AI Remembers the Conversation. Does It Remember the Decision?
By Ravindra Dastikop
You return to an AI conversation after a few days.
“Continue with the plan,” you type.
The response arrives quickly. It sounds familiar. Then you notice something: it includes an option you rejected and leaves out a condition you added later.
The conversation has continued. Your decision has slipped.
This is an illustrative situation, not a reported product test.
Recognize: a conversation contains several versions of your thinking
Imagine planning a community reading event.
At first, you consider an evening programme. Later, volunteers prefer Sunday morning. You discuss charging an entry fee, then agree to keep entry free. An outdoor venue remains possible, but nobody has confirmed it.
All these statements belong in the conversation. They do not have the same status.
Some are suggestions. Some are decisions. Some remain questions.
When you ask AI to continue, the useful starting point is the latest agreed position.
Understand: history and working state serve different purposes
Conversation history records how the discussion developed. A working state records where things stand now.
AI may use information available in the conversation to prepare its response. Depending on the tool and its settings, it may also have other memory features. Neither should be treated as a guaranteed record of approved decisions.
A summary can preserve the main topics while losing the distinction between “we considered this” and “we agreed to this.”
That distinction changes the next action.
For the reading event, a useful working state might look like this:
Example: current event decisions
Item
Current position
Status
Event time
Sunday morning
Agreed
Entry
Free
Agreed
Venue
Outdoor space under consideration
Unconfirmed
Evening programme
No longer being pursued
Rejected
This small record gives the next response a clearer foundation.
Connect: give the next task a checked starting point
Before asking AI to draft the invitation, ask it to identify the current position.
Try this prompt with a short, nonconfidential conversation:
Create a working-state note from this conversation. Separate confirmed decisions, rejected options, constraints and unresolved questions. Include the statement supporting each confirmed decision. If later messages conflict, flag the conflict. Do not turn suggestions into agreements or fill gaps with assumptions.
Read the result yourself.
Correct anything that misrepresents the discussion. Then ask:
Using the checked working-state note below, draft the event invitation. Leave placeholders for unconfirmed details. Do not reintroduce rejected options.
The note becomes a bridge between the discussion and the next piece of work.
A small action you can try today
Choose one AI conversation in which your thinking changed.
Ask for the working-state note. Check whether it correctly captures one decision, one rejected option and one unresolved question.
Then use the corrected note for a single follow-up task.
The accomplishment: a usable next output that respects your latest decisions and makes unfinished matters visible.
This is a proposed exercise; its value depends on what your own check reveals.
The useful insight
As a conversation grows, keeping track of its current position becomes part of the work.
A short, checked decision record can help you resume with greater clarity—and help another person understand where to begin.
Your turn
When you resumed an AI conversation, what had it lost: a decision, a constraint or the reason behind your choice?
Share one example without private details. A specific mismatch gives us something useful to examine in a future post.
“I Can Do It When You Show Me.” What Happens When You’re Not There?
Turning digital assistance into growing independence.
By Ravindra Dastikop
This is an illustrative story. Suresh and his daughter are fictional;
the proposed practice is not a reported learning outcome.
Every month, Suresh asked his daughter to help him download an electricity bill.
She would open the website, find the bill, and save it on his phone.
“There,” she would say. “It’s easy.”
It was easy while she was doing it.
The following month, Suresh would ask again.
One evening, he explained:
“I recognise the screen when you show me. But I don’t know
what to do when it looks different.”
That changed the problem.
See differently
Suresh could follow familiar steps. What he lacked was a way to decide
what to do when a step no longer matched.
A changed button, an unexpected message, or a downloaded file he could
not find could interrupt the whole task.
The next learning opportunity was inside that interruption.
The hidden assumption
If someone has seen the steps, they can repeat the task independently.
A demonstration can help. But watching another person complete a task
does not necessarily reveal how they recognise the right action.
Suresh needed opportunities to notice, choose, and check for himself.
PO: The helper is not allowed to touch the phone.
Use this as a temporary learning experiment.
What changes when the learner holds the device and makes every choice?
His daughter could describe the purpose of a step, ask what he noticed,
and give him time to respond.
If he wanted her to take over, they could pause the experiment.
Independence would remain something he chose to develop.
