AI Glossary

The repeating cycle an AI agent follows: plan what to do, execute a step using tools, observe the result, decide whether to continue or adjust.

Using AI agents to write, test, and debug code autonomously — planning multi-step development tasks and executing them with minimal human intervention.

A multi-step process where AI agents autonomously plan, execute, and refine work — going beyond simple prompt-response to handle complex tasks end-to-end.

An AI system that can take actions autonomously — browsing the web, running code, calling APIs, and completing multi-step tasks with minimal human intervention.

The study and practice of ensuring AI systems are developed and used in ways that are fair, transparent, safe, and respectful of human rights and values.

The policies, processes, and frameworks that guide how an organisation develops, deploys, and manages AI systems — covering risk, ethics, compliance, and accountability.

The ability to understand, use, and evaluate AI tools effectively. Not coding — knowing what AI can do, when to use it, and how to get good results.

An organisation's preparedness to adopt and benefit from AI — covering data quality, skills, processes, culture, and infrastructure.

An organisation that designs its processes, products, and culture around AI as a core capability rather than treating AI as an add-on to existing operations.

A way for software to communicate with other software. APIs are how developers connect AI capabilities to websites, apps, and business tools.

Software that can perform tasks that normally require human intelligence, such as understanding language, recognising patterns, and making decisions.

Using technology to perform tasks without manual human effort. AI automation goes beyond traditional rule-based automation by handling unstructured tasks like writing, analysis, and decision-making.

A technique where you ask the AI to explain its reasoning step by step before giving a final answer. This dramatically improves accuracy on complex tasks.

An AI task that assigns input to predefined categories. Spam detection, sentiment analysis, and image recognition are all classification tasks.

The field of AI that enables machines to interpret and understand visual information from images and videos, including object recognition, scene understanding, and visual analysis.

The maximum amount of text an AI can process at once. Think of it as the AI's working memory — everything it can see and consider when generating a response.

A subset of machine learning that uses neural networks with many layers to learn complex patterns. The 'deep' refers to the number of layers, not the depth of understanding.

The process of using digital technology — including AI — to fundamentally change how a business operates, delivers value, and competes. Not just adopting new tools but rethinking processes.

A numerical representation of text (or images, audio, etc.) that captures its meaning. Embeddings let AI measure how similar two pieces of content are.

A prompting technique where you provide the AI with a few examples of the desired input-output pattern before giving it your actual task. This dramatically improves output consistency.

Training an existing AI model on your specific data to improve its performance on your specific tasks. Like giving the AI specialised on-the-job training.

AI that creates new content — text, images, code, audio, video — rather than just analysing or classifying existing data.

A git feature that creates an additional working directory linked to the same repository, allowing multiple branches to be checked out simultaneously — essential for parallel agent development.

A specialised processor originally designed for rendering graphics but now essential for training and running AI models. GPUs can perform thousands of calculations simultaneously.

Constraints, rules, and safety mechanisms built into AI systems to prevent harmful, incorrect, or out-of-scope outputs and actions.

When AI generates confident but incorrect information. The AI is not lying — it is producing statistically plausible text that happens to be wrong.

A system design where AI handles execution but a human reviews, approves, or intervenes at critical decision points before actions are taken.

The process of an AI model generating output from your input. Every time you send a prompt and get a response, that is inference.

A type of AI trained on vast amounts of text to understand and generate human language. ChatGPT, Claude, and Gemini are all LLMs.

The time delay between sending a request to an AI model and receiving the first part of the response. Lower latency means faster, more responsive AI interactions.

A type of AI where systems learn patterns from data instead of following explicitly programmed rules. The system improves its performance through experience.

A technique where you ask the AI to write or improve prompts — using the AI itself to generate better instructions for the AI, creating a self-improving feedback loop.

An open standard that allows AI models to connect to external tools, data sources, and services through a unified interface. MCP lets AI access real-time data and take actions beyond text generation.

The numerical values inside a neural network that determine how it processes information. Weights are what the model learns during training — they encode its knowledge and capabilities.

