AI & Agentic AI Glossary

Large Language Model (LLM)

An AI system trained on vast amounts of text that can understand, reason about and generate human language at expert level.

LLMs are the reasoning engine behind modern AI agents. They process instructions, understand context, generate responses and decide which actions to take. Models like Claude, Gemini and GPT-4 are LLMs. The quality of an LLM directly determines the quality of the AI agents built on top of it.

Example: An LLM reads a 40-page contract, identifies all renewal clauses and termination conditions, flags three clauses that deviate from standard terms and drafts a summary for the legal team — in under 30 seconds.

Also known as: foundation model, language model, AI model, generative model, transformer model

Generative AI

A class of AI models that create new content — text, images, code, audio, video — by learning patterns from existing data and generating novel outputs.

Generative AI is the technology behind tools like Claude, Gemini and GPT-4. It can write reports, draft emails, generate code, create images and summarise documents. In enterprise, it is most powerful when combined with agentic capabilities — moving from content that is generated to content that is acted upon.

Example: A marketing team uses generative AI to produce 50 personalised email variants for a campaign in 20 minutes — each adapted to a different customer segment and communication stage.

Also known as: GenAI, creative AI, AI content generation, foundation model outputs

Prompt Engineering

The practice of crafting precise instructions for AI models to produce accurate, consistent and useful outputs for specific business tasks.

The quality of an AI agent's output depends heavily on the quality of its instructions. Prompt engineers design, test and refine the instructions that guide AI agents — defining their role, constraints, tone, output format and how to handle edge cases. It is a critical skill for deploying reliable AI in production.

Example: A well-engineered prompt for a customer support agent specifies: always respond in the customer's language, never promise refunds without checking order status, escalate if sentiment is negative for more than two exchanges, and format responses in bullet points.

Also known as: prompt design, prompt optimisation, instruction engineering, system prompt design

Fine-tuning

The process of training an existing AI model on your specific business data to make it significantly more accurate and relevant for your use case.

A general-purpose LLM knows a lot about the world but nothing specific about your company, products or industry jargon. Fine-tuning adapts the model to your context by training it on your domain examples — making it faster, cheaper and more accurate for your specific tasks.

Example: A legal firm fine-tunes an LLM on 10,000 past contracts, enabling the model to identify non-standard clauses with 94% accuracy — compared to 67% for a general-purpose model on the same task.

Also known as: model fine-tuning, domain adaptation, model customisation, supervised fine-tuning

Artificial Intelligence (AI)

The simulation of human intelligence by machines, enabling them to perform tasks that typically require human cognition such as reasoning, learning, problem-solving and decision-making.

Also known as: AI, machine intelligence

Machine Learning

A subset of AI where systems learn from data and improve their performance over time without being explicitly programmed for each task.

Also known as: ML, statistical learning, predictive modelling

Deep Learning

A subset of machine learning that uses neural networks with many layers to learn complex patterns in large datasets. It powers most modern AI breakthroughs including language understanding.

Also known as: neural networks, DL, deep neural networks

Neural Network

A computing system inspired by the human brain, composed of interconnected nodes that process information in layers. The foundation of modern deep learning.

Also known as: artificial neural network, ANN, deep learning model

Natural Language Processing (NLP)

A branch of AI focused on enabling computers to understand, interpret and generate human language. NLP powers chatbots, translation, sentiment analysis and AI agents.

Also known as: NLP, computational linguistics, text AI

Transformer (Architecture)

The neural network architecture that powers virtually all modern large language models. Introduced by Google in 2017, it uses attention mechanisms to process text in parallel — enabling the scale that makes LLMs possible.

Also known as: transformer model, attention-based model, self-attention architecture

Attention Mechanism

A component of transformer models that allows the AI to focus on the most relevant parts of the input when generating each output. It enables LLMs to understand long-range context and relationships in text.

Also known as: self-attention, cross-attention, multi-head attention

Tokenisation

The process of breaking text into smaller units (tokens) that an AI model can process. API costs are calculated per token, making tokenisation awareness important for enterprise AI budgeting.

