AI GLOSSARY
Plain-language, sourced definitions of the AI terms used across our Indexes and Reports — from agentic AI to world models. Jump to a letter, or skim the whole set.
Thank you for reading this post, don't forget to subscribe!
Agentic AI
AI that pursues goals autonomously across multiple steps, using tools and feedback with minimal human input.
AI Agent
An autonomous AI that plans, uses tools, and executes multi-step tasks on a user’s behalf.
AI Inference
Running a trained model to produce a response — the cost dropped ~280-fold from 2022 to 2024.
Benchmark
A standardized test — SWE-bench, MMLU, GPQA, HLE — used to compare AI models head to head.
Capex
Capital expenditure — hyperscaler AI capex is projected to top $600B in 2026.
Context Window
How much text (in tokens) a model can consider at once when generating a response.
Embodied AI
AI that perceives and acts in the physical world — humanoid robots, drones, autonomous vehicles.
EU AI Act
The EU’s comprehensive AI law — most obligations apply from August 2026.
Fine-Tuning
Further training a pre-trained model on specialized data to adapt it to a task or domain.
Foundation Model
A large model pre-trained on broad data, then adapted to many downstream tasks.
Frontier Model
The most capable general-purpose models at any given time — OpenAI, Anthropic, Google, DeepSeek.
Generative AI
AI that creates text, images, code, or audio rather than just classifying or ranking.
GPU
Graphics processing unit — the dominant hardware for training and serving modern AI.
Hyperscaler
AWS, Azure, Google Cloud, Meta, Oracle — the cloud giants that host most large-scale AI.
Large Language Model (LLM)
A transformer trained on vast text to predict the next token and generate language.
MLOps
The practices and tooling for deploying, monitoring, and maintaining ML models in production.
Multiagent System
Multiple specialized agents that coordinate — Gartner’s named successor to single-agent AI.
Multimodal AI
AI that understands and generates across text, images, audio, and video in one model.
Retrieval-Augmented Generation (RAG)
Grounding an LLM’s answers in retrieved documents to improve accuracy and reduce hallucination.
Sovereign AI
A nation’s own compute, models, and capital — behind every 2026 national AI strategy.
Tokens
The basic units of text an LLM processes — the unit in which AI usage is priced.
World Model
A model that learns environment dynamics — the lab-language successor to text-only LLMs.
Don’t see a term? It probably appears inside a report — start at the Indexes or use search. Last reviewed June 2026.