# Practical AI and systems glossary

Plain-language definitions connected to Tanveer Riaz’s public work.

## Agent skill

A focused, reusable procedure that gives an AI agent domain instructions, constraints, and a reliable workflow for a specific job.

**Why it matters:** It moves useful behavior out of one-off prompts and into versioned, repeatable practice.

**Related:** Agent harness, Human in the loop

## Agent harness

The tools, context, permissions, memory, feedback loops, and runtime around a model that determine how useful an agent is in practice.

**Why it matters:** Model quality matters, but the harness decides whether work can be inspected, corrected, and shipped.

**Related:** Agent skill, Evals

## Local-first AI

AI software designed to keep models, data, or core workflows on a person's own device whenever practical.

**Why it matters:** It can improve privacy, resilience, latency, and user control - especially for personal or sensitive workflows.

**Related:** Inference, Human in the loop

## RAG

Retrieval-augmented generation: finding relevant source material first, then giving it to a model as context for an answer.

**Why it matters:** It grounds answers in a defined body of evidence instead of relying only on model memory.

**Related:** Evals, Data provenance

## MCP

Model Context Protocol: a standard way for AI applications to discover and use tools, data sources, and reusable context.

**Why it matters:** It reduces bespoke integration work and makes agent capabilities easier to compose.

**Related:** Agent harness, Tool use

## Evals

Repeatable tests that measure whether an AI system behaves well enough on the tasks and failure modes that matter.

**Why it matters:** A convincing demo is not evidence of dependable behavior; evaluation makes quality discussable and trackable.

**Related:** Human in the loop, Data provenance

## Human in the loop

A workflow that deliberately gives a person review, correction, or approval control at consequential steps.

**Why it matters:** It keeps automation proportional to risk and creates a place for judgment when rules or models are uncertain.

**Related:** Evals, Agent harness

## Data provenance

A clear record of where data came from, how it changed, and what claims it can legitimately support.

**Why it matters:** Without provenance, polished interfaces can make invented or stale data look authoritative.

**Related:** RAG, Evals

## ISO 20022

A global financial messaging standard that gives payments and related business processes a richer, structured data model.

**Why it matters:** It illustrates the systems discipline behind regulated change: common semantics, complex migrations, and many connected parties.

**Related:** Systems delivery, Data provenance

## Vibe coding

Building software through rapid natural-language collaboration with coding agents, with the human steering intent, taste, and acceptance.

**Why it matters:** It speeds up making, but still demands product judgment, verification, security, and ownership of the result.

**Related:** Agent harness, Evals