Explore with AI
Together, they could ask an AI assistant to help structure the practice:
REUSABLE COACHING PROMPT
Help me support a beginner who wants to download an electricity bill
independently. The learner will hold the phone. Give me one coaching
question at a time, starting with what they want to accomplish and
what they can see.
Do not invent button names or assume the website layout. Help them
explain their choice before acting and check the result afterwards.
If the screen differs from expectations, ask for a description.
Never request passwords, OTPs, account numbers, or payment details.
AI could help the daughter slow down and ask clearer questions.
They would check each suggestion against the actual screen.
Instead of saying, “Press that button,” she might ask:
“Which option looks connected to viewing your bill?
What makes you think so?”
After the action:
“What changed? Does that bring you closer to the bill?”
Move from following steps to recognising progress
They could divide the task into three purposes:
Three checks for completing the task
Purpose
What Suresh would check
Reach the right service
Is this the official provider website or app they previously verified?
Find the bill
Does the displayed bill match the intended month and account?
Save and retrieve it
Can he open the downloaded bill from his phone’s files?
The accomplishment would include finding the saved file again.
A download message alone would not complete his intended task.
Where could this generate its own next possibilities?
Suppose Suresh downloaded the bill but could not find it.
That difficulty could generate the next small lesson:
“Where do files go on my phone?”
Once he understood that, he could practise finding another document
he had already saved.
His own questions could begin shaping the learning:
How do I recognise a downloaded PDF?
How can I give it a name I understand?
How do I find it next month?
The situation would acquire the capacity to generate its own next
possibilities when each attempt revealed a question the learner
wanted to answer next.
Accomplish something that matters
Choose one digital task you want to do yourself.
Practise with a willing helper:
Keep the device in your hands.
Say what you want to accomplish.
Explain what you notice before choosing.
Check what happened after each action.
Record one useful discovery in your own words.
Then try the task again later, with help available if needed.
Look for a concrete result:
Can you complete the task, explain how you know it worked,
and recognise when you need help?
Your turn
Which digital task can you do when someone guides you—but
would like to manage yourself?
That is a useful place to begin.
Share the task in the comments without personal account details.
What part would you like to understand better?
The Customer Asked for Something You Don’t Sell. What Could That Become?
Discovering a small business possibility inside an unanswered request.
By Ravindra Dastikop
This is an illustrative story. Kavita and her shop are fictional;
the proposed experiment is not a reported business outcome.
At her neighbourhood tailoring shop, Kavita heard a familiar question:
“Can you help me decide what to do with this sari?”
The customer did not need a blouse stitched. She had brought an old
sari she loved but rarely wore.
Kavita explained the services she offered. The customer folded the
sari and left.
Later, Kavita wondered:
Had someone asked for a service—or revealed a possibility
she had never considered?
See differently
A small business usually records what customers buy.
But what happens to the things customers ask for that the business
cannot yet provide?
“Can you make this easier to carry?”
“Could you teach me to maintain it?”
“Can you help me choose?”
Some requests fall outside the business. Others may reveal a useful
extension of what it already does.
The challenge is to distinguish them without spending heavily on
every idea.
The hidden assumption
My services define which customer requests matter.
This helps a business stay focused. But it can also make unfamiliar
requests disappear before the owner understands them.
Kavita had heard “alteration request.”
The customer might have meant:
“I want to keep something meaningful and find a way to use it.”
Those are different starting points.
PO: Let the unanswered requests write the next service menu.
Take this as a provocation to explore.
What would Kavita need to learn before adding anything to her menu?
She could begin by recording the next few requests she could not
fulfil, without names or personal details.
For each, she could note:
What did the customer ask?
What were they trying to accomplish?
Why could the shop not help?
What would she need to ask before suggesting a solution?
Explore with AI
Kavita could use an AI assistant to organise these observations.
REUSABLE PROMPT
I run a small tailoring shop. Below are anonymised requests I could
not fulfil. Group them by what customers appear to be trying to
accomplish. Separate their actual words from your interpretations.
Suggest three small service experiments using my existing skills
and equipment. For each, identify what I must ask customers, what
could make it impractical, and the smallest way to test it. Do not
treat requests as proof that people will pay.
My existing skills and equipment: [add details]
Requests: [paste anonymised notes]
AI might suggest a “sari possibilities consultation”: a short
conversation about whether an unused sari could become something
the customer would wear or use.
That is a hypothesis to examine.
