Reference

The AI Glossary

Plain-English definitions of the concepts behind AI-native organizations, infrastructure, hiring, search and strategy, for executives, not engineers.

AI Infrastructure

Large Language Model (LLM)

A model trained on vast text to predict and generate language, the engine behind tools like ChatGPT and Claude. It powers chat, drafting, extraction and reasoning over natural language.

Retrieval-Augmented Generation (RAG)

A pattern that grounds an LLM in your own, current data by retrieving relevant documents at query time, reducing hallucination and keeping answers up to date without retraining.

Vector Database

A database that stores embeddings (numeric representations of meaning) and finds the most similar items fast, the backbone of semantic search and RAG.

Embedding

A numeric vector that captures the meaning of text (or images) so machines can compare similarity. Embeddings power search, recommendations and RAG retrieval.

Fine-tuning

Further training a model on your examples to change its behavior, style or format. Best for consistent behavior, not for injecting fresh facts (use RAG for that).

Prompt Engineering

The practice of structuring instructions and context so a model reliably produces the output you need. It is craft, not magic, and increasingly systematized.

AI Agent

An LLM-driven system that plans and takes multi-step actions using tools (APIs, search, code). Powerful for bounded, supervised workflows; risky when unsupervised in high-stakes tasks.

Model Context Protocol (MCP)

An open standard for connecting AI models to tools and data sources through a common interface, so agents can use your systems without bespoke integrations for each one.

MLOps

The practices and tooling that make machine-learning models reliable in production, deployment, monitoring, versioning, retraining and cost control.

Inference

Running a trained model to produce an output. Inference cost and latency, not training, dominate the economics of most production AI features.

Token

The unit of text an LLM processes, roughly a word-piece. Usage and pricing are measured in tokens, so prompt and output length directly drive cost.

Context Window

The maximum amount of text (in tokens) a model can consider at once. Larger windows allow more context but cost more and can dilute focus.

Hallucination

When a model produces confident but false or unsupported output. Mitigated with retrieval (RAG), grounding, citations and human review in sensitive contexts.

Guardrails

Controls that constrain what an AI system can say or do, input/output filtering, policy checks, least-privilege tool access, so it behaves safely in production.

Foundation Model

A large, general-purpose model trained on broad data that can be adapted to many tasks, the base you build on rather than train from scratch.

Multimodal AI

Models that work across more than one data type, text, images, audio, video, enabling use cases like document understanding and image-based support.

Human-in-the-Loop

A design where a person reviews or approves AI output before it acts. Essential for high-stakes workflows and a safe way to widen automation over time.

AI Observability

Monitoring the behavior, quality, cost and safety of AI systems in production, logging prompts, outputs, latency and failures, so you can debug and improve them.

Evaluations (Evals)

Systematic tests that measure an AI system's quality on representative tasks. Good evals turn 'it feels better' into evidence and prevent silent regressions.

Agentic Workflow

A process where an AI agent chains multiple steps and tools toward a goal. Works best when bounded, observable and reversible, with a human on high-stakes actions.

The Weekly Brief

Intelligence for building AI-native organizations.

One email a week: the sharpest thinking on AI hiring, infrastructure, teams and strategy, for the people building the future of work.

Join operators, founders and CTOs. No spam, unsubscribe anytime.