An AI architecture where multiple specialised agents collaborate to complete complex tasks, coordinated by an orchestrator or a defined protocol.

AI that can process and generate multiple types of content — text, images, audio, and video — within a single model. Claude, GPT-4o, and Gemini are all multi-modal.

The branch of AI focused on enabling computers to understand, interpret, and generate human language in useful ways.

A computing system loosely inspired by the human brain, made of layers of interconnected nodes that learn to recognise patterns in data.

A manager agent that coordinates multiple specialist agents by planning tasks, delegating work, reviewing results, and delivering the final output.

The total number of adjustable values in an AI model. A model with more parameters can capture more complex patterns but requires more computing power to train and run.

A feature that reuses previously processed prompt content across API calls, reducing latency and cost when the same system prompt or context is sent repeatedly.

A technique where the output of one AI prompt becomes the input to the next, creating a multi-step workflow that produces better results than a single complex prompt.

The skill of writing instructions to AI that consistently produce useful, accurate, high-quality output.

A security vulnerability where malicious text in user input or external data tricks an AI system into ignoring its original instructions and following the attacker's instructions instead.

Automated checkpoints between AI generation and human review that catch specific types of errors — format, factual, tone, completeness, and consistency.

A prompting pattern that combines Reasoning and Acting — the AI thinks through the problem step by step, takes an action (like a search), observes the result, and repeats until the task is complete.

An AI task that predicts a numerical value based on input data. Sales forecasting, price estimation, and demand prediction are all regression tasks.

A machine learning approach where an AI learns by trial and error, receiving rewards for good outcomes and penalties for bad ones. Used to train game-playing AI and to fine-tune LLMs.

The practice of developing and deploying AI in ways that are ethical, transparent, accountable, and aligned with societal values — translating AI ethics principles into operational reality.

A technique that connects AI to your own documents and data so it can answer questions using your specific information, not just its general training.

A prompting technique where you ask the AI to generate multiple independent answers to the same question, then select the most common response — improving accuracy through consensus rather than relying on a single output.

A feature that constrains AI to respond in a specific structured format — typically JSON — making the output reliably parseable by code instead of requiring human interpretation.

An AI agent spawned by another agent to handle a specific sub-task, operating with its own context window and returning results to the parent.

A machine learning approach where the model learns from labelled examples — input data paired with correct answers. The most common type of machine learning in business applications.

A set of persistent instructions given to an AI that shapes its behaviour for an entire conversation. System prompts define the AI's role, tone, rules, and output format.

A setting that controls how creative or conservative AI output is. Low temperature = predictable and focused. High temperature = varied and creative.

The volume of data an AI system can process in a given time period — typically measured in tokens per second or requests per minute. Higher throughput means more work done faster.

The smallest unit of text an AI model processes. Roughly 3-4 characters or three-quarters of a word. AI pricing is typically measured in tokens.

The component that converts text into tokens — the numerical units an AI model processes. Different models use different tokenisers, which affects how they handle text.

The ability of an AI model to interact with external tools — search engines, code interpreters, APIs, databases — to take actions beyond generating text.

A custom AI chip designed by Google specifically for machine learning workloads. TPUs are optimised for the tensor operations that neural networks rely on.

The dataset used to teach an AI model. The quality, size, and composition of training data directly determines what the AI can and cannot do well.

The neural network architecture behind modern AI assistants like ChatGPT and Claude. Introduced in 2017, it processes all words simultaneously using an attention mechanism.

A prompting technique where the AI explores multiple reasoning paths in parallel, evaluates each branch, and selects the most promising solution — mimicking how humans consider alternatives before deciding.

A machine learning approach where the model finds patterns in data without being given correct answers. Used for discovering hidden structure, grouping similar items, and detecting anomalies.

A specialised database designed to store and search embeddings — the numerical representations of text, images, or other data used in AI applications.

A sequence of connected steps that accomplish a specific business task. In AI context, a workflow combines human actions and AI processing to complete work efficiently.

Asking an AI to perform a task without providing any examples — relying entirely on the model's training and your instructions. The simplest prompting approach.