Also known as: text tokenisation, subword tokenisation, BPE

Hallucination

When an AI model generates information that sounds confident and plausible but is factually incorrect or entirely fabricated. A key risk to manage in enterprise AI deployments.

Also known as: AI confabulation, model fabrication, LLM error

Token

The basic unit of text processed by a language model. A token can be a word, part of a word, or punctuation. LLMs have token limits that define how much text they can process at once.

Also known as: token limit, context token, input token

Inference Cost

The computational cost of running an AI model to generate outputs. In enterprise AI, inference cost is the primary ongoing expense after deployment and directly impacts ROI calculations.

Also known as: inference pricing, API cost, token cost, compute cost

Model Distillation

The process of training a smaller, faster model to replicate the behaviour of a larger, more capable model. Distilled models achieve near-equivalent performance at a fraction of the compute cost.

Also known as: knowledge distillation, model compression, teacher-student training

Benchmark

A standardised test used to evaluate and compare AI model performance on specific tasks. Common benchmarks include MMLU (general knowledge), HumanEval (coding) and HellaSwag (reasoning).

Also known as: AI evaluation, model evaluation, performance benchmark

Backpropagation

The algorithm used to train neural networks by calculating how much each parameter contributed to the error and adjusting weights accordingly. The mathematical engine behind modern deep learning.

Also known as: backprop, gradient backpropagation, reverse-mode differentiation

Parameters (AI Model)

The numerical values a model learns during training that determine how it processes inputs and generates outputs. GPT-4 has an estimated 1.8 trillion parameters.

Also known as: model weights, learned weights, model parameters

Agentic AI

AI systems capable of autonomous planning, decision-making and execution across complex, multi-step workflows — without requiring human input at every stage.

Agentic AI goes beyond generating text or answering questions. It perceives context, sets sub-goals, selects tools, executes actions and iterates until a task is complete. It is the evolution from AI as a tool you use to AI as a system that works for you.

Example: A finance team deploys an agentic AI that monitors bank statements nightly, flags anomalies, matches transactions against invoices, updates the ERP and generates a reconciliation report — ready for the CFO each morning.

Also known as: agentic systems, autonomous AI, agent-based AI, proactive AI, self-directed AI

AI Agent

A system that autonomously executes multi-step tasks across tools and data sources to achieve a defined business goal — without human intervention at each step.

Unlike a chatbot that answers questions, an AI agent acts. It perceives its environment, reasons about what to do, takes real actions — updating databases, sending emails, calling APIs — and adapts based on the results. AI agents are the core building block of modern enterprise automation.

Example: A sales AI agent receives a new lead, enriches contact data from LinkedIn, scores the lead, updates the CRM, sends a personalised follow-up email and notifies the sales rep — all in under 60 seconds.

Also known as: autonomous agent, agentic system, AI assistant, intelligent agent, LLM-powered agent

AI Orchestration

The coordination of multiple AI agents, models and tools into a unified workflow that delivers a single, coherent business outcome.

In an orchestrated system, an orchestrator agent breaks a complex goal into sub-tasks, assigns each to a specialist agent, manages dependencies, collects outputs and assembles the final result. Orchestration is what makes AI scale across an entire organisation.

Example: An orchestrated system handles a customer complaint end-to-end: one agent classifies the issue, another retrieves order history, a third drafts the response, a fourth updates the CRM and a fifth triggers a refund if needed.

Also known as: multi-agent orchestration, agent coordination, AI workflow orchestration, agentic workflow

RAG (Retrieval-Augmented Generation)

A technique that makes AI models more accurate by retrieving relevant information from your own data before generating a response — dramatically reducing hallucinations.

A standard LLM only knows what it was trained on. RAG gives it access to your specific documents, databases and knowledge bases in real time. The system retrieves the most relevant information and feeds it to the model, which then generates a grounded, accurate response.

Example: A customer support agent uses RAG to query internal product documentation, past tickets and pricing tables in real time — giving customers precise, up-to-date answers instead of generic responses.

