systems exchange data. The basis for connecting AI to your existing tools.

Automation

A process that is triggered and completed without manual intervention – the most common first benefit of AI in everyday work.

B

Benchmark

A standardized test used to compare models. Useful for orientation, but rarely identical to your use case.

C

Chain of Thought

Intermediate steps a model generates while solving a task. More thinking time leads to better results on complex tasks.

Context Window

The amount of text a model can process at once. Limits how much documentation or code fits into a single working step.

E

Embedding

Numerical representation of text that makes semantic similarity measurable. The basis for search and RAG.

EU AI Act

European AI regulation with a risk-based approach. Sets obligations depending on risk class – from transparency to prohibition.

Evaluation

Systematic check of whether an AI workflow reliably delivers what it should. Without evaluation, every prototype remains a gut feeling.

F

Fine-Tuning

Retraining a model on your own data. In practice usually unnecessary – good prompting and RAG solve most cases more cheaply.

G

Guardrails

Technical and organizational guardrails that prevent an AI system from producing unwanted outputs or leaking data.

H

Hackathon

A time-compressed service in which interdisciplinary teams go from idea to a working prototype in hours instead of months.

Halluzination

A plausible-sounding but false model output. Mitigated through source grounding, RAG and human approval.

Human in the Loop

A setup in which a human reviews or approves AI results before they take effect. Standard for anything customer-facing.

I

Inference

The actual call to a model – the moment an input becomes an answer. The driver of ongoing costs.

K

Knowledge Base

A structured collection of your documents that an AI system cites from. The quality of the base determines the quality of the answer.

L

LLM

Large Language Model – a language model trained on huge amounts of text that understands and generates language.

M

MCP

Model Context Protocol – an open standard through which AI applications access tools, data and systems.

Multimodal

Models that process images, audio or video in addition to text. Relevant for document review and quality control.

N

No-Code

Building applications without programming skills, via interface or natural language. The core of the Corporathon stack.

O

Orchestrierung

The interplay of multiple models, tools and rules into a reliable process – e.g. with n8n or Make.

P

Prompt Engineering

The deliberate crafting of instructions to a model – role, context, structure, examples – for reliable results.

Prototyp

A working early version of a solution. Shows the benefit in a real process before budget flows into full development.

R

RAG

Retrieval Augmented Generation – the model pulls relevant excerpts from your documents before answering, staying factually grounded.

Reasoning Model

A model class that internally runs through longer solution paths before answering. Stronger at analysis, slower and more expensive to run.

S

Skill

A reusable capability for an AI assistant that is installed with one command and used in every project.

System Prompt

The persistent base instruction to a model: role, tone, boundaries. The most important lever for consistent results.

T

Token

The smallest processing unit of a model, roughly a word fragment. The compute and cost measure for AI usage.

U

Use Case

A concrete application with a trigger, stakeholders and measurable benefit – the prerequisite for any meaningful challenge.

V

Vektordatenbank

A store for embeddings that finds semantically similar content instantly. The technical backbone of many RAG systems.

Vibe Coding

Creating software by describing instead of typing: you state the goal, the model produces and corrects the code.

W

Workflow-Automation

Chaining systems and AI steps into an automated process – with tools like n8n or Make.