AI & Agents
AI Agent
An AI agent is a software system that pursues a goal by taking actions, not just producing text.
Agentic AI
Agentic AI is the general term for AI systems that plan and act autonomously toward a goal rather than responding to one prompt at a time.
Multi-Agent System
A multi-agent system is an architecture where several specialised AI agents work on one goal, typically coordinated by a supervisor agent that decomposes the task, routes each part to the right specialist, and checks results before accepting them.
Tool Calling
Tool calling is the mechanism that lets a language model invoke external functions — querying a database, calling an API, sending an email.
RAG (Retrieval-Augmented Generation)
RAG is a technique where an AI system retrieves relevant passages from your own documents at query time and generates an answer grounded in them, with citations.
Vector Database
A vector database stores text, images or other content as numerical embeddings and retrieves items by semantic similarity rather than exact keyword match.
Embedding
An embedding is a list of numbers representing the meaning of a piece of content, produced by a model so that semantically similar items sit close together in the resulting space.
MCP (Model Context Protocol)
MCP is an open standard for exposing tools and data to AI agents through a consistent interface.
Human-in-the-Loop (HITL)
Human-in-the-loop is a design pattern where an AI system pauses and requires a person to approve, modify or reject an action before it proceeds.
AI Guardrails
Guardrails are the constraints that limit what an AI system is permitted to do, enforced outside the model rather than requested inside a prompt.
Prompt Injection
Prompt injection is an attack where text placed in content the model reads — a document, a web page, an email, a tool response — is crafted to look like instructions, causing the system to act against its operator's intent.
Fine-Tuning
Fine-tuning adjusts a model's weights by training it on examples of desired behaviour, changing how it responds rather than what it knows.
Agent Orchestration
Agent orchestration is the layer that decides which AI agent handles which part of a task, in what order, and what happens when one fails.
Agent Evaluation
Agent evaluation is the practice of scoring an AI agent against a fixed set of real tasks with verified expected outcomes, so that every prompt change, model swap or tool update can be measured rather than guessed at.
Semantic Search
Semantic search finds results by meaning rather than by matching words, so a query about "reducing customer churn" can surface a document about "improving retention" even though they share no vocabulary.
Function Calling
Function calling is the mechanism by which a language model emits a structured request — a function name plus JSON arguments — instead of free-form text, so that host code can execute a real operation and return the result.
Context Window
A context window is the maximum amount of text a language model can consider at one time, measured in tokens and shared by the system prompt, conversation history, retrieved documents and tool results.
Agent Memory
Agent memory is the set of mechanisms an AI agent uses to retain information beyond a single model call: short-term memory held in the context window for the current task, and long-term memory written to an external store and retrieved when relevant.
ReAct Pattern
The ReAct pattern — short for Reason + Act — is an agent control loop in which a language model alternates between producing a reasoning step and taking an action: it thinks, calls a tool, reads the observation, and thinks again until the task is complete.
Chain of Thought
Chain of thought is a technique in which a language model produces intermediate reasoning steps before its final answer instead of answering directly.
AI Hallucination
Hallucination is the generation of fluent, confident output that is false or unsupported — invented citations, wrong figures, functions that do not exist.
LoRA (Low-Rank Adaptation)
LoRA — Low-Rank Adaptation — is a parameter-efficient fine-tuning method that freezes a model's original weights and trains small low-rank matrices alongside them.
LLM Observability
LLM observability is the practice of tracing and measuring every model call in production — prompts, completions, tool invocations, latency, token usage and cost — so that non-deterministic systems can be debugged and regressions caught.
Eval Harness
An eval harness is the infrastructure that runs an AI system against a fixed set of inputs with verified expected outcomes and scores the results, turning every prompt change, model swap or tool update into a measurable event.
Autonomous Agent
An autonomous agent is an AI system that carries a multi-step task from goal to completion without per-step human direction — planning, calling tools, checking its own results and deciding when it is done.
Agent Sandboxing
Agent sandboxing is isolating the environment an AI agent executes in — file system, network access, credentials and compute — so that a wrong or manipulated agent cannot reach anything outside its granted boundary.
Inference
Inference is running a trained model on new input to produce output — the serving phase, as opposed to training.
Architecture & Engineering
Microservices
Microservices are an architecture where an application is split into small independent services, each owning one business capability and its own data, communicating over a network.
