The previous module introduced agents Microsoft has already built. Here, you'll make one of your own without writing code.
You begin by giving Cowork a real task with several steps, while keeping any action that changes something behind your approval. The next lessons cover scheduled workflows, agents that capture meeting decisions, and task agents that help plan a week. In each case, you'll check the AI's output against what happened in the source or system. You then use Agent Builder to make a focused agent over your own documents, push it with questions it shouldn't answer, and share it through SharePoint or a self-service entry point only when the evidence supports that step.
The first build may take only a few minutes. The careful part comes afterwards. You'll define the boundary, test the weak spots, and verify the result before anyone else relies on it. Availability can differ between tenants because some of these features are still rolling out. Buttons may move too. The habits in this module hold up across those changes: set a clear scope, ground answers in evidence, and keep a person in charge of consequential decisions. Bring a harmless practice project, and you'll leave with an agent you have tested yourself.
What you'll walk away with
Delegate outcome-level work to Cowork behind an approval boundary
Build an approval-gated workflow that never sends after a rejection
Verify meeting records and task-agent output against real evidence
Create a grounded no-code agent with Agent Builder and test it properly
Deploy SharePoint and self-service agents to colleagues responsibly