Also known as: retrieval augmented generation, knowledge-grounded AI, document-grounded LLM, contextual AI

Multi-agent System

An AI architecture where multiple specialised agents collaborate to complete a complex task that no single agent could handle alone.

Each agent has a specific role — one researches, one writes, one reviews, one publishes. An orchestrator coordinates them, passing outputs from one to the next and resolving conflicts.

Example: A marketing team runs a multi-agent content system: Agent 1 researches topics, Agent 2 drafts the article, Agent 3 checks brand tone, Agent 4 generates social posts, Agent 5 schedules publication.

Also known as: multi-agent orchestration, agent swarm, collaborative AI agents, agent network, agentic pipeline

Context Window

The maximum amount of text an AI model can process at once — determining how much information an agent can consider when making a decision.

Think of the context window as the AI's working memory. Everything the agent needs — the user's request, previous messages, retrieved documents, instructions — must fit within it. Larger context windows enable agents to handle longer documents and more complex tasks.

Example: An AI agent reviewing a 200-page annual report needs a large enough context window to hold the entire document while answering questions about it.

Also known as: token limit, model memory, input window, prompt window

Vector Database

A database that stores data as mathematical representations enabling AI agents to find semantically similar information instantly — powering fast, accurate knowledge retrieval.

When you ask an AI agent a question, it needs to find the most relevant information from potentially millions of documents. A vector database converts content into mathematical vectors for millisecond semantic retrieval — regardless of exact keyword matches.

Example: A legal AI agent uses a vector database containing all company contracts. When asked which agreements include automatic renewal clauses, it retrieves relevant sections from hundreds of contracts in under a second.

Also known as: semantic database, embedding database, AI memory store, similarity search database

Agent Loop

The iterative cycle an AI agent follows: perceive the current state, reason about the best next action, execute that action, observe the result, and repeat until the goal is achieved.

Also known as: reasoning loop, action loop, perceive-reason-act cycle, ReAct loop

Tool Calling

The ability of an AI model to invoke external tools or functions during a conversation to retrieve information or take actions, then incorporate the results into its response.

Also known as: function calling, tool use, API calling, external tool integration

Grounding

The process of connecting an AI agent's outputs to verified, real-world information — preventing hallucinations by anchoring responses in actual data sources.

Also known as: factual grounding, knowledge grounding, source grounding, data anchoring

Agentic Pipeline

A sequence of connected AI agent steps where the output of one step becomes the input of the next, enabling complex multi-stage workflows to be automated end-to-end.

Also known as: agent chain, agent workflow, sequential agent execution

Agent Escalation

The process by which an AI agent recognises it cannot handle a situation within its defined scope and transfers the task to a human or a more capable agent.

Also known as: human escalation, agent handoff, exception routing, intelligent escalation

Agent Observability

The ability to monitor, trace and understand what an AI agent is doing in real time — including which tools it called, what decisions it made and why.

Also known as: agent monitoring, agent tracing, AI observability, agent logging

Sandboxing (AI Agent)

Isolating an AI agent in a controlled environment where it can act without affecting live systems — used for testing and validation before production deployment.

Also known as: AI sandbox, agent testing environment, safe execution environment

Subagent

A specialised AI agent that operates within a larger multi-agent system, receiving a specific sub-task from an orchestrator and returning its result.

Also known as: worker agent, specialist agent, child agent, delegate agent

AI Governance

The framework of policies, controls and audit mechanisms that ensures AI systems operate safely, ethically and within defined boundaries in an organisation.

Without governance, AI agents can take unauthorised actions, expose sensitive data or make decisions that violate regulations. AI governance defines what agents can do, logs everything they do, and ensures humans retain oversight of high-stakes decisions. It is non-negotiable for enterprise AI deployment.

Example: A finance AI agent is governed by rules that prevent it from initiating any payment above €10,000 without CFO approval, log every transaction in an immutable audit trail and alert compliance if it accesses data outside its scope.

Also known as: AI oversight, AI compliance, responsible AI deployment, AI risk management

Human-in-the-loop

A design where humans retain approval authority over specific agent actions, ensuring critical decisions always receive human validation before execution.