Fabric condition, sentimental value, labour, cost, and the customer’s
expectations would determine whether any proposed transformation
made sense.
Move from an idea to a small test
Kavita could invite one interested customer to explore the possibilities.
Before cutting anything, she would:
Ask what the customer wanted to preserve.
Examine the fabric and explain its limitations.
Sketch two feasible options.
Estimate the price and delivery time.
Obtain agreement on the exact work.
The customer might choose an option, request a change, or decide
to keep the sari intact.
Each response would help Kavita understand the request more accurately.
Where could this generate its own next possibilities?
Suppose the customer liked one idea but said:
“I wish I could see how it would look before deciding.”
That response could reveal the next useful improvement: a clearer
sketch or a sample made from spare material.
If Kavita used an AI-generated visual, she would need to explain
that it was an illustration and check that it did not promise details
she could not produce.
The service could develop through a simple cycle:
Customer request → small proposal → customer response →
better next proposal.
The customer’s participation would help shape what the business
offered next.
Accomplish something that matters
This week, capture five requests your business usually
turns away.
Choose one that fits your capabilities and explore it with a
willing customer.
Look for an observable accomplishment:
Can you reach a clear, feasible, priced agreement that helps
the customer do something they value?
A paid trial would provide stronger evidence than enthusiasm alone.
A declined proposal could still reveal why the service needs
changing—or why it should stop.
Your turn
What do customers repeatedly ask your business for that
you do not currently offer?
Write down their exact words before deciding what those words
might become.
Share one request in the comments, without customer names or
identifying details. What would you need to learn before testing it?
Retired from the Job. What Could Your Experience Become?
Finding a new use for what you already know.
By Ravindra Dastikop
The following is an illustrative story. Meera is a fictional
character; the experiment is an invitation to try, not a reported outcome.
For thirty-two years, Meera managed purchasing for a small manufacturing company.
She knew which questions exposed an unreliable quotation, when a low
price concealed a costly condition, and how to disagree with a supplier
without damaging the relationship.
After retirement, people often asked her:
“What will you do to keep busy?”
Meera had a different question.
“Could something I know still help someone?”
See differently
Retirement ends a particular role. It does not automatically reveal
where the experience gained in that role might become useful next.
Much of Meera’s knowledge lived inside ordinary incidents: an urgent
delivery, a confusing invoice, a promise that needed checking.
She had never thought of those incidents as something another person
could learn from.
The hidden assumption
To contribute professionally, I need another professional position.
That assumption made her choices feel narrow: return to employment,
start a consultancy, or leave that part of life behind.
What other possibilities could she generate?
PO: Retire from the position. Keep one useful question open.
Meera chose a question she had answered many times:
“What should you check before accepting a quotation?”
She did not commit to a business or a course. She decided to explore
whether one small piece of her experience could help one person.
Move from experience to possibility
She described a fictional purchasing situation to an AI assistant,
using no former employer documents or confidential details.
Then she used this prompt:
REUSABLE PROMPT
Interview me about how I would evaluate a supplier’s quotation.
Ask one question at a time. Help me uncover the reasoning behind
my decisions. Afterwards, draft a one-page checklist for someone
making their first purchase. Mark anything you add that did not
come from my answers, so I can review it.
The conversation brought out questions she usually asked almost automatically:
Does the quotation cover everything we requested?
What could change the final cost?
What happens if the delivery is late?
Which promises need to be recorded in writing?
AI helped her put that reasoning into words. Meera checked whether
the resulting checklist reflected her experience.
Where could this generate its own next possibilities?
The interesting moment would come when someone tried the checklist.
Suppose a first-time buyer said:
“I understand the questions. But how do I ask them without sounding suspicious?”
That response could reveal a second useful contribution: examples
of polite questions to send a supplier.
A small cycle becomes possible:
Share one useful insight → observe its use → hear the next
difficulty → develop the next contribution.
Meera would not need to predict an entire project before beginning.
Another person’s experience could help shape what came next.
Accomplish something that matters
Try a modest experiment this week:
Recall one situation you learned to handle well.
Ask AI to interview you about how you handled it.
Turn the conversation into one useful page.
Offer it to someone who wants that help.
Ask: “What did this help you do—and what remained difficult?”
The first accomplishment is concrete:
one person can take a useful next step with something you helped create.
Look for an observable result: a question they could ask, a choice
they could make, or a task they could complete. Record what helped
and what still needs work.
