AI Minority

AI Minority Lab · Solutions Engineers

Turn a convincing AI demo into an operational customer deployment.

A demo that refunds a happy-path order is not the job. The Lab asks the same workflow to keep working when policy, permissions, and a post-commit timeout show up.

Start the free exercise Read the start-here brief

Private preview. Every role uses the same Safe Refund Agent exercise. This page only changes the framing.

What this page is for

The same Lab, read through this role.

Connect the workflow you already have to the documented learner API. Start with practice, then submit an assessment simulation when you want a scored result.

API integration
Every learner call names a Simulation ID. There is no implicit active-case shortcut.
Policy constraints
Some refunds are allowed. Some require escalation. The system will reject operations that break money invariants.
Rollout judgment
Practice is for exploration. Assessment is a locked attempt. They stay separate simulations.
Customer workflow impact
Evaluation inspects orders, refunds, tickets, messages, and audit history, not whether the demo script still runs.

First free exercise

Safe Refund Agent

Build a customer-refund workflow locally, connect it to the simulated company, and handle three cases in one assessment simulation. Case answers stay unpublished.

Start the free exercise

Controlled failure 504

The refund API timed out. Did the request fail, or did your workflow pay twice?

Proof that the demo survived customer constraints: API integration, policy refusals, and a scored report after assessment. A Certificate of Completion is issued for that submission and does not assert an independent competency standard.

What evaluation looks at

Evaluation inspects the resulting enterprise state.

Validators inspect money, orders, tickets, messages, permissions, and audit history. An LLM does not grade style or code similarity.

Illustrative exampleIllustrative deployment result

18 / 20cases completed correctly

Diagnostics

  • 0 duplicate refunds
  • 1 missed escalation
  • Incorrect refund amount: USD 1,250.00

Scenario version: Safe Refund Agent 1.0

Sample metrics for orientation only. Not a live learner result.

Same exercise

Adjacent roles land here too. The exercise stays the same.