
Building an AI Workloads Register with AWS Agent Registry
A spreadsheet can track five AI workloads. An enterprise can't. Following the regulators' expectations, this post shows how to stand up a scalable, searchable AI Workloads Register on AWS Agent Registry — embedding accountability, approvals, and auditability into your AI pipeline.
The National AI Centre (NAIC) is the Australian Government’s lead body helping organisations understand and use AI safely. NAIC recently released their Guidance for AI Adoption . This framework consolidates earlier initiatives into six essential practices that every organisation should implement:
- Decide who is accountable
- Understand impacts and plan accordingly
- Measure and manage risks
- Share essential information
- Test and monitor
- Maintain human control
Why an AI Register Matters
One of the most practical tools the NAIC recommends is an AI Workloads Register — a central record of every AI workload your organisation uses, including: 1/ Who owns it 2/ What it does and why 3/ What data it processes 4/ Its risk level 5/ Known limitations and potential for misuse, and, 6/ Key stakeholders affected
The register creates visibility, prevents duplication, reduces shadow AI, and gives leaders evidence of control when scrutiny arrives — whether from boards, regulators, or customers.
Designing the AI Workloads Register
The NAIC provides a template to get started , including a spreadsheet which is fine for five workloads. However, an enterprise-grade register needs a purpose-built, scalable solution that allows:
- Discovery — "Which AI workloads process customer data?"
- Workflow — Draft → Review → Approve lifecycle
- Access control — Who can register vs. who can approve?
- Automation — Alerts when new workloads need review
- Auditability — Who approved what, and when?
AWS Agent Registry
AWS Agent Registry (Preview) stores metadata for every agent, tool, MCP server, agent skill, and custom resources as a structured record. It captures who published each record, what protocols it implements, what it exposes, and how to invoke it. The registry supports established standards like MCP and A2A natively, with the flexibility to define custom record types for your organisation.
While it's designed for AI agent and tool discovery, its Custom record type makes it a powerful platform for implementing a full AI workloads register.
Capability | How it supports an AI Register |
|---|---|
Custom record type | Model a flexible registry — map directly to NAIC template fields |
Approval workflow | Built-in Draft → Pending Approval → Approved/Rejected lifecycle |
Semantic search | Find AI workloads by purpose, not just name ("customer data processing") |
Authorisation | Control who can register, who can approve, who can search |
Embed into your SDLC | Automated registration into your CI/CD pipelines with API and MCP support |
Notifications | Notify governance teams when new workloads need review |
Customising AI Workloads Register
The NAIC guidance encourages organisations to adapt the register — add fields you need, remove fields that don't add any value.
AWS Agent Registry also stores Agent (A2A) records, MCP server records and Agent Skills which can extend your AI workload register to reference other registered components. In this example, I am extending the registry model with Agents and Models.
NAIC Guidance Field | Type | Example |
|---|---|---|
System name and version | string | "Customer Service Chatbot", "2.1" |
System owner | object | {"name": "Sarah Chen", "role": "Customer Success Manager"} |
System status | enum | "pilot", "operational", "decommissioned" |
Source | string | "Internal IT team" |
Purpose / business goals | string | "Reduce customer service response times" |
Intended use case(s) | array | ["Answer common customer questions"] |
Known limitations and prohibited use | array | ["Cannot handle complex complaints"] |
Any foreseeable misuse | array | ["Manipulate or mislead customers"] |
Data sources and type | array of objects | [{"source": "Customer tickets", "type": "text"}] |
Registered date | ISO date | "2026-01-15" |
Screening tool result | enum | "normal", "elevated", "unacceptable" |
Key stakeholders affected | array | ["Customers", "Support staff"] |
(Extended) Agents | array of objects | [{"agent": "Customer Service Agent", "role": "..."}] |
(Extended) Models | array of objects | Amazon Nova Pro |
Implementing the Register
You can create and update the registry using multiple methods, including AWS SDK and AWS CLI. You can also use the registry's MCP endpoint to allow your AI Agents to discover registered Agents and MCP Servers. Each registry exposes an MCP-compatible endpoint following the 2025-11-25 specification .
For your Agents and MCP Servers, you can also point Agent Registry to their endpoints for data synchronisation - regardless if the agents or the MCP servers are hosted on AWS.

Once you create your register, you can start adding your AI workloads. You should ideally embed the registry into your CI/CD pipeline, so you don't miss an update or a new AI workload.

You can enable approvals to make sure records are aligned with your Responsible AI Principles.

Here, our - currently - humble AI Workloads Register already has two workload records.

Agent Registry supports both semantic and keyword search. In this example, I am searching for a particular business executive to find out which AI Workloads she is accountable for.

Conclusion
A well-designed AI Workloads Register is more than a compliance artifact—it becomes the operational backbone of responsible AI. It gives organisations a real-time view of their AI landscape, embeds accountability into delivery pipelines, and creates a defensible position when governance questions arise. By combining NAIC’s guidance with AWS Agent Registry, you can maintain a cohesive view of your AI landscape.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article