Glossary

The agentic AI glossary.

Plain-English definitions of the terms behind agentic AI in financial services — from RAG and MCP to model risk and human-in-the-loop. Each links to a deeper guide.

Browse all resources →

A C E F G H L M P R S V

Agentic AI

Agentic AI is AI that pursues a goal autonomously — perceiving, reasoning, planning, and taking actions across tools and systems, then adapting on the results — rather than only producing an answer.

Agentic RAG is retrieval-augmented generation restructured as a reasoning loop: an agent decides whether to retrieve, what to search for, which source to use, and whether the evidence is good enough — retrying or escalating when it is not.

AI agent

An AI agent is a system that uses a model to plan and take actions toward a goal — calling tools, observing results, and iterating across steps — instead of returning a single response.

Audit trail

An audit trail is a durable, tamper-evident record of how an AI decision was made — the context and sources used, the reasoning, the tool calls and actions taken, the decision, and the human's view — so it can be reconstructed later.

Context window

A context window is the maximum amount of text, measured in tokens, that a language model can consider at once — its prompt plus everything generated so far. Anything outside the window is invisible to the model.

Embedding

An embedding is a numeric vector representation of text (or other data) that captures meaning, so similar items sit close together in vector space. Embeddings let a system retrieve by semantic similarity rather than exact keywords.

EU AI Act

The EU AI Act (Regulation (EU) 2024/1689) is the European Union's comprehensive, risk-tiered law for AI. It classifies systems by risk and imposes obligations — heaviest on 'high-risk' uses such as credit scoring and life/health insurance pricing.

Explainable AI

Explainable AI (XAI) is the ability to state, in human terms, why a model produced a particular output. In lending and insurance it means giving specific, accurate reasons for a decision — not just a score.

Fine-tuning

Fine-tuning is further training of a pre-trained model on a focused dataset to adapt its behavior, style, or skill. Unlike RAG, it changes the model's weights rather than supplying knowledge at query time.

Generative AI

Generative AI is AI that produces new content — text, code, images — in response to a prompt. It answers, drafts, or summarizes; a person then reviews and acts on the output.

Guardrails

Guardrails are the constraints that keep an AI system inside acceptable bounds — input/output filters, policy checks, allowed-action limits, and escalation rules — so it cannot take unsafe or non-compliant actions.

Hallucination

A hallucination is when an AI model produces confident output that is factually wrong or unsupported — inventing a fact, source, or detail that is not grounded in real data.

Human-in-the-Loop (HITL)

Human-in-the-loop (HITL) AI is a design in which a person reviews, approves, or can override an AI system's consequential actions. A risk gate routes low-risk actions to automatic execution and high-risk ones to a human.

Large Language Model (LLM)

A large language model (LLM) is an AI model trained on vast amounts of text to predict and generate language. LLMs are the reasoning engine behind most generative and agentic AI systems.

Model Context Protocol (MCP)

The Model Context Protocol (MCP) is an open standard that lets an AI application connect to external tools and data through one consistent interface — 'a USB-C port for AI' — instead of a custom integration per system.

Model Risk Management (MRM)

Model risk management (MRM) is the discipline of governing the risk that a model is wrong or misused — through inventory, independent validation, ongoing monitoring, and clear ownership. In US banking it derives from SR 11-7.

Prompt engineering

Prompt engineering is the practice of designing the instructions and examples given to a model to steer its behavior at inference, with no training. It is the fastest, cheapest way to improve an AI system's output.

Retrieval-Augmented Generation (RAG)

Retrieval-augmented generation (RAG) connects a language model to an external knowledge source, retrieving relevant information at query time and supplying it as context — so answers are grounded in real, current data rather than only training.

Straight-Through Processing (STP)

Straight-through processing (STP) is the automated handling of a transaction end-to-end without manual intervention. In insurance it means taking a submission or claim to a priced indication or settlement for cases that fall within defined rules.

Vector database

A vector database stores data as embeddings and retrieves it by semantic similarity — finding the items closest in meaning to a query rather than matching keywords. It is the common backbone of retrieval in RAG systems.

Want these applied to your stack?

Book a demo