Not every AI action should be fully autonomous. Human-in-the-loop systems define thresholds — above a certain value, risk level or sensitivity, the agent pauses and requests human approval before proceeding.

Example: A procurement AI agent automatically approves supplier invoices under €5,000. For invoices above that threshold, it drafts the approval request and routes it to the CFO — who approves with one click.

Also known as: human oversight, human approval workflow, supervised AI, human-AI collaboration

AI Literacy

The ability to understand what AI can and cannot do, how to use it effectively and how to identify high-value opportunities to apply it in a business context.

AI literacy is not about knowing how to code. It is about understanding AI's capabilities and limitations well enough to make informed decisions about where to deploy it, how to evaluate its outputs and how to manage the risks.

Example: After a PAATCH AI Workshop, Celio's 70 employees could identify which daily tasks were strong AI automation candidates, articulate what data the agents would need and flag decisions that should remain human.

Also known as: AI fluency, AI readiness, AI skills, digital literacy, AI competency

AI ROI

The measurable return generated by an AI deployment — calculated as value created (time saved, costs reduced, revenue generated) minus the cost of building and running the AI system.

AI ROI is measured in hours of manual work eliminated, error rates reduced, processing volumes increased and revenue cycles accelerated. Defining ROI metrics before deployment determines whether an investment is justified and scalable.

Example: A finance team spending 40 hours/month on bank reconciliation deploys an AI agent completing the same work in 2 hours. At €80/hour that is €3,040 saved per month — €36,480 per year. Payback period: 5 months.

Also known as: AI return on investment, AI business value, AI impact measurement, AI cost savings

AI Use Case

A specific, defined application of artificial intelligence that solves a real business problem with a measurable impact on time, cost or quality.

The best use cases combine high task volume, clear inputs and outputs, access to relevant data and significant time currently spent by humans. Identifying and prioritising use cases is the critical first step before any AI deployment.

Example: After mapping Celio's workflows, PAATCH identified 10 priority AI use cases ranked by ROI and feasibility — including automated stock replenishment alerts, supplier follow-up sequences and monthly sales reporting.

Also known as: AI application, AI opportunity, automation use case, AI deployment scenario

AI Workshop

A structured, hands-on session where enterprise teams learn to use AI tools, identify automation opportunities and build a prioritised roadmap of AI use cases.

An AI Workshop is not a conference or a presentation. Participants actively use AI tools on real business problems. The output is a ranked list of AI use cases ready to move into deployment. Workshops are most effective after a diagnostic phase.

Example: PAATCH ran 3 AI Workshops for Celio across 70 employees, covering Gemini, n8n and Make. Result: 10 priority AI applications identified and ranked, ready for deployment planning.

Also known as: AI training session, AI upskilling workshop, AI hackathon, AI discovery session

AI Strategy

A structured plan that defines how an organisation will adopt, deploy and scale artificial intelligence to achieve specific business objectives over time.

Also known as: AI roadmap, AI transformation plan, enterprise AI strategy, AI adoption plan

AI Maturity

A measure of how advanced an organisation is in its AI adoption journey, from initial experimentation to scaled, governed, production-grade deployment.

Also known as: AI readiness, AI adoption stage, digital maturity, AI capability level

Responsible AI

The practice of designing, developing and deploying AI in a way that is ethical, transparent, fair, safe and accountable to individuals and society.

Also known as: ethical AI, trustworthy AI, AI ethics, safe AI deployment

AI Automation

The use of AI to execute business tasks — including complex, variable and judgement-based ones — without human intervention. Unlike traditional automation, AI automation handles variability and unstructured data.

Also known as: intelligent automation, cognitive automation, AI-powered automation, hyperautomation

Change Management (AI)

The structured process of preparing, equipping and supporting people through the adoption of AI tools and workflows to ensure lasting, successful behavioural change.

Also known as: AI adoption management, AI transformation, people-centred AI, AI cultural change

AI Center of Excellence (CoE)

A dedicated team within an organisation responsible for defining AI standards, governance frameworks, best practices and supporting business units in AI adoption.

