Save version one
Use the four approved fictional property facts. Ask for an email summary under 100 words with one invitation to request a showing. Save both request and draft.
AI Foundations 03 · archive + current practice
How do I improve a useful prompt without losing what worked?
Save the request and its result, identify one defect, and change one instruction. Re-test with the same facts and criteria, then keep the better version with a short explanation. This current exercise accompanies an archived course recording.
Ryan Wanner · original course recording published January 2026
All classes and lessonsThe original lesson, with its history intact
Archive edition. The video teaches the older HOME framework. Current practice below uses plain instructions, reliable facts, review and versioning. It does not teach HOME or CRAFT as a current method.
Every class on this page is free to read and watch. No account needed.
Preserve version one. Make version two explainable.
Fictional approved fact sheet: two bedrooms · one bathroom · covered parking · showing by appointment. No dimensions, neighborhood claims or accessibility certification are supplied.
These are requests, not generated results. Run the comparison in your approved workspace and record what actually happens. Edits stay in this tab.
Use the same acceptance check for both results
Under 100 words. Supported property facts only. One showing invitation. Missing information separate. No invented buyer assumptions. Draft remains unsent.
A small, reversible improvement
Inspect the result
Mark invented features, missing facts and tone problems. Check every claim even when the writing sounds polished.
Change one instruction
Add the missing restriction or output requirement. Write why this edit should repair the defect; preserve the earlier version.
Compare like with like
Run the revised request with the same facts. Check whether it fixes the defect without losing accurate information or the useful next action.
Keep a reusable version
Replace the property facts with an explicit input field. Save a representative accepted result, review checklist and change note. Re-test when the task or tool changes.
The parts of the archive worth carrying forward
So you want to save your best prompt.
I highly recommend keeping some type of versioning
Build whole bridges that get you what you want
Changing everything at once
Make one targeted revision so you can tell what helped.
Losing the previous version
Save the request and result before replacing them.
Using a formula to cover missing facts
No wording technique supplies an absent property fact.
Keep the revision history
Create a free student portal account to download this class's files.
Download the revision log and fact sheetRead the log and checks without downloading
# Prompt revision log Current practice companion to the archived AI Foundations 03 recording · November 8, 2026. No acronym framework is required. ## Approved fictional property facts Two bedrooms. One bathroom. Covered parking. Showing by appointment. No dimensions, school information, neighborhood claims or accessibility certification are supplied. ## Version history | Version/date | Request | Observed defect | One change | Result against same criteria | Keep or revert | |---|---|---|---|---|---| | 1 | | | | | | | 2 | | | | | | ## Acceptance checks Under 100 words. All four facts preserved when relevant. No invented amenity, location claim or buyer assumption. One invitation to request a showing. Draft remains unsent. ## Reusable request Task: Source facts field: Output format: Limits: Review criteria: Accepted example: Change note: Keep old versions so a failed change can be reversed. Re-test when the task or tool changes.
This is a current authored companion. It does not claim to recover the original prompt library mentioned in the recording.
Questions about the archive and current practice
Editorial learner questions.
Why does the video teach a framework the page does not?
It is an archive edition. The current exercise uses direct instructions, source facts and review; it does not teach the older acronym method.
Do I need a special prompt formula?
No. State the work, available facts, required format, limits and review criteria.
What should I save?
The request, its required inputs, a representative accepted output, review checks and a short change note.
Should I replace a prompt for every new model?
Re-test the same representative task first. Keep the prior version so changes are reversible.
What if information is missing?
Ask for it or mark it unknown. A clearer sentence cannot repair an absent fact.
Selected passages on saving and revision
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.
7:17 · Use understandable language
one of my favorites is actually to tell it to rewrite this in a way that would be understandable to a fifth grader, right?
the people who are really making an impact are making things as simple as they possibly can.
10:47 · Save a working request
So you want to save your best prompt. So when you find something that worked like a charm, got you that output, you've iterated it, and now you're here at the output, you can then ask the AI to create the prompt that basically shortcuts that route, right?
11:22 · Keep a version history
So I recommend, you know, keeping some type of simple format.
I highly recommend keeping some type of versioning because you may end up changing the prompt and then being like, "Oh, this is now performing much worse. Let's go back and look at what changed, how it changed, and what we need to revert back to.
12:09 · Finish one useful workflow
Don't try to build a bunch of things because then you never complete the bridge, right? Build whole bridges that get you what you want and then go back and start the next one.
Make the next request stronger
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.
Build a Context Card That Stays Useful
Keep relevant facts current and resolve conflicting notes.
From a saved request to useful work
Keep the version that survives a real test.
A saved prompt is one part of a working process. Review the Agent Constraint Accelerator if you want help deciding which process deserves attention first.