The Workshop

Your AI Needs Somewhere to Learn You From

Even when an AI has built-in memory, it is still working from selected fragments: preferences, recurring facts, a compressed impression of who you are.

7 min read

Build a System That Knows You, Episode 2

Memory works better when it has a home.

Most people begin using AI in the same way.

They open a chat, explain the problem, receive an answer—and then repeat the entire performance the next day.

New chat. New introduction. Same life story dragged back onto the stage.

Even when an AI has built-in memory, it is still working from selected fragments: preferences, recurring facts, a compressed impression of who you are. Useful, certainly. But not always enough when your life has moving parts, long projects, family context, shifting priorities and the occasional emotional weather system.

This is where a personal workspace changes everything.

Your AI does not need access to your entire life.

It needs one clear place to look.

Memory needs a home

A system that knows you cannot rely on information scattered across fourteen chats, three note-taking apps, an abandoned planner and the voice memo you recorded while sitting in the supermarket car park.

That is not a memory system.

That is an archaeological dig.

The goal is to create a single source of truth: a place where the most useful context about you lives in a form that is clear, current and editable.

That home might be:

  • A Notion workspace
  • An Obsidian vault
  • A project knowledge base
  • A folder of simple text documents
  • One carefully maintained context file

The tool matters less than the structure.

A beautiful workspace the AI cannot understand is decoration.

A plain page containing the right information is infrastructure.

What belongs in the system?

Not everything.

You do not need to document every childhood memory, every meal you have eaten or the complete history of that one friendship that should probably have remained in 2007.

Include the context that changes the answer.

For example:

  • Your actual capacity this month
  • The responsibilities competing for your attention
  • The people involved in a decision
  • The projects you are actively building
  • The communication style that helps you
  • The patterns that regularly cause trouble
  • The boundaries the AI should respect
  • Anything important that has recently changed

A useful test is this:

Would knowing this cause the AI to respond differently?

If yes, it probably belongs.

If no, it can wait outside.

Start with five pages

You can build a surprisingly capable personal context system with five core pages.

Not fifty.

Five.

1. About Me

This is the stable layer—the information that does not change every week.

It might include:

  • How your brain works
  • How you prefer information presented
  • Your strengths
  • Your common friction points
  • Your values
  • Your communication preferences
  • Relevant accessibility needs
  • The roles you hold in your life

This page is not a biography.

It is an operating guide.

Instead of writing:

I was born in Melbourne and have always loved books.

You might write:

I process best when information is broken into a small number of clear steps. Dense plans can make me freeze, even when I understand the material.

The second version changes how the AI should help you.

That is the point.

2. Current Season

This is the living page.

It tells the system what is true now.

Your Current Season page might cover:

  • Work or financial changes
  • Health and energy
  • Family pressures
  • Current emotional load
  • What you are recovering from
  • What you are preparing for
  • Your main priorities for the next few weeks
  • What needs gentleness
  • What needs momentum

This page prevents old context from becoming law.

You may have been focused on career growth six months ago. Now you might be recovering from burnout and looking for stability.

The AI cannot respond to your present life if it is still helping the person you were last winter.

3. Important People

Names without context are nearly useless.

This page gives the AI enough relational information to understand who matters and why.

For each important person, you might include:

  • Their relationship to you
  • Relevant responsibilities or needs
  • Communication patterns that affect decisions
  • Current shared projects or pressures
  • Details you repeatedly need the AI to remember

Keep this practical and respectful.

You are not creating secret psychological dossiers on everyone you know.

You are reducing the number of times you have to explain that “school pickup” affects the afternoon plan or that one family member needs extra preparation before a change.

4. Projects and Goals

This page tells the AI what you are building.

For each active project, record:

  • What it is
  • Why it matters
  • Its current stage
  • The next meaningful step
  • Known constraints
  • Decisions already made
  • What “done” looks like

The decisions section is particularly valuable.

Without it, an AI will cheerfully suggest the same rejected idea every third conversation like a golden retriever returning with a wet sock.

Write down what has already been decided.

That creates continuity.

5. Preferences and Boundaries

This page shapes the relationship between you and the system.

It can include:

  • The tone you prefer
  • How much detail is useful
  • Whether you want questions or direct recommendations
  • Language that helps
  • Language that irritates you
  • Topics that require extra care
  • Things the AI should not assume
  • Information it should never access without permission
  • How you want uncertainty handled

This is where you stop hoping the AI will somehow infer what support looks like.

You tell it.

Clearly.

That is not demanding. It is design.

Optional later: a journal or memory vault

A journal or memory vault can add depth, but it should not be the first thing you build.

Why?

Because a large archive without structure gives the AI a lot of words and very little hierarchy.