Also known as: AI competency center, AI governance team, digital innovation center

AI Upskilling

Training employees to effectively use AI tools in their daily work — not to become AI engineers, but to become capable, confident AI users who identify and leverage AI opportunities.

Also known as: AI training, AI reskilling, AI capability building, workforce AI readiness

Process Mining

The analysis of event logs from business systems to discover, monitor and improve real-world processes. Process mining identifies which processes are best suited for AI automation.

Also known as: process discovery, workflow analysis, process intelligence, event log analysis

Shadow AI

The use of AI tools by employees without official approval from IT or management, creating security, compliance and data governance risks.

Also known as: rogue AI, unsanctioned AI, employee AI usage

API (Application Programming Interface)

A set of protocols that allows different software systems to communicate. AI agents use APIs to connect to external tools, databases and services — APIs are the hands of an AI agent, enabling real-world actions.

Also known as: application programming interface, API integration, AI connector, system integration

Webhook

A method for one application to automatically send data to another when a specific event occurs. Used by AI agents to trigger actions in real time across connected systems.

Also known as: event trigger, HTTP callback, real-time notification

Data Pipeline

A series of automated steps that collect, process and move data from one system to another. AI agents rely on data pipelines to access the information they need to operate.

Also known as: ETL pipeline, data flow, data integration pipeline

Microservices

An architectural approach where a software application is built as a collection of small, independent services that communicate via APIs. Enables scalable, flexible AI deployments.

Also known as: microservice architecture, service-oriented architecture, SOA

LLMOps

The operational practices and tools for managing large language models in production, including monitoring, versioning, evaluation and cost optimisation.

Also known as: LLM operations, MLOps for LLMs, model operations, AI operations

No-code AI

AI tools and platforms that allow non-technical users to build, configure and deploy AI-powered workflows and applications without writing any code. Examples include Make and n8n.

Also known as: visual AI, citizen developer AI, low-code AI, drag-and-drop AI

Semantic Search

Search technology that understands the meaning and intent behind a query rather than matching exact keywords — enabling AI agents to find relevant information even when the exact words differ.

Also known as: meaning-based search, intent search, conceptual search, vector search

Event-Driven Architecture

A software design pattern where actions are triggered by events (a file uploaded, a form submitted, a threshold crossed). AI agents in event-driven architectures react instantly to business events.

Also known as: event-based architecture, reactive architecture, trigger-based automation

Data Quality

The accuracy, completeness, consistency and timeliness of data used by AI agents. Poor data quality is the most common cause of AI agent failures in production.

Also known as: data integrity, data accuracy, data reliability, data hygiene

Real-time Processing

Processing data and triggering AI agent actions immediately as events occur, with minimal latency. Essential for customer-facing AI applications and time-sensitive business processes.

Also known as: stream processing, real-time inference, live processing, instant processing

GDPR

The European Union regulation governing how organisations collect, store and process personal data. All AI deployments handling EU citizen data must comply with GDPR, including AI agents that process employee or customer data.

Also known as: General Data Protection Regulation, EU data law, data privacy regulation

EU AI Act

The European Union's comprehensive regulation for artificial intelligence, establishing risk-based requirements for AI systems deployed in the EU. High-risk AI systems require rigorous documentation, testing and human oversight.

Also known as: AI regulation, European AI law, AI compliance framework

Audit Trail

A chronological record of all actions taken by an AI agent — what it did, when, with what data and what outcome. Essential for compliance, debugging and governance.

Also known as: action log, activity log, compliance trail, agent history

Prompt Injection

A type of attack where malicious instructions are embedded in user inputs to manipulate an AI agent into performing unintended or harmful actions. A key security threat for enterprise AI.

Also known as: AI jailbreak, adversarial prompt, input manipulation, LLM attack

Zero Trust

A security model that requires strict verification of every user and device attempting to access resources, regardless of whether they are inside or outside the network perimeter.

Also known as: zero trust security, never trust always verify, ZTA

SOC 2

A security compliance framework that verifies an organisation's controls for data security, availability, processing integrity, confidentiality and privacy. Increasingly required by enterprise clients evaluating AI vendors.

