Glossary

A

• Agent

  An AI system that plans steps on its own, calls tools and pursues a goal instead of returning single answers.

• AI adoption

  How much AI is actually used in daily work — not the number of licences bought.

• AI champion

  A person inside a team who drives AI use, spreads knowledge and surfaces blockers.

• AI enablement

  Everything that enables teams to use AI productively and safely in their daily work.

• AI literacy

  A basic understanding of the capabilities, limits and risks of AI systems — mandated by the EU AI Act.

• AI prototype

  A running minimal solution that proves concrete value before bigger investment.

• AI readiness check

  A structured review of data, tools, skills and governance before a project starts.

• API

  An interface through which systems exchange data and functions — the base of most AI integrations.

• Automation

  A workflow that runs steps without manual triggering, usually combining trigger, logic and action.

B

• Benchmark

  A standardised test that makes models or workflows comparable.

• Blaze

  The largest Corporathon service: three to five days, several departments, production-ready results.

C

• Chain of thought

  A technique where a model spells out intermediate steps and becomes more reliable on complex tasks.

• Chatbot

  A conversational interface on top of a model; without data and tools it stays a help desk.

• Company brain

  A searchable knowledge base of a company that AI applications access in a controlled way.

• Competence record

  Documentation that employees are trained in using AI — relevant for the EU AI Act.

• Context window

  The amount of text a model can consider at once — the hard limit for inputs.

D

• Data governance

  Rules, roles and controls that define which data may be used, how and by whom.

• Deployment

  Moving a prototype into productive operation, including permissions and monitoring.

E

• Embedding

  A numeric representation of text used to find content by meaning rather than keyword.

• EU AI Act

  The EU regulation on AI with risk classes, transparency duties and an AI literacy obligation.

• Evaluation

  A systematic review of output quality against defined test cases instead of gut feeling.

F

• Fine-tuning

  Retraining a model on your own examples, useful only once prompting and context are exhausted.

G

• Guardrails

  Technical and organisational limits that keep an AI workflow from doing damage.

H

• Hackathon

  A time-boxed format in which teams build on real problems instead of listening.

• Hallucination

  A confidently phrased but wrong model output — the main reason for source grounding and review steps.

• Human in the loop

  A process design where people approve critical steps before an action takes effect.

I

• Ignite

  The mid-size Corporathon service: two days, parallel teams, several prototypes.

• Integration

  Connecting an AI workflow to existing systems such as CRM, ERP or ticketing.

L

• Latency

  Time between request and response; decides whether a workflow is accepted day to day.

• LLM

  A large language model that continues text and thereby solves tasks from summarising to code.

M

• MCP

  An open standard through which AI clients access tools and data sources in a uniform way.

• Model cost

  Cost per request, usually billed by tokens — the lever for viable business cases.

• Multimodal

  Models that process text, image, audio or video together.

N

• No-code

  Tools that let business teams build applications without classic programming.

O

• Onboarding

  A structured start for new users into tools, prompts and rules of the AI stack.

• Orchestration

  Coordinating several models, tools and steps into one reliable overall workflow.

• Output

  The result of a sprint: running solutions and documentation instead of slide decks.

P

• Pilot

  Limited productive use with real users to measure value and risk.

• Prompt

  The instruction to a model including role, context, format and limits.

• Prompt library

  A maintained collection of reviewed prompts so quality does not depend on individuals.

R

• RAG

  Retrieval augmented generation: relevant documents are retrieved and passed to the model.

• ROI

  The ratio of value to effort of an AI project, usually in saved time and cost.

• Roles and permissions

  An access model defining who may use which data and actions inside an AI workflow.

S

• Shadow AI

  Unapproved tools employees use privately — a governance and data risk.

• Spark

  The compact Corporathon service: one day, one problem, one working prototype.

• Sprint

  A fixed period in which a team delivers a defined result.

• System prompt

  A persistent instruction that sets behaviour, tone and limits of an assistant.

T

• Token

  The smallest processing unit of a model; the basis for context limits and billing.

• Tool stack

  The combination of models, automation, data and collaboration tools a team uses.

U

• Use case

  A concrete application with a trigger, participants and measurable value.

V

• Vector database

  A store for embeddings that finds content by similarity — the base for RAG.

W

• Workflow

  A sequence of steps, data and decisions that produces a work result.

Z

• Zero shot

  A task without examples in the prompt; works well for clear, standardised tasks.