EveryDay Carry: Why You Need a Backpack For AI

The goal isn't to download somebody else's ‘perfect second brain’ template. The goal is to establish enough durable identity and structure that a new intelligence can enter your environment and understand how to operate. Capture isn't…

Nicholas McGill

iamheroik

October 4, 2026

I have spent a few years teaching machines who I am. One generation of AI after another. 

Teaching the next machine. Then teaching the first machine something it forgot. Then moving information between them. Then figuring out which one had the current version. Then wondering why something from one project was suddenly showing up in another.

AI memory has gotten dramatically better. The underlying problem hasn't disappeared.

If you use ChatGPT, Claude, Grok, Gemini, or whatever comes next as the primary home for your operational memory, you are building your memory inside somebody else's machine.

That creates a problem every time you want to leave.

So I started thinking about AI context differently.

Your AI needs a backpack. A space you control that contains enough of your operational context that you can pick it up, walk into another AI environment, and say:

Here's my stuff. Get oriented.

We've Been Fighting This Problem for Years

People who have used AI for a while may remember when there effectively was no memory.

Every new conversation started cold. If you wanted consistency, you had to provide the context yourself. So we built giant prompts. Then prompts about the prompts.

We explained who we were, what we were doing, how we wanted things written, what had already happened, what mattered, and which details absolutely could not be omitted.

We learned quickly that merely providing information wasn't always enough. The AI might technically have the context and still fail to appreciate its significance.

So you added context about the context.

Then context windows improved.

  • Projects arrived.
  • Memory arrived.
  • File uploads improved.
  • Connectors arrived.
  • Custom instructions improved.

The models got better at deciding what mattered.

And all of that is useful.

I use it. I also don't completely trust it. Nor do I think that trust is warranted yet.

Models still forget things. They prioritize information differently than you would. Context can get mangled. Old information can linger. Information from one scenario can occasionally contaminate another.

More importantly, different AI environments still know different versions of you.

You teach one platform for months, then another company releases something incredible.

Now what? Start over?

Stop Making the Model the Canonical Version of You

This is the mistake.

Your AI provider should not own the canonical operational version of you.

Let the model remember things, learn preferences and maintain working memory. Let it use its project features, connectors; all the good stuff.

But your durable operational memory should exist somewhere outside the model.

Somewhere you can inspect, correct and organize. 

Somewhere another system can potentially access.

Somewhere that doesn't disappear from your operating model simply because one AI provider changes a feature, changes a policy, suffers an outage, raises its price, hits a usage limit, or gets leapfrogged by somebody better.

That is the AI backpack, a critical part of your Digital EDC (everyday carry for the uninitiated). 

Why a Backpack?

I like everyday carry because the philosophy is practical.

You carry the things that help you operate. It is the portable system that allows useful things to move with you. That's what I want from my AI context.

Claude suddenly fixes whatever temporary traumatic brain injury is making it spit gibberish at me?

Great.

Here's the backpack.

ChatGPT becomes dramatically more cost-effective for a certain kind of work?

Great.

Here's the backpack.

Grok gets some capability the others don't have?

Cool.

Here's the relevant part of the backpack.

A new model appears next year and makes today's leaderboard look prehistoric?

Even better.

Here's my stuff.

Get oriented.

The point is continuity across models.

AI Portability Matters More Than AI Loyalty

People already talk about AI platforms like sports teams.

"I'm a Claude guy."

"I use ChatGPT."

"I switched to Gemini."

Why?

These companies are competing.

Models leapfrog each other.

  • Prices change.
  • Limits change.
  • Features change.
  • Policies change.
  • Reliability changes.

The best system for one kind of work may not be the best system for another.

And none of these companies has a special obligation to preserve your individual operating advantage.

So I don't want my architecture built around loyalty to any one of them.

I want optionality.

ChatGPT goes down, move to Claude.

Claude hits a limit, move to ChatGPT.

Grok gets something useful, use Grok.

A specialized model is better for a particular job, use that.

That isn't merely redundancy. It is robustness.