Continuing is a choice. So is stopping after that single contribution.
Later life also makes room for rest, curiosity, friendship, and interests
pursued for their own sake.
Your turn
What do people still ask your help with—even though it was
never written in your job description?
That may be a starting point for a later-life possibility.
Share one example in the comments.
What might someone else be able to do with that experience?
Edward de Bono’s Concept Fan: Are Your Ten AI Ideas Really One?
Old Knowledge. New Possibilities.
AIYING · By Ravindra Dastikop
You ask AI for ten ideas.
The wording changes. The examples change. Yet several suggestions may still rely on the same approach.
For this edition, we revisit Edward de Bono’s concept fan: a structure for exploring alternatives at different levels.
The past practice: organise the search for alternatives
De Bono’s concept fan belongs to his toolkit for deliberate creative thinking. His own Serious Creativity article refers to using it to organise concepts; the De Bono Group also identifies concept extraction and the concept fan among its lateral-thinking tools. These sources establish attribution and use, rather than proving effectiveness in every setting. [1][2]
For this activity, we will use a simplified working structure:
Four levels of the working structure
Level
Question
Purpose
What are we trying to accomplish?
Direction
What broad route could help achieve it?
Concept
What general mechanism could serve that direction?
Idea
What specific action could put that mechanism into practice?
The useful movement goes both ways: from a particular idea back to the approach behind it, then outward to other ways of achieving the purpose.
PO is not required for this exercise. The concept fan supplies the structure.
The pattern beneath the tool
When several ideas lead down the same route, step back to the purpose and explore other routes.
This is our practical interpretation for the adaptation.
GenAI’s proposed contribution is to help classify suggestions, expose repeated mechanisms, and develop branches we have not explored.
It can also create the appearance of variety by giving similar ideas different labels. We therefore need to inspect the relationships, not count the suggestions.
What must hold: The purpose must reflect the actual need, and each branch must have a plausible connection to it.
Where the pattern breaks: Categories can overlap, important routes can be omitted, and an orderly map can contain impractical ideas. The map does not establish effectiveness.
Begin with a purpose that leaves room to think
Consider this fictional request:
“How can we get more people to attend our Sunday morning clean-up?”
Attendance may matter. But suppose the organiser’s real concern is accumulated litter.
A broader purpose might be:
“Reduce litter in our neighbourhood with a manageable burden on volunteers.”
That opens possibilities beyond one event.
However, broadening the purpose is a decision. If bringing neighbours together is itself essential, removing the gathering could lose something important.
Confirm what must be preserved before expanding the search.
Build a small fan
Here is an illustrative map—not a report of solutions already tested.
Purpose: Reduce neighbourhood litter without overloading volunteers.
Illustrative concept fan
Direction
Concept
Concrete idea
Prevent some litter
Make disposal easier at a recurring litter point
Ask the responsible authority about a suitable bin location and collection arrangements
Increase participation
Lower the time commitment
Offer a fifteen-minute clean-up option
Distribute the work
Give small areas clear ownership
Invite willing groups to adopt a short stretch
Improve collection
Coordinate with existing services
Agree how collected bags will be removed before organising the event
These proposals require local checks. A bin without collection arrangements, for example, could create another problem.
The table’s value is that the ideas act through different mechanisms. The directions are not claimed to be exhaustive or mutually exclusive.
Use GenAI to examine your first list
Write down three ideas of your own before consulting AI.
Then use this prompt:
Help me build a simplified concept fan inspired by Edward de Bono’s method.
Situation: [Describe.] Purpose: [State the desired outcome.] What must be preserved: [Include important social or practical values.] Constraints: [Time, money, access, permissions.] My first three ideas: [Insert.]
First, identify the general mechanism behind each idea. Flag ideas that use substantially the same mechanism.
Suggest up to three broader directions. Explain how each could contribute to the purpose. Ask me to confirm them before expanding.
After confirmation, propose two concepts under each direction and one concrete idea under each concept.
Show the chain: purpose → direction → concept → idea.
Identify overlaps, assumptions, missing information, and proposals outside our control.
Do not claim the map is exhaustive or that a plausible idea will work. Keep generation separate from evaluation.
The branch counts are instructions for this exercise, not a rule attributed to de Bono.
