Digital collar glossary

The vocabulary around AI and work moves fast, and many terms are used loosely. These are the definitions we use on this site, in plain words. Where a term has an official or research definition, we follow it and link the source.

Agentic AI

AI systems that carry out multi-step tasks on their own, work inside software and make decisions based on context, rather than only answering a single prompt. The ILO notes that models have been gaining these abilities since 2023. (source)

AI agent

A piece of software, usually built on a large language model, that is given a goal and takes actions to reach it, such as searching, writing files, calling tools or sending messages.

AI literacy

The skills, knowledge and understanding that let people make informed use of AI systems and be aware of their opportunities, risks and possible harms. Defined in the EU AI Act, which requires providers and deployers to support it among their staff. (source)

Automation bias

The tendency to rely, or over-rely, on the output of an automated system. People overseeing high-risk AI systems under the EU AI Act must stay aware of it. (source)

Blue collar

Work that is mainly physical and hands-on, such as trades, manufacturing, construction and transport. The name comes from the blue work shirts manual workers wore.

Brief

The instruction given to an AI agent: the goal, the context it cannot guess, the limits and what a finished result looks like.

Deployer

Under the EU AI Act, an organization or person using an AI system under its authority, as opposed to the provider that built it. Deployers of high-risk systems carry duties such as human oversight and keeping logs. (source)

Digital collar

Work where the main job is directing AI agents and automation, checking what they produce and owning the result. It describes a kind of work, not a separate group of people.

Exposure (to generative AI)

A measure of how much of an occupation's current tasks could be done with generative AI. The ILO groups occupations into four gradients, from low to highest exposure. Exposure shows potential, not job loss. (source)

Generative AI

AI that produces new content such as text, images, audio, video or code from a prompt.

Hallucination

When an AI model produces information that sounds right but is false, such as an invented figure, quote or source. The main reason AI output needs checking.

Human in the loop

A setup where an AI system proposes actions and a person approves each one before it takes effect.

Human on the loop

A setup where an AI system acts on its own while a person monitors it and can intervene or stop it.

Human oversight

The ability of people to understand, monitor, override and stop an AI system while it is in use. Required for high-risk systems by Article 14 of the EU AI Act. (source)

Large language model (LLM)

An AI model trained on large amounts of text to predict and generate language. Most current AI agents are built on one. (source)

Prompt

The text or other input given to an AI model to tell it what to do.

Reskilling and upskilling

Training workers for a different role (reskilling) or for new demands in their current role (upskilling). The WEF report estimates that 39% of workers' existing skill sets will change or become outdated between 2025 and 2030. (source)

Risk management framework

A structured approach to finding and handling the risks of a technology. The NIST AI RMF organizes it into four functions: Govern, Map, Measure and Manage. (source)

White collar

Work that is mainly done at a desk with information, such as administration, finance and management. Named after the white shirts and collars of office workers.