NamerynHow to Name a Fleet of AI Agents
One agent can get away with a cute name. Thirty need a convention: a name that says whose agent it is, what it does and where it runs, a person who answers for it, and a review that catches the ones nobody uses. This is that convention, built on what Microsoft, AWS and the A2A protocol ask of an agent's names.
The names one agent carries
| Name | Who reads it | What the platform says |
|---|---|---|
| Display name | People, in the admin center, Teams and Outlook | Microsoft Entra calls it the human-friendly name of an agent, and it can be changed (Microsoft Learn, read October 19, 2026). |
| Agent card name | Other agents | A required field in an A2A agent card: a human readable name for the agent (the A2A specification, read October 19, 2026). |
| Registry record | Your teams, when they search for an agent before building another | As announced on April 9, 2026, AWS Agent Registry, in preview, keeps a record per agent searchable by name and description; records start as drafts and are visible to the organization once approved (AWS). |
| System name | Logs, scripts and access policies | Your convention: the part this guide is about. |
A convention that scales
Put the team, the job and the environment in the system name, in that order, and keep it the same shape for
every agent: Agent-HR-Onboarding-prod, Agent-Sales-LeadTriage-dev,
Agent-Finance-InvoiceMatch-test. Microsoft's guidance for Entra Agent ID gives the same advice,
prefixing display names with the department or function, as in Agent-HROnboardingBot, so agents
are recognizable in logs and the admin center (Microsoft Learn).
Then give the display name to the people who work with it: HR Onboarding, or a short persona name if the team talks to it every day. The AI agent team name generator names a whole set at once, and the persona card writes the A2A agent card that carries the name to other agents.
One agent per environment
A test agent should never share a name, or credentials, with the one in production. Microsoft recommends separate blueprints or environment-specific credentials for dev, test and prod, and custom security attributes such as Environment, Department and DataSensitivity that access policies can read, for example to keep non-production agents away from production data (Microsoft Learn). The suffix makes the same split visible to a person reading a log.
A sponsor and an owner for every agent
Microsoft Entra Agent ID records a sponsor on an agent identity: the person or group accountable for the agent, used for things like contacting a human during a security incident (Microsoft Learn). Its best practices say every agent identity must have a sponsor and a technical owner, assigned when it is created and checked again when people change jobs (Microsoft Learn). Write both into the register next to the name, with a review date: the AI agent fleet register keeps one row per agent and exports it.
The reviews that keep the fleet tidy
- Every 6 to 12 months, the sponsor confirms the agent is still needed and properly set up; an agent nobody confirms is a candidate for decommissioning.
- Every quarter, look for agents with no sponsor, outdated details or no recent activity.
- Every year at least, rotate the certificates the agents use.
Source: Microsoft's Entra Agent ID best practices, read October 19, 2026. Microsoft Agent 365, generally available since May 1, 2026, keeps these agents in one registry in the Microsoft 365 admin center (Microsoft Learn).
FAQ
What is a good naming convention for AI agents?
A system name that says the team, the job and the environment, such as Agent-HR-Onboarding-prod, plus a friendly display name people see, such as HR Onboarding. Microsoft's own example prefixes display names with the department or function, like Agent-HROnboardingBot.
Who is the sponsor of an AI agent?
The person or group accountable for what the agent is for. Microsoft Entra Agent ID records a sponsor on every agent identity, to reach a human when a security incident happens, and Microsoft's guidance says to assign a technical owner as well.
Should development and production agents share a name?
No. Give each environment its own agent and a suffix such as -dev, -test or -prod. Microsoft recommends separate blueprints or credentials per environment, and an Environment attribute you can use in access policies.
How often should agents be reviewed?
Microsoft suggests sponsors confirm every 6 to 12 months that each agent is still needed, a quarterly check for agents with no sponsor or no recent activity, and rotating certificates at least once a year.
Is the display name the same as the agent card name?
Not necessarily. The display name is what people see in tools such as Teams and Outlook and can be changed; the A2A agent card carries its own required name, the one other agents read. Keep them the same unless there is a reason not to.