The language of the AI-native workspace.
The AI-native workspace is one team of people and AI agents, working from one memory. These are the words we work by.
Start here.
Last updated
People move on and so do AI agents. What stays is the organisation: its roles, its rules and what it has learned. Its memory holds the work as it happens: requests, plans, decisions and their sources.
AI agent
An AI agent is a piece of software that does a job for the team, with a person responsible for it.
The team.
Who does the work: people and AI agents, each with a role and an owner.
OrganisationWorkspace
Your organisation’s place in Stellr: chat, sources, AI agents, flows, memory and settings, separated from every other organisation’s.
PositionRole
A job that stays, for a person or an AI agent. Who fills it changes. The work stays with the workspace.
Line managerOwner
Who answers for an AI agent. Every AI agent has an owner in your organisation, and a person is accountable for it.
HireHire
The same word for a person and an AI agent. Either joins with a role; an AI agent also gets an owner.
ExpertiseModel
The AI system an AI agent uses to read, reason and respond. The workspace holds the memory, not the model.
Job descriptionAgent instructions
Define an AI agent’s role, expected behaviour, boundaries and when to ask a person for help.
The work.
How a request becomes a plan, is approved and gets done.
Email threadRequest
Work arriving. It lands in one place with its context, from email or a WhatsApp export, marked with its source.
ProcessPlan
What a request becomes: steps with owners, some people and some AI agents. The Flows page holds them.
Sign-offApproval
One person’s yes before a plan runs. The moment human judgement gets written down.
BriefingPrompting
Give an AI agent a goal, constraints and requirements for the result.
Work transferHandoff
Passing a task between a person and an AI agent, with its context. The record shows who did each step.
The record.
What the work leaves behind, and what the organisation keeps.
Institutional knowledgeMemory
What your organisation knows, kept in order in the workspace. People and AI agents work from it and add to it.
Background informationContext
What an AI agent has in front of it for the current task: instructions, information and results. Memory can feed it.
ReferenceSource
Where an answer or a record came from. Every answer from the memory shows its source.
Paper trailRecord
What the work leaves behind as it happens: who did each step and what was decided.
AmendmentCorrection
A change to what the memory holds. A memory with sources can be checked, and what was learned in error undone.
Standard procedureSkill
A reusable package of instructions and resources an AI agent can load for a particular kind of work.
The rules.
What people and AI agents work under.
PolicyRule
A policy for people and AI agents alike, set in Settings and carried by each AI agent.
Scope of authorityBoundary
The limit of what a person or an AI agent may see and do. In Settings you decide who may see what.
Operating controlsGuardrails
Checks and controls around an AI agent’s inputs, actions or outputs. A person’s approval stays a separate gate.
The field’s words.
These describe how AI agents are built, not what Stellr does.
Work environmentHarness
The software around the model that manages its work loop, supplies context, connects tools and handles execution.
Staff a taskSpawn
Start an additional AI agent instance for a defined task within configured limits. Stellr’s word is Hire.
CoordinationOrchestration
A system coordinates AI agents and tools, routes tasks and combines results, with human involvement where required.
Application useTool call
An AI agent requests a specific function, such as searching records or updating an approved system.
System connectionConnector
An integration that makes a system’s data or actions available to an AI agent under configured permissions.
Quality assuranceEvals
Repeatable tests that assess AI agent behaviour and results against explicit criteria, alongside human review.
Activity logTrace
A recorded sequence of an AI agent run, including captured tool calls, handoffs, results and errors.
Changes.
2026-09-16: the first fifteen terms.
2026-09-24: twelve terms added and the page reorganised by team, work, record and rules.
References.
Two essays from Anthropic. The original fifteen terms draw on them.
Building effective agents
Anthropic engineering, December 2024.
Effective context engineering for AI agents
Anthropic engineering, September 2025.
This is also how we work with you: one team first, these words throughout, and a record of everything.
The list is not finished. Send us a term, or bring a team and learn the words on your own work.