Check whether the fan contains real alternatives
For each proposed branch, ask:
Connection: How could this contribute to the purpose?
Difference: Does it work through a different mechanism, or repeat another branch?
Specificity: Is the idea concrete enough to try?
Conditions: What information, permissions, resources, or cooperation would it need?
Compare the map with your original three ideas. Record one useful distinction it added—or record that it did not add anything useful.
Choose an action using criteria suited to the situation, such as likely contribution, cost, accessibility, volunteer effort, and feasibility.
Test one branch
For the fictional clean-up, suppose the group selects the fifteen-minute option.
Before organising it, ask potential participants:
“Would a fifteen-minute option change whether you could participate? What would still make it difficult?”
Their answers may support the proposal—or reveal that equipment, accessibility, timing, or bag removal matters more.
If the group proceeds, record:
Before the test
After the test
Expected participation
Actual participation
Planned volunteer effort
Actual effort
Intended litter removal
What was removed and how collection worked
Assumptions about access
Difficulties participants reported
A small test can inform the next decision. It cannot establish a general rule about community participation.
If the activity has not happened, label the result “Not yet tested.”
Your accomplishment
Finish with:
A purpose you can explain.
A map containing genuinely different approaches.
One selected action with a reason for choosing it.
A test record—or a clearly labelled plan if it has not happened yet.
You can build the fan entirely on paper. Judge AI assistance by whether it helps reveal useful alternatives, not by how many branches it produces.
Open the next possibility
A test may reveal a constraint that belongs in the next version of your map.
For example, if people have time but cannot reach the meeting point, reducing the time commitment may not address the real obstacle. Revisit the purpose and explore another route.
Sources checked on 16 November 2026. The four-level structure is presented as a simplified teaching explanation. The neighbourhood example, prompt, and testing procedure are proposed applications, not evidence of a completed experiment.
PATTERNS OF THE PAST · V12.0 · #002
Peter Drucker’s Feedback Analysis: Can AI Help You Learn from Your Own Decisions?
Old Knowledge. New Possibilities.
AIYING · By Ravindra Dastikop
A decision worth learning from
You introduce a new activity, change a work process, or accept a responsibility.
Later, you decide whether it went well.
But what exactly did you expect before you began?
For this edition, we revisit Peter Drucker’s feedback analysis: preserving expectations so that experience has something specific to challenge.
The past practice: record expectations, then compare
Drucker described feedback analysis as a way to investigate personal strengths. Record important decisions or actions alongside their expected results, then compare expectations with outcomes nine to twelve months later. The September 12 entry in The Daily Drucker presents this as a repeated practice extending over years. [1]
The historical method therefore involves more than reflecting after an event. It preserves an expectation that can later be examined.
This edition adapts that structure. It does not claim that Drucker proposed an AI system or that the adaptation below has been validated.
The pattern beneath the practice
Preserve an expectation before acting. Compare it with evidence afterwards. Use the difference to question how you work.
The pattern card
Pattern element
What we preserve
Proposed GenAI adaptation
Expectation
A dated account of what you think will happen
Ask AI to identify vague wording and unstated assumptions
Action
What you decided to do
Keep the original decision separate from later changes
Result
Evidence of what happened
Organise observations without filling missing information
Comparison
Expected versus actual outcomes
Generate possible explanations and counter-explanations
Learning
A decision about future practice
Examine recurring differences across several records
What must hold: Expectations and results need enough detail to compare.
Where the pattern breaks: Results reflect circumstances, other people’s contributions, chance, and measurement choices. A successful outcome does not establish a personal strength; a disappointing result does not establish incapacity.
A present difficulty: repeating activity without accumulating learning
Consider a fictional teacher who introduces a new classroom activity.
Afterwards, the teacher remembers that students seemed enthusiastic. The activity enters next term’s lesson plan.
But the original purpose was to help students explain a concept independently.
Enthusiasm and independent explanation are different outcomes.
A decision record could preserve that distinction before the activity begins.
Make your first expectation card
Choose one modest decision with an observable result. Complete this before acting:
Field
What to write
Decision ID and date
D01; date recorded
Situation
What needs attention?
Action
What will I do?
Expected result
What specifically do I expect to happen?
Reason
Why do I expect that result?
Assumptions
What am I taking for granted?
Evidence
What will I observe or collect?
Review point
When should a meaningful result be visible?