Also known as: SOC2, Service Organization Control 2, security certification

AI Transparency

The ability to understand and explain how an AI system reaches its decisions, outputs or recommendations. A legal and ethical requirement for many enterprise AI applications.

Also known as: AI explainability, model interpretability, decision transparency, XAI

Claude (Anthropic)

A large language model developed by Anthropic, designed with a focus on safety, reliability and helpfulness. One of the leading AI models used in enterprise applications and integrated into the PAATCH platform.

Also known as: Claude AI, Anthropic model, Claude Opus, Claude Sonnet

Gemini (Google)

Google's multimodal AI model family capable of processing text, images, audio and video. Available in several sizes optimised for different tasks. Integrated into the PAATCH ecosystem.

Also known as: Google Gemini, Google AI, Gemini Pro, Gemini Ultra

Make

A no-code automation platform that allows users to build automated workflows connecting hundreds of apps and services. A key tool in the PAATCH ecosystem for building AI-powered automations.

Also known as: Make.com, Integromat, workflow automation, no-code automation

Gamma

An AI-powered presentation and document creation tool that generates professional slides and documents from text prompts. Part of the PAATCH AI tools ecosystem.

Also known as: Gamma.app, AI presentations, AI slides

MuleRun

An integration and automation platform used to connect enterprise systems and data sources. Part of the PAATCH technology ecosystem for enterprise AI deployments.

Also known as: MuleRun integration, enterprise connector

Cursor

An AI-powered code editor that enables developers to write, edit and debug code using natural language. Popular among PAATCH builders for accelerating AI agent development.

Also known as: AI code editor, AI IDE, developer AI tool, coding assistant

PAATCH Platform

An Agentic AI Orchestration Platform that connects the best AI technologies, elite builders and enterprise operators in a single governed environment.

Organisations use PAATCH to explore, build and deploy production-grade AI agents. The platform provides access to 20+ AI tools, a community of 2000+ vetted builders and enterprise governance built-in. PAATCH is a platform, not an agency — you access it, use it, deploy from it.

Example: Clients including Celio, Eramet, Damart, Publicis Groupe, Celonis and Alinea use PAATCH to deploy AI agents across Sales, HR, Finance, Marketing, Operations and Customer Support.

Also known as: AI agents hub, agentic AI platform, enterprise AI platform

AI Workshop (PAATCH)

A structured hands-on session for enterprise teams that combines AI literacy building, tool exploration and structured use case identification.

Workshops follow a diagnostic phase that maps existing processes. Teams work hands-on with AI tools on real business problems. The output is a prioritised roadmap of AI applications ranked by ROI and feasibility.

Example: PAATCH ran 3 workshops for Celio with 70 employees across 3 sessions, covering Gemini, n8n and Make — identifying 10 priority AI use cases ready for deployment.

Also known as: AI training session, AI immersion day, AI upskilling workshop

AI Agent Template (PAATCH)

A pre-built, pre-configured agent designed for a specific business function — ready to deploy and customise in hours rather than building from scratch in weeks.

Templates package all the prompt engineering, tool connections, data integrations and governance configurations for a specific use case. Teams deploy immediately and customise to their specific context.

Example: A company deploys PAATCH's HR Agent template in two days — connecting it to their ATS and customising screening criteria. The same deployment built from scratch would have taken 6 weeks.

Also known as: pre-built AI agent, agent starter kit, AI agent blueprint, ready-made agent

Playground (PAATCH)

A live experimentation environment on the PAATCH platform where builders and enterprise teams can test AI agents, explore tools and prototype use cases in a real setting.

Also known as: AI sandbox, experimentation environment, PAATCH Playground

PAATCH Builder Community

A vetted network of 2000+ AI developers, automation specialists, agent architects and LLM integration experts accessible to enterprise clients through the PAATCH platform.

Also known as: AI builder network, PAATCH community, vetted AI builders

AI Agents Hub

The core of the PAATCH platform, connecting AI technologies on one side and elite AI builders on the other, with enterprise clients at the centre — a single governed environment for agentic AI.

Also known as: PAATCH hub, AI orchestration center