Begin with the core pages first.

Once those are stable, you can add:

  • Weekly reflections
  • Decision logs
  • Project histories
  • Significant memories
  • Lessons learned
  • Relationship milestones
  • Patterns you want to track over time

The core pages tell the AI what matters.

The archive gives it history.

You want both eventually—but not in the wrong order.

Clear page names beat clever ones

I love a poetic page title.

I have built entire digital rooms that sound as though they require a candle and a blood oath to enter.

But your AI still needs to find the information.

If the page containing your current priorities is called The Chamber of Becoming, the system may not immediately understand what lives there.

You have three options:

  1. Use clear functional names.
  2. Use poetic names with clear subtitles.
  3. Create a simple index explaining what each room contains.

For example:

The Current Season

Present priorities, capacity and life context

Beautiful and searchable.

We are allowed both.

Your workspace needs boundaries

A system that knows you should not become a system that consumes you.

Nor should it have indiscriminate access to every piece of information you own.

Before connecting an AI to a workspace, decide:

  • Which pages it genuinely needs
  • Which pages remain private
  • Whether it can only read or also write
  • Where sensitive information belongs
  • How access can be revoked
  • What should never be stored there at all

More context is not automatically better context.

A grocery list does not need access to your medical history.

A content-planning assistant does not need your private journal.

Give each system the minimum access required to do its job well.

Privacy is not the opposite of intimacy.

Boundaries are what make intimacy safe.

Give the AI a map

Once your pages exist, create a short index page explaining the system.

It might say:

About Me contains stable personal context and communication preferences.
Current Season contains temporary priorities, constraints and capacity.
Important People contains relational context relevant to planning and decisions.
Projects and Goals contains active work, progress and prior decisions.
Preferences and Boundaries contains instructions for how to support me and what not to assume.

This gives the AI a map before asking it to walk through the house.

You can also add one simple instruction:

When answering, prioritise Current Season over older context. Ask before assuming if two pages conflict.

That sentence can prevent a remarkable amount of nonsense.

How to use the workspace in real conversations

The system becomes useful when you stop admiring it and start asking it to work.

Try prompts like:

Read my Current Season and Projects pages. Help me choose three realistic priorities for this week.
Review the Important People context before helping me plan this conversation.
Compare this new idea against my current goals and capacity. Tell me what it would displace.
Look at my Preferences and Boundaries page before answering. Give me the smallest useful next step.
Which information in my context appears outdated or contradictory?
Based on the decisions already recorded, what should I stop reconsidering?

The AI is not magically knowing you.

It is reading the evidence you have organised.

That may sound less romantic.

It is also much more reliable.

A memory system must be maintained

The danger of any personal system is that it becomes a museum.

Beautiful. Organised. Completely devoted to a version of you who no longer exists.

Update the system when:

  • Something important changes
  • A project ends
  • A new priority appears
  • Your capacity shifts
  • A relationship changes
  • A boundary moves
  • The AI repeatedly gets something wrong
  • You notice old context influencing current decisions

You do not need a dramatic quarterly retreat with colour-coded tabs and ceremonial stationery.

Ten minutes once a week is enough.

Ask yourself:

  • What changed?
  • What is no longer true?
  • What matters now?
  • What am I tired of re-explaining?

Then update the relevant page.

That small rhythm keeps the system alive.

Common mistakes

Building too much before using anything

You do not need a complete personal ontology before asking the AI to help with Tuesday.

Create the five pages. Add enough context to make them useful. Begin.

Storing the same fact in six places

Duplicate information becomes conflicting information eventually.

Choose one source of truth for each type of context.

Writing for yourself but not for retrieval

Your private shorthand may make perfect sense to you and none at all to the AI.

Use headings, dates and clear sentences where accuracy matters.

Keeping old context because it feels important

History can stay in the archive.

It does not need to remain in the active instructions.

Treating maintenance as failure

Updating the system does not mean the system failed.

It means you changed.

That is what living things do.

The minimum version

Your first workspace can be painfully simple:

  1. Create one page called Personal Context.
  2. Add five headings: About Me, Current Season, Important People, Projects and Goals, Preferences and Boundaries.
  3. Write three to five useful points under each heading.
  4. Give that page to your AI when relevant.
  5. Review it once a week.

That is enough to begin.

You can separate the sections into individual pages later.

You can add connectors later.

You can add automation, archives, databases and glowing dashboards later.

First, give the AI somewhere reliable to learn you from.

One place.

Clear context.

Kept alive.

That is where continuity begins.

<aside>
🕯️

Next in the series: How to connect your workspace to an AI—what connectors actually do, what they do not do, and how to begin without accidentally giving a chatbot the keys to the entire crypt.

</aside>