Glossary of Terms

Agent memory is how an agent retains information across steps or sessions. Short-term memory holds the current task context; long-term memory persists facts, preferences, and past outcomes in a vector or structured store for later retrieval.

Agentic AI

Agentic AI describes systems where AI models take actions to reach a goal rather than only producing text. An agent plans, calls tools, observes results, and iterates — operating with a degree of autonomy instead of answering one prompt at a time.

The practical difference from a chatbot is the loop. A chatbot responds once; an agent decides what to do next, executes it, checks the outcome, and continues until the goal is met or a stop condition fires.

Agentic AI

An AI agent is a program that uses a language model to decide which actions to take toward a goal. It has access to tools, memory of prior steps, and a policy governing when to act autonomously versus escalate to a human.

Agentic AI

AI governance is the set of policies and controls determining how AI systems are approved, monitored, and held accountable — covering risk classification, human oversight, audit logging, and defined ownership for model behaviour in production.

Governance & Compliance

API-first means designing and agreeing the API contract before building the implementation or interface. Every consumer — including AI agents — works against the same documented surface rather than logic embedded in a UI.

Engineering

Attack surface management continuously discovers and monitors everything an organisation exposes to the internet — domains, APIs, cloud assets, forgotten infrastructure — and prioritises what an attacker would most plausibly exploit.

Security

Autonomous operations describes business processes executed end-to-end by software agents without routine human intervention. Humans set policy and handle exceptions; the system handles the standard path, monitoring itself and escalating when confidence drops.

Agentic AI

Continuous integration and continuous delivery automate building, testing, and releasing software. Changes are integrated and verified frequently so defects surface early and releases become routine rather than high-risk events.

Engineering

Cloud security posture management continuously inspects cloud configuration for misconfigurations and policy violations — public storage, over-permissive roles, unencrypted volumes — and flags or auto-remediates them against benchmarks like CIS.

Security

Cloud-native describes software designed for elastic cloud infrastructure from the outset — containerised, independently deployable, horizontally scalable, and resilient to individual instance failure rather than assuming fixed servers.

Cloud & Infrastructure

Computer vision is the field of extracting information from images and video — detecting objects, reading text, classifying defects, tracking movement. In industry it underpins quality inspection, document processing, and physical safety monitoring.

AI & ML

Data residency is the requirement that data be stored and processed within a specific country or region. It commonly applies to government, healthcare, and financial workloads, and shapes where models and databases may be hosted.

Governance & Compliance

Decentralized finance provides financial services — lending, exchange, derivatives — through smart contracts rather than intermediaries. Users retain custody of assets, and protocol logic rather than a firm enforces the rules.

Blockchain

DevSecOps embeds security checks into the software delivery pipeline — dependency scanning, static analysis, secret detection, container scanning — so vulnerabilities surface at commit time rather than in a pre-release audit.

Security

Digitalization is the redesign of business operations around software and data rather than paper or manual coordination. It differs from digitisation, which only converts existing records into digital formats without changing the process.

Cloud & Infrastructure

Edge computing processes data near where it is produced rather than sending everything to a central cloud. It reduces latency and bandwidth cost, and keeps operating when connectivity is intermittent — important for IoT and industrial systems.

Cloud & Infrastructure

An embedding is a numeric vector representing the meaning of a piece of text or an image. Items with similar meaning sit close together in vector space, which is what makes semantic search and clustering possible.

AI & ML

Endpoint detection and response continuously monitors laptops, servers, and workstations for malicious behaviour, records forensic telemetry, and enables remote containment. It detects behaviour rather than relying only on known malware signatures.

Security

The EU AI Act is European legislation regulating AI by risk tier. High-risk systems face requirements covering data quality, documentation, human oversight, transparency, and post-market monitoring, with obligations falling on providers and deployers.

Governance & Compliance

Extended detection and response correlates telemetry across endpoints, network, identity, and cloud into a single detection layer. Correlating signals that look benign individually is what surfaces multi-stage attacks that single-domain tools miss.

Security

A fallback policy defines what an agent does when it cannot complete a task confidently — retry with a different approach, hand off to another agent, or escalate to a human. It prevents silent failure and unbounded retry loops.

