One class, two client academies.
This is one Alpine Academy class, shown the way two different companies' staff see it.Same class, different tools, rules, examples and brand.Nothing below is hand-edited per client: each academy is one config file.
Fictional example companies
Agents with Approvals
Build an agent with approval checkpoints
Agents are only useful if you can trust them with real work.Design an agent for a process your team runs every week.Decide what it may do alone, what needs approval and what it must never touch.Then run it with checkpoints, and a log that shows exactly what happened.
- 01Break a real process into steps an agent can carry out
- 02Set a permission level for every step: automatic, approval or blocked
- 03Write approval requests a busy manager can decide in seconds
- 04Log every action so anyone can see what the agent did and why
- 05Know when to widen an agent's autonomy, and when to pull it back
- AI assistantsMicrosoft Copilot
- AutomationPower Automate
The models built into Microsoft 365 Copilot, in the Ashgrove tenant
Use Microsoft Copilot signed in with your Ashgrove account, and Power Automate in the Ashgrove environment.Other AI tools are not approved for client or firm data.
Client data never leaves approved tools
Non-public personal information is protected under GLBA and Reg S-P.Use Copilot signed in with your Ashgrove account and nothing else.Never paste names, account numbers or balances into a public AI tool.
Outputs that inform decisions are models
If an AI output feeds a credit, pricing or suitability decision, it falls under model risk management (SR 11-7).Document what it does, validate it, and keep a human making the call.
No unreviewed AI in credit decisions
Fair lending rules (ECOA and Reg B) apply whatever the tool.AI can prepare a credit memo.It cannot approve, decline or price a loan.
Automations leave an audit trail
Every flow logs what it read, what it changed and who owns it.If an examiner asks what happened, the answer should take minutes, not a week.
An agent that assembles an exam document request, with Compliance approving every file before it leaves the firm.
What changed between the two.
Both columns are read from the two configs for Agents with Approvals.Change the class above and the table follows.
| Difference | Ashgrove Financial | Northwind Logistics |
|---|---|---|
| Tools named | Microsoft Copilot and Power Automate | ChatGPT Enterprise and Zapier |
| Model and tool policy | The models built into Microsoft 365 Copilot, in the Ashgrove tenant Use Microsoft Copilot signed in with your Ashgrove account, and Power Automate in the Ashgrove environment.Other AI tools are not approved for client or firm data. | The models available in the Northwind ChatGPT Enterprise workspace Use ChatGPT Enterprise signed in with your Northwind account, and Zapier in the Northwind workspace.Other AI tools are not approved for customer or shipment data. |
| Role example | Compliance & Risk An agent that assembles an exam document request, with Compliance approving every file before it leaves the firm. | Operations An agent that prepares a receiving discrepancy report, with a supervisor approving before it goes to the carrier. |
| Risk themes |
|
|
| Brand | Ashgrove AI Academy
| Northwind AI Academy
|
Your company is the third column.
We scope your tools, rules and roles, then build your version.