AI Glossary

A software system that can perceive its environment, make decisions, and take actions to achieve a specific goal — without needing a human to approve every step. Unlike chatbots that just respond, AI agents plan and execute multi-step tasks autonomously.

AI systems that operate with a high degree of autonomy and goal-directed behavior. Agentic AI can break down complex objectives into sub-tasks, use tools, browse the web, write code, and complete entire workflows independently.

Cloud-based software powered by autonomous AI agents that independently complete complex tasks based on high-level user goals — rather than requiring manual operation through traditional dashboards and buttons.

A set of rules that lets two software applications talk to each other. When an AI tool connects to your CRM or email platform, it does so through an API. Think of it as a waiter taking your order to the kitchen.

Using software to complete repetitive tasks without human input. Unlike AI, traditional automation follows fixed rules. AI automation can handle exceptions and make judgment calls.

An AI chatbot built by OpenAI that uses a large language model (LLM) to generate human-like text responses. Launched in November 2022, it reached 900 million weekly users by early 2026 — the fastest consumer tech adoption in history.

The maximum amount of text an AI model can read and remember at one time. A larger context window means the AI can process longer documents or longer conversations without "forgetting" earlier content.

An AI assistant built by Anthropic, designed to be safe, accurate and helpful. Competes with ChatGPT and Gemini. Known for longer context windows and strong reasoning capabilities.

The process of training a pre-existing AI model on new, specific data to make it better at a particular task. Like taking a general-purpose chef and training them to specialise in French cuisine.

A large AI model trained on massive amounts of data that can be adapted for many different tasks. GPT-4, Claude and Gemini are all foundation models. They form the "base" that other AI products are built on top of.

AI that can create new content — text, images, code, audio or video — rather than just analysing existing content. ChatGPT, Midjourney and GitHub Copilot are all generative AI tools.

The type of AI architecture used by OpenAI's models. Pre-trained means it learned from huge amounts of internet text before you started using it. Transformer is the technical architecture that makes it work.

When an AI model confidently states something that is false. The model doesn't know it's wrong — it generates plausible-sounding text based on patterns, not verified facts. Always double-check AI outputs for critical information.

An AI model trained on billions of words of text that can understand and generate human language. GPT-4, Claude, and Gemini are all LLMs. The "large" refers to the billions of parameters (settings) the model uses.

The set of practices for deploying, monitoring and maintaining LLM-based applications in production. Similar to DevOps but specifically for AI products. Covers prompt management, model versioning, cost tracking and evaluation.

A type of AI where computers learn from data rather than being explicitly programmed with rules. The more data it sees, the better it gets. Most modern AI — including LLMs — is built on machine learning.

AI that can process and generate multiple types of content — text, images, audio and video — not just one. GPT-4o and Gemini Ultra are multimodal. You can show them a photo and ask questions about it.

Tools that let you build AI-powered products without writing code. Platforms like Zapier, Make, and Bubble now include AI features so non-developers can create automated workflows and AI applications.

The instruction or question you give to an AI model. The quality of your prompt directly affects the quality of the output. "Prompt engineering" is the skill of crafting prompts that get the best results from AI systems.

The internal settings or "weights" of an AI model that it learned during training. More parameters generally means a more capable model. GPT-4 reportedly has over a trillion parameters — that's what makes it so capable.

A technique that gives an AI model access to external documents or databases before generating a response. Instead of relying only on what it learned during training, it can look up current, specific information first — reducing hallucinations.

A training technique where humans rate AI responses and the model learns to produce more of what humans prefer. Used by OpenAI and Anthropic to make their models more helpful, accurate and safe.

Software delivered over the internet on a subscription basis — you don't install it, you just log in. Salesforce, Slack, and Notion are all SaaS products. Most business software today is SaaS.

A set of instructions given to an AI before the conversation starts that shapes how it behaves. When you use a customer service chatbot, the company has written a system prompt telling it to stay on topic, use a certain tone, and follow specific rules.

The basic unit AI models use to process text. A token is roughly ¾ of a word. "Hello world" is about 3 tokens. AI pricing is often calculated per token — so longer conversations cost more to process.

The massive collection of text, images or other data an AI model learns from. Most LLMs trained on large chunks of the public internet, books, and academic papers. The quality and diversity of training data shapes what the model knows.

A database designed to store and search AI-generated numerical representations of text or images (called embeddings). Essential for building RAG systems and AI-powered search. Pinecone and Weaviate are popular vector databases.