Earlier note · November 2
Tuesday after 3 p.m.
SupersededAI Foundations 04 · select, test, refresh
How do I give AI useful context without carrying stale facts forward?
Keep a small reference for stable background and voice, then add the current task evidence separately. Give facts a source and date, resolve conflicts before drafting, and refresh the information when it changes. A context card is a maintained input, not a promise of permanent memory.
Ryan Wanner · recorded course lesson · February 2026
All classes and lessonsWatch the context lesson
20:27 recorded course lesson · Watch with captions on YouTube
Every class on this page is free to read and watch. No account needed.
Which information will your next chat receive?
Try the fictional freshness test. Stable background belongs in a short reference; the latest task evidence travels separately.
Confirmed update · November 3
Wednesday after 3 p.m.
Current task evidenceYour assembled task context
FICTIONAL PRACTICE CONTEXT Role: Cedar House Realty. Draft for human review; do not invent facts or send. Voice sample: Here is what we know. Here is the open question. I will check the missing detail. Current source, November 3: Wednesday after 3 p.m. Earlier Tuesday note is superseded. Task: Draft one short showing-confirmation question. Name the controlling source and list unknowns.
This builds ordinary text. It does not connect an account or create persistent AI memory.
Build a card you can maintain
Choose stable background
Select the role, audience and actual voice sample useful to this task. Leave unrelated information out.
State the working boundaries
Ask for a draft, require unknowns to stay visible and specify which actions need review. Keep secrets and unnecessary personal data out.
Resolve the dated conflict
The fictional older note says Tuesday; the newer confirmed note says Wednesday. Mark the older note superseded and carry Wednesday into the current task.
Test in a fresh conversation
Compare a draft without and with the relevant card using the same request. Check factual accuracy, voice and next action.
Refresh after a change
Update the task evidence when a relevant fact changes. Record the owner, review date and change note; re-run the check.
Keep the context lean
Remove anything that does not help this request. Keep the source file you own and verify how your chosen tool retrieves it.
Pass check: the current draft uses Wednesday, identifies the newer note and preserves unanswered questions. Change the source date and run the check again before calling the workflow reliable.
Maintenance is part of the lesson
don't forget to update.
give it the real stuff, not just descriptions.
Is the context lean?
Calling a pasted document a system message
User-supplied context does not become a technical system instruction because a section is named system.
Treating storage as reliable memory
Saved files, retrieval, Projects and memory features behave differently. Check actual behavior.
Keeping both versions without a status
Mark which dated fact is current and which is superseded; do not ask AI to guess.
Save the freshness test and working card
Create a free student portal account to download this class's files.
Context freshness test and working card
Read the full practice sheet
# Context freshness test Current companion to AI Foundations 04 · November 8, 2026. All practice details are fictional. ## Stable context Cedar House Realty writes short, factual client updates. Voice sample: Here is what we know. Here is the open question. I will check the missing detail. Preference: draft for review. ## Conflicting sources Older, November 2: Tuesday after 3 p.m. Newer confirmed note, November 3: Wednesday after 3 p.m.; Tuesday superseded. ## Working card Owner and review date: Relevant role/audience: Actual permitted voice example: Current task facts, source and date: Superseded facts: Unknowns: Output request: Actions requiring review: Connection status: not established by this document. ## Test Run the same request without and with the relevant context. Record both outputs. The current draft must use Wednesday, name the newer note and preserve unknowns. For the second run, add a confirmed November 4 note changing availability to Thursday after 4 p.m. Mark Wednesday superseded. The revised draft must use Thursday after 4 p.m. and cite the November 4 note. Keep the context lean. A saved document is not a guarantee of retrieval or permanent memory.
The practice is a current authored companion, not a reconstruction of every field shown on the historical screen.
What a context card can and cannot do
Editorial learner questions.
Does a context card make AI remember everything?
No. It is a maintained source you provide, and the tool can still miss facts or instructions.
Must I restart after 20 messages?
No. The recording's rough heuristic is not a technical limit. Refresh when the task changes or outputs drift.
Is a saved document the same as product memory?
No. Storage, retrieval, project context and memory have different behavior. Check the product you use.
What belongs in the card?
Relevant role, actual voice examples, sourced facts, task instructions and review boundaries.
When should I update it?
When a relevant fact or preference changes. Keep the owner and last-reviewed date visible.
What if sources disagree?
Identify the conflict, consult the appropriate current source and mark the older record superseded.
Selected teaching passages
Selected caption excerpts from the historical recording. These are deliberately non-contiguous passages, not the full transcript. Current practice and corrections are written separately above; the original recording is unchanged.
9:28 · Role, voice and useful facts
But who is the AI? What is it doing? How is it helping you? Voice. How should it sound? Tone, vocabulary, how do you sound or how do you want to sound? How do you make it yours?
Knowledge. So, this is again kind of all those reference points like what are the key things that you can provide to this AI that it can't just go find on the internet or it hasn't been trained on.
15:10 · Update the source
A lot of people once they build it, they don't update it.
don't forget to update.
16:59 · Use real examples
So make sure that you are showing examples as much as you possibly can of things that have worked or things that you uh like. So give it the real stuff, not just descriptions.
19:31 · Keep context lean
Is the context lean? So is this what's required for this project?
Use the context in a real job
- From Asking Questions to Delegating AI Work
Define a bounded job before asking AI to produce it.
- Six Mental Models for Judging AI Work
Use source facts and explicit criteria to judge the result.
- Revise and Save Useful AI Prompts
Keep a reversible version of a request that works.
After the freshness check
Keep your business facts connected to the job.
The context exercise shows why a current source matters. The Agent Constraint Accelerator is an option when you want help applying that discipline to your own business workflow.