Monolithic Architecture
A monolithic architecture is an application built and deployed as a single unit, where all functionality shares one codebase, one process and usually one database.
API Gateway
An API gateway is a single entry point that sits in front of backend services and handles concerns common to all of them — authentication, rate limiting, routing, request validation and logging.
Serverless
Serverless is a model where code runs in short-lived instances the cloud provider creates on demand and destroys when idle, billed per invocation rather than per running server.
Containerization
Containerization packages an application together with its dependencies, libraries and configuration into a single image that runs identically on any machine with a container runtime.
Kubernetes
Kubernetes is a system that runs containers across a cluster of machines, deciding where each one should run, restarting them when they fail, scaling them with demand, and routing traffic between them.
CI/CD
CI/CD is the practice of automatically building and testing every code change (continuous integration) and automatically releasing changes that pass (continuous delivery or deployment).
Infrastructure as Code
Infrastructure as code means defining servers, networks, databases and permissions in version-controlled configuration files rather than configuring them by hand.
Blue-Green Deployment
Blue-green deployment runs two identical production environments, with only one serving live traffic.
Load Balancing
Load balancing distributes incoming requests across multiple servers so no single instance becomes a bottleneck, and so the failure of one instance does not take the service down.
Horizontal Scaling
Horizontal scaling adds more machines to handle load, rather than making one machine bigger.
Caching
Caching stores the result of an expensive operation so subsequent requests can reuse it instead of recomputing it.
Message Queue
A message queue holds tasks that one part of a system produces and another consumes later, decoupling the two so the producer does not wait for the work to finish.
Event-Driven Architecture
Event-driven architecture is a design where components publish events describing something that happened, and other components react to them, without the publisher knowing who is listening.
Multi-Tenancy
Multi-tenancy is an architecture where one running application serves many customer organisations, keeping each tenant's data isolated from the others.
SSO (Single Sign-On)
Single sign-on lets users authenticate once with a central identity provider and then access multiple applications without logging in again.
RBAC (Role-Based Access Control)
Role-based access control grants permissions to roles rather than to individuals, and assigns users to roles.
Webhooks
Webhooks are HTTP callbacks that one system sends to another when an event occurs, so the receiver learns about it immediately instead of polling for changes.
Rate Limiting
Rate limiting caps how many requests a client can make in a given period, protecting a service from being overwhelmed by a single caller — whether malicious, misconfigured or simply busy.
Technical Debt
Technical debt is the future cost created by choosing a faster implementation now over a better one.
Core Web Vitals
Core Web Vitals are a set of measurements Google uses to score real-world page experience: how quickly the main content renders, how fast the page responds to interaction, and how much the layout shifts while loading.
GraphQL
GraphQL is a query language and server runtime for APIs in which the client specifies exactly the fields it needs and the server returns exactly that, from a single typed endpoint described by a schema.
Code Review
Code review is the practice of having a second engineer read a change before it merges, checking correctness, design and risk.
Observability
Observability is the ability to explain a system's internal state from its outputs — structured logs, metrics and traces — without deploying new code to find out.
Product & Data
MVP (Minimum Viable Product)
An MVP — minimum viable product — is the smallest version of a product that can be put in front of real users to test the core assumption: that the problem is real and this solves it.
Product-Market Fit
Product-market fit is the point where a product meets a real market need strongly enough that customers adopt it, keep using it and recommend it without being pushed.
Design System
A design system is the shared, governed library of design tokens, components and usage rules that a team builds interfaces from — one canonical implementation of each button, form and modal, consumed everywhere.
Data Pipeline
A data pipeline is the automated sequence of steps that moves data from where it is created to where it is used — extracting from applications and databases, validating and transforming it, and loading it into a warehouse, lake or model.
ETL (Extract, Transform, Load)
ETL — extract, transform, load — is the classic data-integration pattern: pull data from source systems, reshape it into the target schema, then load it into a warehouse.
Data Warehouse
A data warehouse is a central analytical store holding structured, cleaned historical data, optimised for queries that scan and aggregate millions of rows — the system of record for metrics, reporting and dashboards.
Data Lake
A data lake stores large volumes of raw data — structured, semi-structured and unstructured — in native form on cheap object storage, applying schema when data is read rather than when it is written.