EvidencePack — Source-Grounded Review & Approval for Agent-Generated Work Outputs

The Problem

AI-generated work is crossing from “first draft helper” to actual operating layer.

Notion’s own example is compiling customer feedback from Slack, Notion, and email into actionable insights. Google’s Kärcher case chains multiple Gems into a ready-to-review user story. Atlassian highlights weekly reporting agents and sales-support agents answering questions directly in Slack.

All of that saves time. But it also creates a review problem:

• where did this claim come from;
• which sources were ignored;
• which sentence is inference vs direct evidence;
• who approved this output;
• what changed from last version;
• which policies were checked before publishing or sending.

Most current tools generate the work but do not provide a strong review-grade evidence layer for professional teams.