AI Innovators Forum // A.M.A.N.D.A. governance practice
A.M.A.N.D.A. is a six-stage review you can apply to any agentic AI workflow, whether the agents run on your own infrastructure, a cloud platform, or a third-party product. It governs the decision process, not the technology, and keeps a named person responsible for every consequential decision.
This is a learning hub, not a social forum or a hosted platform. There are no accounts, no public posts, and no production data.
The framework splits agentic AI governance into six disciplines instead of one large policy. Engineering, operations, compliance, and finance teams can each adopt the stage that fits their workflow, and each stage produces a document someone can review.
Set the mission charter before any agent action is authorized: the intended outcome, authority boundaries, excluded actions, and the named person accountable for the decision.
Record every input, data source, and model output in a decision trace, with its source, timestamp, and confidence, before any claim reaches a decision point.
Check the agent's outputs and permissions against your control requirements: least-privilege API access, data-handling policy, output accuracy, and regulatory fit.
Route every agent action through three separate steps: the agent proposes, a human or system decides, and execution happens only after the decision.
Scope credentials and make actions reversible before granting any agent write, spend, or delete authority. Test the rollback for every path.
Turn the evidence into a decision owned by a person, naming the approver, the monitoring plan, and the condition that stops the work.
Set the objective, the accountable function, the workflow, the stage, and the consequence level. The route updates as you change them and runs entirely in your browser.
The route builder needs JavaScript. The default example is shown below.
A.M.A.N.D.A. SIX-STAGE GOVERNANCE ROUTE Governance objective: Complete a supervised governance action Accountable function: Enterprise AI Platform Teams Workflow domain: CI/CD pipeline automation Current stage: Learning and planning Consequence level: Medium (internal systems or routine operations) 1. AMY (Mission): Define the mission charter, authority boundary, excluded actions, and the person accountable for the decision. 2. MAYA (Evidence): Collect the pipeline manifest, deployment history, approval logs, and current access-control policy before any change proposal is drafted. 3. ARIA (Audit): Audit build integrity, test coverage, secrets handling, least-privilege API access for CI/CD credentials, and compliance with the change-management policy. 4. NANCY (Routing): Separate the agent's deployment proposal from the human decision to promote, and the decision from the automated execution. 5. DAWN (Defense): Scope CI/CD credentials to least privilege, require signed commits, and keep a tested one-click rollback before enabling any autonomous deploy step. 6. ASTRA (Decision): Approve production promotion only after the audit trail confirms the change matches the reviewed proposal. STAGE RULE: Use sandboxed examples and isolated test data. Do not connect live production systems or credentials. CONSEQUENCE RULE: Internal or routine operations require evidence review, access-control checks, and a named approver. A planning checklist, not legal, compliance, or security advice.
No. It is a curated Hyde Workshop learning hub. It creates no accounts, public profiles, or user-generated content, and it processes no production data. The route builder runs in your browser to illustrate the six-stage method.
No. Amy, Maya, Aria, Nancy, Dawn, and Astra name the six control disciplines of the framework. They are not live agents. Using this page does not start an AI runtime, approve an action, or authorize deployment in any connected system.
No. A.M.A.N.D.A. is system-agnostic by design. The same six stages apply whether the agents run on your own infrastructure, a cloud AI platform, or a third-party product, because the framework governs the decision process, not the technology stack.
No. The route is a planning checklist built from templates and your selections. It is not legal, compliance, or security advice, and it does not certify anything. Have a qualified person review it against the rules that apply to you.
This page gives the system-agnostic overview and a governance route across common business workflows. AI Forge builds a deployment contract with six controls, and the A.M.A.N.D.A. analysis turns an objective into a short brief.
Run any agentic action, such as a deployment, a CRM update, or a financial approval, through the A.M.A.N.D.A. gates before granting production authority.