Agentic AI

Fine-tuning continues training a pretrained model on a narrower dataset so it adopts a specific style, format, or domain vocabulary. It changes model weights, unlike RAG which changes only the context supplied at inference time.

AI & ML

FinOps is the practice of managing cloud spend as an engineering concern — attributing costs to teams and features, rightsizing resources, and making cost visible at the point where architectural decisions are made.

Cloud & Infrastructure

The General Data Protection Regulation governs personal data of people in the EU. It establishes lawful bases for processing, data subject rights including access and erasure, breach notification timelines, and constraints on international transfers.

Governance & Compliance

A hallucination is a confident but false statement produced by a language model. It arises because models predict plausible text rather than verify facts, which is why grounding, retrieval, and citation checking matter in production systems.

AI & ML

Headless architecture separates business logic and data from the presentation layer, exposing everything through APIs. Any interface — web, mobile, or an AI agent — consumes the same backend, which is a precondition for agentic automation.

Cloud & Infrastructure

HIPAA is US legislation governing protected health information. For software it drives requirements around encryption, access control, audit logging, breach notification, and business associate agreements with any vendor touching patient data.

Governance & Compliance

Human-in-the-loop means a person reviews or approves specific agent decisions before they take effect. It is typically triggered by confidence thresholds, monetary limits, or regulatory categories rather than applied to every action.

Agentic AI

Inference is running a trained model to produce an output. In production it is the recurring cost and latency centre of an AI system, which is why batching, caching, and model-size choices materially affect unit economics.

AI & ML

Infrastructure as code defines servers, networks, and cloud resources in version-controlled configuration files rather than manual console changes. Environments become reproducible, reviewable, and rebuildable from source.

Cloud & Infrastructure

The Internet of Things refers to networks of physical devices with sensors and connectivity that report data and can be controlled remotely. Enterprise IoT covers asset tracking, condition monitoring, and industrial control.

Cloud & Infrastructure

ISO/IEC 27001 is the international standard for information security management. It requires a risk-based management system with documented controls covering access, cryptography, operations, suppliers, and incident response.

Governance & Compliance

ISO/IEC 42001 is the international standard for an AI management system. It specifies how an organisation should govern AI across its lifecycle — risk assessment, impact analysis, controls, and continual improvement — and is certifiable.

Governance & Compliance

Kubernetes is an open-source system for deploying, scaling, and managing containerised applications. It handles scheduling, service discovery, rolling updates, and self-healing across a cluster of machines.

Cloud & Infrastructure

Know Your Customer and Anti-Money Laundering are the identity verification and transaction monitoring obligations placed on regulated financial platforms. In crypto they extend to wallet screening and sanctions checks before transactions settle.

Blockchain

A large language model is a neural network trained on large text corpora to predict likely continuations of a sequence. This general capability lets it summarise, classify, translate, write code, and reason over text without task-specific training.

AI & ML

Microservices is an architecture that splits an application into small services owning distinct capabilities, each deployable independently. It improves team autonomy and scaling granularity at the cost of added operational and networking complexity.

Cloud & Infrastructure

A minimum viable product is the smallest release that delivers real value and tests the core assumption behind a product. Its purpose is learning from actual usage, not shipping a reduced version of the eventual feature set.

Engineering

The Model Context Protocol is an open standard for connecting AI models to external tools and data sources. It gives agents a consistent interface for discovering and invoking capabilities instead of requiring bespoke integration per system.

AI & ML

Multi-agent orchestration coordinates several specialised AI agents working on one problem. An orchestrator assigns tasks to domain agents, manages message passing between them, resolves conflicts, and decides when the overall job is complete.

It is used when a single agent's context window or skill set is insufficient — for example separating a retrieval agent, a reasoning agent, and a compliance-checking agent so each stays narrow and testable.

Agentic AI

Observability is the ability to understand a system's internal state from its outputs — logs, metrics, and traces. For agentic systems it extends to recording each decision, the tools invoked, and the reasoning that led there.

Engineering

An orchestrator agent is the coordinating layer in a multi-agent system. It decomposes a goal into sub-tasks, routes each to the appropriate specialist agent, aggregates their outputs, and decides whether to iterate or return a result.

