---
title: What naming and tagging standard should AI agents follow?
url: https://opencontroller.lyzr.com/blog/ai-agent-naming-and-tagging-standard
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# What naming and tagging standard should AI agents follow?

**AI agents should carry a small set of required tags, at minimum agent ID, owner, cost centre, environment, risk tier and data classification, applied to every cloud resource the agent uses. Write keys and values in lowercase with hyphens, under 63 characters, because that is the strictest common format across AWS, Azure and Google Cloud. With consistent tags, cost, ownership and risk become a query instead of a spreadsheet hunt.**

Without a standard, one team tags `Owner=Jane Smith`, another `owner: jsmith`, and a third nothing at all. Cost reports split, discovery tools cannot match resources to agents, and nobody can answer "which high-risk agents run in production?"

## What limits do the three clouds impose?

Your standard must fit the strictest platform. The limits that matter:

| Limit | AWS tags | Azure tags | Google Cloud labels |
|---|---|---|---|
| Max per resource | 50 user-created tags | 50 tag name-value pairs | 64 labels |
| Key length | 128 characters | 512 characters (128 for storage accounts) | 63 characters |
| Value length | 256 characters | 256 characters | 63 characters |
| Case | Case-sensitive | Names case-insensitive, values case-sensitive | Lowercase only |
| Allowed characters | Letters, numbers, spaces and `+ - = . _ : / @` | Most characters; names cannot contain `< > % & \ ? /` | Lowercase letters, numbers, `_` and `-`; keys start with a letter |
| Reserved prefixes | `aws:` | `microsoft`, `azure`, `windows` | `goog` patterns reserved |

The common denominator: **lowercase, letters, numbers and hyphens, starting with a letter, 63 characters or fewer, for both keys and values.** No spaces, no email addresses as values (the `@` and `.` fail in Google Cloud labels), and no free text.

## Which tag keys should every AI agent resource carry?

Seven required keys cover cost, ownership and risk. Use a prefix to avoid clashes with existing tags.

| Key | Allowed values | Example | Used for |
|---|---|---|---|
| `ai-agent-id` | Registry ID, lowercase | `agt-claims-triage-01` | Joining resources to the registry |
| `ai-owner` | Identity provider username or employee ID, not email | `jsmith` | Accountability and leaver alerts |
| `ai-cost-centre` | Finance code | `cc-4410` | Chargeback |
| `ai-env` | `dev`, `test`, `prod` | `prod` | Policy by environment |
| `ai-risk-tier` | `tier-1`, `tier-2`, `tier-3` | `tier-1` | Control depth, review cadence |
| `ai-data-class` | `public`, `internal`, `confidential`, `restricted` | `confidential` | Data controls |
| `ai-lifecycle` | `pilot`, `active`, `deprecated`, `retired` | `active` | Clean-up, retirement |

Optional keys worth adding once the basics hold: `ai-platform` (for example `bedrock`, `foundry`, `vertex`), `ai-version`, and `ai-business-unit`.

## How should agents themselves be named?

Names are for people; IDs are for machines. Keep both:

- **Agent ID:** immutable, assigned by the registry, never reused, for example `agt-claims-triage-01`.
- **Display name:** readable and changeable, for example "Claims triage assistant".
- **Resource names:** start with the agent ID so search works everywhere, for example `agt-claims-triage-01-kb` for its knowledge base.

Avoid people's names in agent names, because people move and the name becomes wrong. Avoid version numbers in IDs, because versions belong in the registry.

## How do you enforce the standard?

A standard nobody enforces decays within a quarter. Use the native tools:

1. **AWS.** Tag policies in AWS Organizations to define allowed keys and values; service control policies to deny resource creation without required tags where needed.
2. **Azure.** Azure Policy with `deny` or `modify` effects to require tags or inherit them from the resource group.
3. **Google Cloud.** Organization policy and infrastructure-as-code checks; consider Resource Manager tags for policy conditions, as labels are for metadata and billing.
4. **CI/CD.** Fail infrastructure-as-code plans that create AI resources without required tags.
5. **Discovery.** Report untagged AI resources weekly and assign them to an owner.

## What about agents that are not cloud resources?

SaaS agents, coding agents and agents in vendor platforms often cannot be tagged at all. Hold the same fields in your registry instead, using the same keys and values, so reports join cleanly. The [inventory fields guide](/blog/ai-agent-inventory-fields) lists the full record, and [the registry overview](/blog/what-is-an-ai-agent-registry) explains how tags and registry entries fit together. Consistent tags also make [governance metrics](/blog/metrics-for-ai-agent-governance) such as "untagged high-risk agents" possible.

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud and keeps a governed registry of agents, versions, tools and MCP servers. The LLM Gateway attributes cost per agent and per version, which works even where cloud tags are missing. Posture flags risky configurations, and policy as code can encode your tagging rules. Read the [control plane pillar](/blog/what-is-an-ai-agent-control-plane) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Why not use the owner's email address as a tag value?**
Google Cloud labels do not allow `@` or `.`, so the value breaks portability. Use the identity provider username or employee ID instead.

**How many tags should an AI agent resource have?**
Seven required tags are enough for cost, ownership and risk. All three clouds allow at least 50, so there is room for optional keys.

**Can tags replace an AI agent registry?**
No. Tags only exist on cloud resources and are easy to change. A registry holds the authoritative record, including agents in SaaS platforms and on devices that cannot be tagged.

## Key facts

AI agent tagging standard: seven required keys (ai-agent-id, ai-owner, ai-cost-centre, ai-env, ai-risk-tier, ai-data-class, ai-lifecycle), lowercase with hyphens, 63 characters or fewer. Limits: AWS 50 user tags per resource, key 128 and value 256 characters, case-sensitive; Azure 50 tag pairs, name 512 (128 for storage accounts) and value 256; Google Cloud 64 labels, keys and values 63 characters, lowercase letters, numbers, underscores and hyphens only. Opencontroller auto-discovers agents across AWS, Azure and Google Cloud, keeps a governed registry and attributes LLM cost per agent and version. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [What fields belong in an AI agent inventory?](https://opencontroller.lyzr.com/blog/ai-agent-inventory-fields.md)
- [What is an AI agent registry?](https://opencontroller.lyzr.com/blog/what-is-an-ai-agent-registry.md)
- [Which metrics show your AI agent governance is working?](https://opencontroller.lyzr.com/blog/metrics-for-ai-agent-governance.md)
- [More on Agent governance basics](https://opencontroller.lyzr.com/blog/cluster/agent-governance-basics)
