Glossary
Agentic AI, in plain English.
Short, practical definitions of the terms enterprise leaders meet when they evaluate autonomous AI systems.
Agentic AI
AI that pursues a goal across multiple steps: it plans, uses tools and software, checks its own work and adapts, instead of producing a single reply to a single prompt. In business, agentic AI is what lets a system carry a task such as reconciling invoices or resolving a customer request from start to finish.
AI agent
A single software component, built on an AI model, that can take actions toward a goal using tools. An agent is a building block. An autonomous business function usually combines several of them with integrations, approvals and monitoring. See what comes after agents.
Agentic AI vs AI agents
Agentic AI describes the capability and approach: systems that act with some independence toward outcomes. AI agents are the individual units that implement it. A chatbot that only answers questions is neither; an agent that reads an invoice, matches it to a purchase order and prepares the payment for approval is both.
Autonomous organization
An organization in which most routine operational work is carried out by governed AI systems working together across functions, while people set direction, handle judgement and audit results. It is a direction of travel, not a switch. See our research direction.
Levels of autonomy
A scale describing how much authority an AI system has. We use five: L1 Assist, L2 Execute on approval, L3 Operate within limits, L4 Run the function and L5 Coordinate the organization. Authority is set per function and per action type and earned through evidence. Read autonomy is a dial, not a switch.
Human-in-the-loop
A design in which a person reviews or approves an AI system's action before it takes effect. It is the defining feature of Level 2 autonomy and is typically retained for sensitive actions even at higher levels.
Guardrails
Hard limits that bound what an AI system can do, such as spending thresholds, permitted counterparties, allowed tools and data it may access. Guardrails are enforced outside the model, so they hold even if the model makes a mistake.
Orchestration
Coordinating several AI systems or steps so they share context and hand work to each other, for example a sales system passing a signed deal to a finance system for invoicing. Orchestration across functions is the focus of Level 5 research.
Audit trail
A complete, time-stamped record of what an AI system saw, decided and did, including who approved it. Audit trails make behaviour traceable and support data-protection and governance obligations. See security and trust.
Evaluation suite
A set of tests run before every release of an AI system to check accuracy, safety and behaviour on real and edge-case examples, so improvements do not introduce regressions.
PDPA
The Personal Data Protection Act. Thailand's PDPA B.E. 2562 governs how personal data is collected, used, stored and disclosed, and Singapore has its own PDPA. AI systems that handle personal data should be designed for purpose limitation, minimization, access control and retention from the start. This is general information, not legal advice.
Retrieval-augmented generation (RAG)
A technique in which an AI system looks up relevant information from your own documents and data before it answers, so replies are grounded in your content instead of only the model's general knowledge.
Model-agnostic architecture
A design that is not tied to one AI model provider, so the best model for each step can be used and swapped as the field advances without rebuilding the system.
From the lab
Read the thinking behind the systems.
Agentic AI vs AI agents: what is the difference?
The two terms are used interchangeably, and that causes confusion when you buy or build. A plain-English comparison, with examples and a simple test.
ArticleAutonomy is a dial, not a switch
Most conversations about AI ask whether to automate something. The better question is how much authority to grant, and how that authority is earned.
ArticleWhat comes after agents: the orchestrated enterprise
Autonomous functions are the first step. The next is functions that share context and act together. A look at where the technology is heading and what it will require.
Go deeper
See how the terms fit together.
The platform page shows how autonomy levels, guardrails and orchestration combine into systems that run real business functions.