Keep the original entry unchanged. Append revisions with dates.
For a short activity, you can rehearse this comparison within days or weeks. That shorter review is our adaptation, not Drucker’s nine-to-twelve-month interval, and it cannot substitute for examining longer-term development.
A worked fictional example
Suppose the teacher records:
Now imagine the following illustrative result:
Sixteen students completed the paired exercise. Nine met both criteria in the individual exit task.
These numbers are invented to demonstrate the comparison.
They create a useful discrepancy: successful participation exceeded demonstrated independent performance.
They do not establish why.
Possible explanations include insufficient practice, partners completing different shares of the work, or a changed example that introduced an additional difficulty. Those explanations need investigation.
Where GenAI could contribute
With several records, an assistant could help look for candidate patterns:
A particular approach works only under certain conditions.
Several apparent failures share an external constraint.
The evidence collected does not match the intended outcome.
These are possibilities to examine—not conclusions about your personality or competence.
GenAI can also produce a coherent explanation that the records do not support. Require it to show which entries support each interpretation.
Use this review prompt
Help me review these decision records using the expectation–result comparison associated with Drucker’s feedback analysis.
This is an AI-assisted adaptation, not a claim that Drucker designed this workflow.
For each record:
Preserve the original expectation exactly.
Separate observed results from interpretations.
Identify the difference between expectation and result.
State whether the evidence actually measures the intended outcome.
Suggest possible explanations, including contextual factors and explanations that challenge my preferred account.
Identify missing evidence. Do not fill gaps.
Across the records, propose candidate recurring patterns. Cite record IDs and identify counterexamples.
Do not infer a stable strength or weakness from one event, or present correlation as causation.
End with one question to investigate and one small change I could test.
Records: [Paste dated entries.]
Check the response against your records yourself. Use anonymised or fictional records for practice; exclude identifiable student, client, or colleague information.
Test whether AI adds value
Before using the prompt, write your own short comparison.
Then examine the AI-assisted review:
Did it reveal a discrepancy you had overlooked?
Did it identify a genuine counterexample?
Did it distinguish evidence from an attractive explanation?
Did it introduce an unsupported claim?
Keep a useful insight only if you can trace it to the records.
If you have no dated expectations, begin a prospective record today. Retrospective recollections may provide context, but label them as recollections.
You can complete the entire exercise on paper. The comparison with your own review helps you assess whether AI contributed anything useful in this instance.
Your accomplishment
Start with one complete decision record.
When the review point arrives, append the evidence, your comparison, and the next question. Until then, mark the outcome “Not yet observed.”
Over time, you can build a collection that helps you investigate how your expectations meet experience.
The first accomplishment is modest and concrete:
A decision you can learn from without rewriting what you originally expected.
Source checked on 16 November 2026. The classroom example, additional record fields, shorter rehearsal interval, and GenAI workflow are proposed adaptations. They are not reported research results or claims that Drucker specified this AI-assisted process.
The Commonplace Book: Can Your Saved Notes Help You Start Something?
Old Knowledge. New Possibilities.
AIYING · By Ravindra Dastikop
You saved the article because it mattered.
You highlighted a passage, copied a quotation, and bookmarked another useful explanation.
Weeks later, you face a problem. Somewhere in those notes is something that might help.
But where—and how?
For our first edition, we revisit the commonplace book: a historical practice of collecting and organising material for later reference. We will explore an adaptation in which GenAI helps examine connections between a small collection of notes and a present difficulty.
A practice from the past
A commonplace book was a place to gather selected material from reading, often organised under headings or through an index.
Harvard Library’s exhibition documents George Ripley’s nineteenth-century notebook, which combined excerpts, poems, observations, and reading notes. Entries included source references. The exhibition also describes John Locke’s indexing method for locating material entered across different notebook pages. [1]
Another notebook in the exhibition illustrates a practical difficulty: assigning pages to subjects in advance could leave some sections empty while others overflowed. Locke’s method addressed that allocation problem. He developed a method within an existing practice; he did not invent the commonplace book. [2]
The past offers both a useful practice and a warning: an organising system can become cumbersome. Ripley himself did not consistently follow the indexing scheme associated with his notebook. [1]
The pattern beneath the notebook
The useful connection is not simply that people once saved quotations on paper.
It is this:
Select something worth retaining, preserve its context, and make it available for another occasion.
