AI agents A to Z

Proposals, contracts and articles about AI now talk about agents. The same few words come up again and again.

Agentic AI

The UK government's AI Playbook was published on 10 February 2025 for people working in government. It says "Agentic AI refers to autonomous AI systems that can make decisions and perform actions with minimal human intervention". It adds that these agents can identify the tools and functions at their disposal and use them to take actions towards their objectives.

AI system

The Organisation for Economic Co-operation and Development (OECD) runs a site called OECD.AI. An article there, dated 29 November 2023, explains an update to the OECD's definition of an AI system. The new wording begins "An AI system is a machine-based system". It goes on to say that such a system "infers, from the input it receives, how to generate outputs" such as predictions, content, recommendations or decisions. The authors note that their opinions do not necessarily reflect the official views of the OECD.

Hallucination and prompt injection

The AI Playbook defines both. "Hallucinations are when the generative AI model responds with information that appears to be truthful but is actually false."

The Playbook calls prompt injection a type of perturbation attack aimed at generative AI systems that use text prompts. It explains that a generative AI model "cannot distinguish between the user prompt and these system instructions". A crafted prompt can then make the model respond in an unintended way. The Open Worldwide Application Security Project (OWASP) runs a Gen AI Security Project, which puts prompt injection first in its 2025 list of risks for large language model applications, as LLM01.

Human in the loop

The fourth principle of the AI Playbook is "You have meaningful human control at the right stages". Where AI is used in situations of high impact or risk, the Playbook asks teams to "establish a human-in-the-loop process to oversee and validate outputs and decisions".

Assurance and monitoring

The Department for Science, Innovation and Technology published an introduction to AI assurance on 12 February 2024. It says "Assurance is the process of measuring, evaluating and communicating something about a system or process, documentation, a product or an organisation". For AI, it measures, evaluates and communicates how trustworthy a system is.

Monitoring is the part that carries on after launch. The National Cyber Security Centre's guidelines, published on 27 November 2023, cover "logging and monitoring, update management and information sharing" once a system is deployed.

More terms

The US National Institute of Standards and Technology (NIST) keeps a longer list, "The Language of Trustworthy AI: An In-Depth Glossary of Terms". It is "being released in beta format as a spreadsheet". NIST says the aim is to give people "a broader awareness of the multiple meanings of commonly used terms". Two sources can define one word differently.

Useful links