Agentic AI

Physical AI refers to AI systems that act in the physical world through robots, vehicles, or connected equipment. It combines perception, planning, and control, and adds safety constraints that purely digital agents do not face.

AI & ML

Prompt injection is an attack where malicious instructions are hidden in content an AI system reads — a web page, document, or email — attempting to override its actual instructions. It is the primary security concern for agents that process untrusted input.

AI & ML

A reasoning model spends additional computation working through a problem before answering, producing intermediate steps rather than responding immediately. This improves accuracy on multi-step tasks such as maths, planning, and complex code changes.

AI & ML

Retrieval-augmented generation retrieves relevant documents from a knowledge base and supplies them to a language model as context before it answers. This grounds responses in current, source-of-truth data rather than only what the model memorised in training.

RAG is usually preferred over fine-tuning when the underlying data changes frequently, when answers must cite sources, or when the corpus is too large to bake into weights.

AI & ML

Role-based access control grants permissions to roles rather than individuals, and assigns people to roles. It simplifies auditing and offboarding, and in agentic systems is used to constrain what actions each agent may perform.

Security

SaaS-to-agentic conversion is the process of restructuring an existing SaaS product so AI agents can operate it. Business logic is decoupled from the interface into an API-first layer that agents call programmatically, turning manual workflows into autonomous ones.

Agentic AI

A SIEM aggregates logs from across an estate, normalises them, and applies correlation rules to surface security events. It provides the long-term searchable record required for investigation and most compliance regimes.

Security

A smart contract is code deployed to a blockchain that executes automatically when its conditions are met. Because deployed logic is typically immutable, correctness must be established before release through testing and audit.

Blockchain

SOC 2 is an audit report on a service organisation's controls across security, availability, processing integrity, confidentiality, and privacy. Type I assesses design at a point in time; Type II tests operating effectiveness over a period.

Governance & Compliance

Sovereign AI describes AI systems deployed under a nation's own control — infrastructure inside its borders, data that does not leave its jurisdiction, and governance under domestic law. It is a common requirement for public-sector deployments.

Governance & Compliance

A strangler fig migration replaces a legacy system incrementally. New functionality is built alongside the old system and traffic is routed across piece by piece, so the legacy platform is retired gradually rather than in one cutover.

Cloud & Infrastructure

Sukuk are Sharia-compliant financial certificates. Rather than paying interest, holders own a share in an underlying asset and receive returns from it — a structure increasingly issued and settled on tokenized infrastructure.

Blockchain

Technical debt is the future cost of choosing an expedient implementation over a sound one. Like financial debt it accrues interest — each subsequent change in the affected area takes longer and carries more risk.

Engineering

Tokenization represents ownership of an asset as a blockchain token. Applied to real-world assets such as property, bonds, or funds, it enables fractional ownership and programmatic transfer subject to the applicable securities regime.

Blockchain

Tool use is an agent's ability to call external functions — querying a database, hitting an API, running code. The model chooses which tool to invoke and with what arguments, then incorporates the result into its next decision.

Agentic AI

The UAE Personal Data Protection Law is the federal framework governing personal data in the Emirates. It sets consent requirements, data subject rights, breach obligations, and conditions for transferring data outside the UAE.

Governance & Compliance

VARA is Dubai's Virtual Assets Regulatory Authority. It licenses and supervises virtual asset activity in the emirate across seven categories including exchange, custody, broker-dealer, and lending, each with its own operational requirements.

Governance & Compliance

A vector database stores text or images as numerical embeddings and retrieves them by semantic similarity rather than keyword match. It is the retrieval layer most RAG systems depend on for finding contextually relevant documents.

AI & ML

Zero trust is a security model that assumes no implicit trust from network location. Every request is authenticated and authorised on its own merits, with least-privilege access enforced continuously rather than granted once at the perimeter.

Security

Zero-downtime deployment releases new software without interrupting service, typically via blue-green or rolling strategies. Traffic shifts to new instances only once they pass health checks, with rollback available if error rates rise.

Cloud & Infrastructure