That is our interpretation of the practice—not a claim that every historical notebook served the same purpose.
The pattern card
Pattern element
What we preserve
What we adapt
Selection
Someone decides what deserves attention
Begin with a small, deliberately chosen collection
Context
Material remains connected to its source
Keep source details beside each note
Organisation
Entries can be found and revisited
Ask AI to propose several possible groupings
Reuse
Earlier material can inform later work
Examine connections to one present difficulty
What must hold: The notes must contain enough context to interpret responsibly.
Where the pattern breaks: A collection can contain mistakes, outdated advice, and incompatible assumptions. Saving something does not establish its truth.
The new possibility
A conventional search might help you find a note containing a particular word.
In this experiment, ask GenAI to examine several notes together:
Which describe a similar relationship?
Which disagree?
Which might inform this situation?
Which connections are merely superficial?
GenAI’s proposed role is to generate candidate connections and explain them. You check whether those explanations fit the notes and the situation.
A convincing connection is still a hypothesis.
Begin with one present difficulty
Use a real, modest problem.
For example, consider this fictional situation:
A community reading group has lively discussions, but members rarely try anything suggested between meetings.
Before consulting AI, write your own first explanation and one possible response.
This creates a starting point for comparison.
Then select five notes that might be relevant. They need not share a keyword.
Use your own observations or short excerpts you can appropriately reuse. Keep quotations separate from paraphrases.
Build five note cards
Give each note an identifier:
Field
What to record
ID
N1, N2, N3, N4, or N5
Content
Short excerpt, paraphrase, or observation
Source
Author, title, page or URL; or the circumstances of your observation
Context
What situation the material originally concerned
Your comment
Why you retained it
Uncertainty
What needs checking
Keep the collection small enough to review yourself. A notebook or ordinary document is sufficient.
Ask AI to explore the connections
Use this prompt:
I am adapting the commonplace-book practice to explore a present difficulty.
My difficulty: [Describe it.] My first explanation and possible response: [Insert.] My five notes: [Paste N1–N5 with their source details.]
Work only from these notes when describing their contents.
Identify up to three candidate relationships across the notes. Cite the note IDs supporting each.
Distinguish shared mechanisms from similarities in wording.
Identify disagreements, missing context, and connections that may be weak.
Propose three different responses to my difficulty. Explain which notes inform each and where the adaptation introduces a new assumption.
Include the possibility that these notes do not adequately support a useful response.
Help me choose one small experiment using criteria we state explicitly.
Do not invent quotations, sources, historical claims, or evidence of effectiveness. Label your proposals as proposals.
Turn one connection into an idea card
Review the AI’s suggestions against the original notes.
Reject any connection you cannot explain in your own words.
Then complete:
For the fictional reading group, one possible adaptation might be:
End a meeting with each willing member choosing a five-minute action and deciding what observation to bring back.
This is an illustrative proposal. It is not a result established by the historical practice or by notes we have actually analysed here.
Test usefulness, not elegance
First compare the selected idea with your original response.
Did examining the notes add a relevant distinction, reveal a weakness, or produce a different action? Record what changed.
Then, where feasible, try the action.
Check
Evidence to record
Source fidelity
Do the cited notes actually support the interpretation?
Contribution
What changed from your initial idea?
Practicality
Can the proposed action be carried out within the constraints?
Outcome
What happened when you tried it?
Limitation
What remains uncertain or unexplained?
For the reading-group example, you might record how many members voluntarily chose an action, attempted it, and returned with an observation. Do not assume participation means learning—or that one successful meeting establishes lasting improvement.
If you have not run the test, write “Not yet tested.”
You can also perform the same exercise on paper by arranging the five cards and looking for connections yourself. Comparing that process with AI assistance may help you judge whether the assistant added value in this task.
Your accomplishment
The intended result is:
One source-linked idea card, one small experiment, and a record of what happened.
You may discover that the notes were useful. You may discover that your first idea was better. You may discover that you need different evidence.
Each result can guide the next inquiry.
Let the collection grow through use
Add one final note:
“I tried this because… What happened was… The next question is…”
Your collection now contains an observation about applying what you saved.
That observation can become material for a later decision.
Historical references checked on 16 November 2026. The pattern card is an editorial interpretation. The AI workflow, reading-group example, and experiment are proposed adaptations, not research-validated outcomes or reports of a completed test.