The words of agentic AI,
in one sentence each.
The terms you meet when you put AI agents to work as a team, as Atipol uses them.
Agentic AI
AI systems that plan, use tools and chain actions to reach a goal, instead of only answering a question. Because they act, they need roles, permissions and controls.
AI agent
A program driven by a language model that reads, edits and calls tools to complete a task. In Atipol, every agent has a job description, a scope, allowed tools and a responsible lead.
Agent lead
The agent that acts for one person: it splits requests into tickets, assigns them to agents, reviews deliveries and integrates them. It never has more rights than the person it works for, and each person has their own.
Sub-agent
An agent that another agent delegates part of its work to. It inherits the framework of the agent that called it.
Ticket
The unit of work given to an agent. It goes through these steps: draft, transmitted, in progress, review, integration, done, and keeps its full history.
Acceptance criteria
The typed conditions a delivery must meet, written before the work starts: a result, a skill to apply or a proof to provide. An agent cannot deliver until it has checked every one.
Scope
The files an agent may and may not change. It is checked on the delivered version itself, not on what the agent says it did.
Review
The check of a delivery by someone other than its author, before it is integrated. In Atipol, an agent never reviews its own work.
Recipe
A reusable job description for an agent: mission, instructions, skills for each phase, ticket templates, suggested scope and recommended model. Recipes can be shared through a library or the marketplace.
MCP (Model Context Protocol)
An open protocol that lets an AI agent call external tools and data sources through MCP servers. Atipol sets each agent's rights tool by tool: allowed, subject to a person's approval, or denied.
Runner
A machine or Docker container where agents run, on your side. Repositories stay there; Atipol receives the reference of the delivered version, never the files.
Companion app
The signed Mac and Windows app that links a machine to your organisation, installs the tools and runs the Atipol MCP server. On Linux servers, the same companion runs as a command-line binary or in a Docker runner.
Execution policy
The per-project rules that set each agent's model and the number of concurrent sessions, and require an explicit, justified confirmation before a more expensive model is used.
Tracking register
A shared table or document where a project follows its work beside the tickets. Registers can be shared with other projects, read only or read and write.
Audit log
The organisation's record of members, roles, security settings, key openings and marketplace actions. It is exportable as CSV and never contains ticket content.
Insights
Measures of deliveries, teams, consumption and estimated cost at three levels: organisation, workspace and project, down to the ticket.
AI agent governance platform
What it governs: roles, permissions, review, audit trails and costs. Atipol is one.
AI agent governance
The rules that set, for each AI agent, its scope, its rights, its review and the budget tracked. It covers roles as well as the audit log that keeps the proof.
Human in the loop
A human decision taken before an action or an integration happens. At Atipol, that means the review before integration and the approval asked before sensitive MCP calls.
Guardrail
A limit set before the action: a file scope, allowed tools, criteria to meet. It bounds what an agent can do, not only what it should do.
Shadow AI
The use of AI tools by people or teams outside the framework set by their organisation. A shared framework, such as common roles and scope, reduces the temptation to step outside it.
Sovereign AI
Keeping control of your data, your keys and where processing runs, instead of handing it all to a third party. At Atipol, repositories stay on your machines, the runners, and project content is encrypted under your organisation's key.
Worktree
A separate Git working copy, on its own branch. At Atipol, every agent works in its own.
GDPR
The European regulation that protects personal data: legal basis, minimisation, retention period and people's rights. It applies to AI agents as soon as they process personal data.
AI Act
The European regulation that classifies AI systems by risk level and sets obligations for each level. It covers AI agents used in companies too, not only model providers.
Cost per accepted delivery
The amount spent on AI measured against the number of deliveries actually accepted, rather than the number of requests or tokens consumed. It is the indicator to track, not just the monthly bill.
Updated on November 11, 2026.