Use one source for people and AI

2 min read

Parallel Training uses the same approved organizational knowledge for human learning and AI assistance. Give people a course built from the source, and give an Agent or connected AI client scoped access to that source.

1. Prepare the shared source

Create a Knowledge Base folder for the process or product.
Import or write the approved documents.
Check titles, images, source accuracy, and folder permissions.
If the source lives in GitHub, connect the intended repository folder and verify a completed sync.

The fictional onboarding document in the Knowledge Base

Import files and links · GitHub sync

2. Build the human learning path

Attach the document to a course, explain the skill learners should demonstrate, and generate the lessons and practice. Review the draft and deliver it through an assignment, academy, or LMS.

The attachment is a snapshot. Refreshing a changed source updates that material, not the authored lessons or questions. Revise and review the affected course content separately.

Add sources to a course

3. Configure AI access

Create an Agent with read access to the approved folder, or a scoped Connector for an external AI client. Give it clear instructions about which process it supports and ask it to cite its sources. Use a Skill when you need reusable instructions for executing that process.

Test an answer whose source you know. Check that citations open the intended document and that the Agent cannot read unrelated private folders. Add a delivery channel only when that audience is ready to use it.

Create an Agent · Connect an AI client · Rules and Skills

4. Keep the two uses aligned

After an approved source update, verify its Knowledge Base version, test a fresh AI answer, and review which course lessons need revision. Publish the reviewed course changes through the configured workflow. A recurring drift-check automation can help identify work; it does not by itself publish a new course.

Use Automations for a scheduled source-review task, and Inspect Agent logs to review the recorded result. A successful run still needs comparison with the source.