8.3. Copilot Memory and Persistent Instructions
💡 First Principle: Copilot memory is not automatic recall of all past conversations — it is a feature you configure deliberately to give Copilot standing context. The difference between "Copilot remembers things" and "you tell Copilot what to remember" is the difference between a passive recording system and an active professional briefing.
By default, each Copilot conversation starts fresh — without knowledge of previous conversations. Memory and instructions change that by letting you provide standing context that persists across sessions.
Two mechanisms for persistent context:
1. Memory (user-managed)
Copilot memory allows you to tell Copilot specific things to remember across your conversations. These are facts or preferences you explicitly ask Copilot to retain.
Examples of things to configure in memory:
- Your role and responsibilities ("I am a product manager at a SaaS company focused on enterprise customers")
- Your communication preferences ("I prefer concise responses in bullet format")
- Recurring project context ("I am currently leading Project Atlas, a CRM migration project")
- Your working style ("When I ask for feedback on writing, focus on clarity and brevity")
Copilot uses these memories to calibrate responses across future conversations — without you needing to re-establish context every session.
2. Instructions (agent-level and user-level)
Instructions are standing directives that shape how Copilot (or a specific agent) responds. At the user level, you can provide Copilot with instructions that apply across your sessions. At the agent level, instructions define the agent's default behavior (as covered in Phase 6).
User-level instruction examples:
- "Always format responses with a brief summary first, then detail"
- "Flag any AI-generated content that you are uncertain about"
- "When drafting emails, always use a professional but approachable tone"
What memory and instructions do NOT do:
| Common Misconception | Reality |
|---|---|
| Copilot automatically remembers everything from all past chats | Memory captures only what you explicitly tell it to remember |
| Memory makes Copilot aware of all your historical work context | Memory holds the specific context items you provide |
| Instructions apply globally across all Copilot apps automatically | Instructions apply within the configured scope (user-level or agent-level) |
| Memory can be accessed by other users | Memory is personal — it applies only to your Copilot experience |
Managing your memory and instructions:
- Access memory settings through the Copilot settings or profile area
- Review, edit, or delete specific memory items at any time
- You remain in control of what Copilot retains about you
⚠️ Exam Trap: The exam will present scenarios where a user expects Copilot to "remember" something from a past conversation that they never explicitly added to memory. The correct answer is always that Copilot does not have automatic total recall — the user must deliberately configure memory for context to persist. This is a feature of transparency and privacy, not a limitation.
Reflection Question: A consultant wants Copilot to always be aware that she works primarily with healthcare clients and should use HIPAA-compliant language assumptions in any drafting help. What is the correct Copilot feature to use, and what action should she take?
