Install and Compare

Installation is the whole point. In Copilot Chat, attach the files, drop them into a Copilot Page, or add them as knowledge for an agent that should always know this project.

If Copilot cannot load it, you built a notebook, not a brain.

Then run the same real request twice: once with no files, once with the brain attached. Compare. If nothing changes, the files are too vague or Copilot did not read them.

A teammate should be able to trigger the same pack and get a draft that sounds like your world, not a generic professional.

Core principles

  1. Pick one home per use: attach for ad hoc, Page or agent knowledge for repeat work.
  2. A/B the same request. Write one sentence on what changed.
  3. If the with-brain draft is still generic, thicken the project pack — do not add adjectives to About Me.
  4. Optional fourth file later: a glossary of MillerKnoll or dealer terms an outside model would miss. Not required to finish this course.

Next: Your AI Usage — see where tokens actually go

Check yourself

How do you know the personal brain is working?

Do this in Copilot

Run the same request with and without the files. Keep the pair that proves the brain earned its keep.

Run A in a chat with no attachments. Run B in a new chat with the three files attached. Then paste both for the score.

A/B the same request

I will run the same task twice. First, I will send only the task. Second, I will attach my About Me, How I Work, and project pack. Task: [YOUR REAL REQUEST]. After the second run, wait for me to paste both outputs labeled A (no files) and B (with files). Score them on: fit to my audience, respect for my constraints, and edits needed before send. Say which is better and why in two sentences.

Techniques: Source material · Claim spot-check

  • Copilot Chat: paperclip or add files from OneDrive. Agents: add the files as knowledge only if that agent should always see them.
Open Copilot

Did you run this in Copilot? Mark complete when you have tried it.

AI

Part of the MillerKnoll collective

Studio · 2026