Mauricio Tapia.
← AI course

Chapter 03 · Foundations

AI needs to know what stays and what changes

Five layers separate lasting rules from today’s work, so everything does not end up in one endless prompt.

When an instruction mixes purpose, old data, open items and today's decisions, it becomes difficult to know what still applies. We can organize context into five layers:

  • Permanent instruction: a lasting role and rules. For example, “separate facts from inferences and do not send messages without approval.”
  • Memory: stable background and agreements that help interpret a task. They need a source and a way to be corrected.
  • Task: the concrete action now. For example, “summarize the incidents received this week.”
  • Campaign: a time-bound objective with several stages. For example, “reduce response times over eight weeks and review progress each Friday.”
  • State: the current picture: what is open, what changed and what remains pending.
A customer-service example

The permanent rule is to cite a source for every claim. Memory holds the approved definition of a “resolved case.” The campaign seeks to improve response times over eight weeks. State tells us that four cases are open today. The current task is to prepare a meeting summary.

These layers let us update what changes without losing useful principles. They also expose conflicts: if an older record says one thing and a newer one another, the system should show both pieces of evidence, not invent a reconciliation.

Your turn

“Four cases remain open today, and one was resolved yesterday.” Which layer fits best: A) permanent instruction, B) memory, C) campaign or D) state?

If you came from LinkedIn, share your answer there before opening the explanation.

Show the explanation

D) State. It describes the current situation and may change tomorrow. A campaign describes a time-bound objective; permanent instructions set the rules; memory retains more durable context.