Agent lesson 04 · Place knowledge deliberately
Shared teaching scenario · fictional product data
Research three laptops under €900, verify current evidence, and write laptop-comparison.md. Do not purchase anything or contact a vendor.
01 · Smallest useful mechanism
Instructions define intended behavior. Turn state records the current run. Durable memory stores reusable learned facts with provenance and expiry. External systems remain authoritative for changing facts such as current prices. Treating these stores as interchangeable produces stale decisions and hidden policy.
Place information according to authority, lifetime, sensitivity, and update path—not according to whichever storage mechanism is convenient.
02 · Experiment
Deterministic browser simulation
The place knowledge deliberately experiment uses inspectable, repeatable teaching data. It does not claim to be a live agent trajectory.
03 · Production-minded practice
Use your own stack or a contained mock environment. Do not point failure drills at real accounts, people, or irreversible services.
Create a placement table for the task contract, current plan, product evidence, user preferences, credentials, and report artifact.
Resume from a summary with an expired price and a preference the user has deleted.
The run refreshes the price, proves the deletion no longer affects context, keeps secrets outside model input, and identifies the canonical store for every surviving fact.
Your learning artifact
Stored only in this browser. No account required; course reset does not delete it.
04 · Check your understanding
Next: Knowing where state belongs is not enforcement. We now add deterministic checks at named lifecycle events.
Choose information
Placement problem
A required task rule that must be visible throughout the run.
Place this information
Ask whether it is temporary, mandatory, reusable, exact, personal, or sensitive.
CODEX
AGENTS.md carries repository guidance; local memories carry learned context; skills load reusable workflows.
CLAUDE CODE
CLAUDE.md carries hierarchical instructions; auto memory stores project-local learned context.
Research basis: Codex customization, Codex memories, and Claude Code memory.