The words behind the work

Optimising content so that AI answer engines — such as ChatGPT, Perplexity or Google's AI Overviews — select and quote it when responding to a question, rather than optimising only for a ranked list of links.

A software system that uses a language model to pursue a goal across multiple steps — planning, calling tools and acting — instead of producing a single one-shot reply.

AI designed to take autonomous, multi-step action toward an objective: deciding what to do next, using tools and adapting, rather than only answering prompts.

A process in which one or more AI agents carry out a sequence of steps — research, draft, review, execute — with limited human intervention.

Writing code alongside an AI assistant that suggests implementations, explains errors and drafts changes, while the developer directs and reviews the work.

A defined way for two pieces of software to communicate, exposing specific operations and data without revealing the underlying code.

A prompting technique that encourages a model to reason step by step before answering, which often improves accuracy on complex tasks.

Continuous Integration and Continuous Delivery: an automated pipeline that tests and ships code changes frequently and safely, cutting manual release effort and errors.

Packaging an application with everything it needs to run into an isolated unit — a container, such as Docker — so it behaves the same across different environments.

The practice of deliberately assembling the right information, instructions and tools into a model's context window so it can perform a task reliably — increasingly seen as the discipline succeeding prompt engineering.

The maximum amount of text, measured in tokens, that a language model can consider at once — including the prompt, any retrieved data and its own output.

An AI assistant embedded in a tool — an editor, spreadsheet or design app — that suggests or performs actions alongside the human, who stays in control.

Running computation close to where data is generated or the user is located, rather than in a distant central server, to reduce latency.

Numerical representations of text, images or other data that capture meaning, letting software measure how similar two things are — the basis of semantic search and RAG.

Getting a model to perform a task with a handful of examples in the prompt (few-shot) or none at all (zero-shot), rather than task-specific retraining.

Further training a pre-trained model on a specific dataset so it performs better on a particular task, domain or style.

A large model trained on broad data that can be adapted to many downstream tasks — the general-purpose base that products are built on top of.

The discipline of making a brand or its content more likely to be surfaced, cited and recommended by generative AI systems.

A query language for APIs that lets a client request exactly the data it needs in a single call, instead of fetching fixed responses from several endpoints.

Rules, filters and checks placed around an AI system to keep its behaviour safe, on-topic and within acceptable bounds.

When an AI model produces confident but false or fabricated information — a core reason human review remains necessary.

Separating a site or app's content and back end from its presentation and front end, so the same content can be delivered to many interfaces through APIs.

A property whereby performing the same operation multiple times has the same effect as doing it once — important for safe retries in APIs and payments.

Running a trained model to produce an output — a prediction or generated text — as opposed to training the model in the first place.

The delay between a request and its response. Low latency means a system feels fast and responsive.

A model trained on vast amounts of text that predicts and generates language, powering chatbots, coding assistants and much of the current AI wave.

An open standard that lets AI applications connect to external tools, data sources and services in a consistent, reusable way.

AI that can understand or generate more than one type of data — for example text, images, audio and video together.

The smallest version of a product that delivers real value and can be released to learn from actual users before investing further.

Coordinating multiple components — models, tools, agents or services — so they work together as a single, reliable workflow.

Crafting the wording, structure and examples in a prompt to get more accurate and useful output from a language model.

An attack in which malicious instructions hidden in input trick an AI system into ignoring its rules or leaking data — a key AI security concern.

Generating a working application, or a large part of one, largely from natural-language descriptions rather than hand-writing all the code.

A website built to behave like a native app — installable, fast and able to work offline — using modern web capabilities.

A technique where a model retrieves relevant documents from a knowledge base and uses them to ground its answer, reducing hallucination and adding up-to-date facts.

Capping how many requests a client can make within a period, to protect a service from overload or abuse.

A common style for building web APIs around resources and standard HTTP methods such as GET and POST.

A system's ability to handle growth — more users, data or traffic — without breaking or slowing unacceptably.

A cloud model where code runs on demand without you managing servers; you pay for actual usage and the platform handles scaling.

An emerging approach where a clear written specification guides AI code generation, keeping AI-built software aligned with intent and reviewable.

Building a website's pages into fixed files ahead of time so they load extremely fast and are cheap and secure to host.

The underlying instructions that set an AI assistant's role, rules and behaviour before the user's own messages are considered.

The future cost of choosing a quick or easy solution now instead of a better one; it accrues interest, slowing later work.

A setting that controls how random or creative a model's output is: lower values are more focused and deterministic, higher values more varied.

The unit of text a language model reads and generates — roughly a word-piece. Pricing and context limits are measured in tokens.

Uptime is the proportion of time a service is available; an SLA (Service Level Agreement) is the promised standard for that availability.

A database optimised for storing embeddings and finding the most similar items quickly — a common backbone for RAG and semantic search.

A system — most commonly Git — that tracks changes to code over time, enabling collaboration, history and safe experimentation.

Building software by describing what you want in natural language and letting an AI generate the code, guiding it through iteration rather than writing most lines by hand. The term entered wide use in 2025.

A way for one system to notify another in real time by sending an automated HTTP message when an event happens.