The agentic-AI terms AIMOCS uses, defined.

AI operator

An AI operator runs one business workflow end-to-end as a managed service. It is productized agentic software with a service wrapper around it — model, tools, guardrails, monitoring, escalation, on-call. You hire an operator the way you hire a contractor: for the outcome, not the toolset.

Think of it as a contractor that does one workflow forever, not a tool you hand to your team.

Agentic workflow

A sequence of decisions and actions a software agent executes on its own — perceiving the state of a system, choosing among options, and taking actions through tools, with humans only stepping in when a decision crosses a defined threshold. Agentic workflows are the building blocks AIMOCS uses to compose an operator.

AI agent

A software primitive that perceives, decides, and acts inside a defined environment. An agent on its own is a building block; turning an agent into something a business can rely on — SLAs, observability, model upgrades, exception handling — is the work of building an operator around it.

MCP (Model Context Protocol)

An open protocol from Anthropic that standardises how an agent exposes tools, data sources, and prompts to an underlying language model. MCP lets the agent call a CRM, a payment gateway, or an internal API through one uniform interface — instead of bespoke glue per tool.

RAG (Retrieval-Augmented Generation)

A pattern where the agent retrieves relevant documents — from a vector store, a search index, or a database — and includes them in its prompt before generating a response. RAG keeps the model grounded on real data without retraining it. AIMOCS uses RAG for everything from knowledge bases to Arabic dialect prompt libraries.

Human-in-the-loop

A control pattern where the operator pauses and routes a decision to a person before continuing — usually because the action crosses a value, sensitivity, or confidence threshold. Done well, it is invisible most of the time and decisive when it matters. Done poorly, it becomes an alert backlog the team ignores.

RPA (and why agents are different)

Robotic Process Automation drives software through its UI as if a person were clicking. RPA breaks the moment a UI changes. AI agents work through APIs and tools and reason about unstructured input — emails, voice, free-form requests — which is where RPA has always failed.

Tool calling

A capability of modern language models to invoke external functions — read a database, send a message, write a file — by emitting a structured request that runtime code executes. Tool calling is what turns a chat model into something that can take action in the real world.

Vision 2030 AI

Saudi Arabia's national programme to embed artificial intelligence across the public and private sectors by 2030, led by SDAIA and aligned to the Kingdom's broader transformation agenda. AIMOCS builds for businesses that need to participate in this transition operationally — not pitch-deck — through agents, integrations, and KSA-native infrastructure.

ZATCA Phase 2 (Fatoora integration)

The integration phase of Saudi Arabia's e-invoicing regime, requiring every B2B invoice to be issued as a UBL 2.1 XML document, cryptographically stamped, and cleared in real time by the ZATCA Fatoora portal. Wave 24 (revenue threshold SAR 375,000) faces a 30 June 2026 integration deadline.

Nafath

Saudi Arabia's national digital identity service. Apps that want to verify a user's identity send the request to Nafath; the user approves through their Absher-linked authenticator. AIMOCS wires Nafath into agents that need verified citizen identity for onboarding, contracts, or regulated workflows.

ALLaM

The sovereign large language model family developed by SDAIA, built for Arabic-first reasoning and Saudi-specific context. Useful when a deployment requires in-kingdom inference and authentic Arabic linguistic patterns. AIMOCS evaluates ALLaM alongside Claude and GPT for engagements where data residency or dialect performance is the constraint.