The language of this field, in plain words
Nobody should have to nod along to a word they have not been given the meaning of. Here are the terms you will hear, each in one sentence, with an example.
- Agent, or AI agent
- Software that can be given a goal and carry out the steps towards it, using tools and making choices along the way.
- For example: A system that reads an incoming claim, checks the documents, and pays it if everything matches.
- Agentic AI
- AI that takes actions, as distinct from AI that only answers questions.
- For example: A chatbot tells you the refund policy. An agent issues the refund.
- Autonomy
- How much an AI system may do without a person stepping in.
- For example: See the six levels.
- Decision rights
- The agreed answer to: who may decide what?
- For example: An agent may approve refunds under a set amount. Anything larger goes to a team leader.
- Delegation
- Handing a task, and the authority to do it, to someone or something else. Accountability stays with whoever delegated.
- For example: A manager lets an agent schedule maintenance, and still answers for the schedule.
- Governance
- The arrangements that make clear who is accountable, what is allowed, and how that is checked.
- For example: A committee that approves which agents may act without sign-off, and reviews them every quarter.
- Guardrails
- Limits built into a system so it cannot do certain things.
- For example: An agent that cannot send money to an account it has not seen before.
- Human in the loop
- A person who reviews or approves what an AI system does before it takes effect.
- For example: An underwriter who signs off each AI-drafted quote.
- Identity
- How your systems know who or what is acting.
- For example: An agent with its own account, so its actions are not recorded as a staff member's.
- Operating model
- How an organisation is set up to get its work done: its structure, roles, processes, decisions and technology.
- For example: Whether claims are handled by one central team, by regions, or by agents with people handling exceptions.
- Orchestration
- Coordinating several agents, tools and people so the work passes between them in the right order.
- For example: One agent gathers documents, a second assesses them, a person approves, a third pays.
- Organisational design
- Deliberately shaping structure, roles, decision rights and ways of working so the organisation can do what it has set out to do.
- For example: Redrawing a team's roles when half its routine work moves to an agent.
- Psychological safety
- People feeling able to speak up, admit a mistake or question a decision without being punished for it.
- For example: A junior analyst saying the AI's answer looks wrong, and being thanked.
- Readiness
- Whether an organisation can take something on and run it well.
- For example: Having the owners, skills, controls and trust in place before an agent goes live.
- Recourse
- A way for someone affected by a decision to challenge it and have it looked at again.
- For example: A customer who can ask a person to review a claim an agent declined.
- Sociotechnical
- Treating the people side and the technology side of work as one system that has to be designed together.
- For example: Redesigning jobs and training at the same time as installing the new system.
- Traceability
- Being able to reconstruct what happened and why.
- For example: A log showing each step an agent took and which approvals it relied on.