Eventually, done well, it starts moving toward antifragility because competition between AI systems becomes useful to you rather than disruptive.

When somebody releases a better engine, you don't have to rebuild the car.

You can take advantage of it.

Notion Is a Pretty Good Backpack Right Now

I increasingly use Notion as part of this architecture.

I am not claiming Notion is the theoretically perfect technology for storing and managing human knowledge.

I've never seen a perfect tool.

And I wouldn't build the larger system so that Notion itself becomes another catastrophic single point of failure.

I use local and synced files. Google Docs. Markdown. Other repositories and tools where they make sense. Important systems can be mirrored. An index can point agents toward information that lives somewhere else.

But Notion occupies an unusually useful position right now.

  • It is human-readable.
  • It is structured.
  • It is relational.
  • It is editable.
  • It is searchable.

Normal people can operate it.

And increasingly sophisticated AI agents can operate against the same environment.

That combination matters.

An AI backpack doesn't need to be the theoretically perfect storage technology.

It needs to reliably carry your shit.

Right now, Notion is pretty damn good at that.

Your Backpack Needs Pockets

This doesn't mean creating one enormous page called:

EVERYTHING ABOUT ME.txt

A backpack without compartments becomes a sack.

Different information has different meaning.

A simple personal AI environment might eventually distinguish between things like:

Memory.
"Nicholas goes by Nicholas, not Nick."

Knowledge.
"This is the current positioning doctrine for the company."

Goals.
"These are the outcomes I'm currently pursuing."

Projects and tasks.
"This is what we're doing and what needs to happen next."

Artifacts.
"This is the final approved version."

State.
"The director hasn't reviewed this yet."

Decisions.
"We considered these options and chose this one for these reasons."

Lore and experience.
"This is how I experienced what happened and what I took from it."

That last one matters.

Lore isn't necessarily objective fact.

My interpretation of an experience might be incomplete. I may later decide I was wrong.

It is still useful.

Human judgment isn't built exclusively from clean, objectively verified facts. We accumulate experiences, interpretations, intuitions, scars, stories, exceptions, and lessons.

Preserve those without pretending they are something they aren't.

Then there are instructions, governance, review mechanisms, and the rules that tell the system how all of this should be interpreted.

Which leads to something even more important.

Give the Backpack a Spine

A portable information dump is not enough.

Your system needs a durable core that explains who you are and how the rest of the environment should be understood.

I think of that as the Spine.

For better or worse, something has to be authoritative.

  • What do you value?
  • What are your goals?
  • What rules govern the system?
  • What matters to you?
  • How should conflicting information be handled?
  • What deserves escalation to you?
  • Where does deeper information live?
  • What should an AI assume, and what should it ask?

Then you can have sections beneath that Spine according to your needs.

Mine will not look like yours.

It shouldn't.

A CEO, pastor, investor, parent, salesperson, researcher, student, writer, and engineer live different lives and solve different problems.

The goal isn't to download somebody else's "perfect second brain" template.

The goal is to establish enough durable identity and structure that a new intelligence can enter your environment and understand how to operate.

Capture Isn't Enough

There is another requirement I consider fundamental.

The backpack needs a scoring layer.

Capture is cheap.

I want to preserve things that matter or reasonably could matter.

But capturing everything without evaluating it creates a digital junk drawer.

So AI should perform an initial relevance assessment against rules you establish.

That might include your:

  • goals,
  • values,
  • Needs, Deals and Wants,
  • current projects,
  • relationships,
  • priorities,
  • governing rules,

and whatever else actually matters to your life and work.

The score isn't truth. It is analysis.

And importantly, the score should be capable of changing.

An old conversation might have been irrelevant six months ago.

Then you decide you need to hire a particular kind of executive.

Suddenly that old conversation with someone who knows exactly that kind of person matters.

The record didn't change.

Your relationship to it changed.

So relevance should be dynamic.

  • Time matters.
  • Current goals matter.
  • Relationships matter.
  • Subsequent information matters.
  • Your changing priorities matter.

That is why "AI memory" is too small a concept.

You don't merely need something capable of remembering.

You need something capable of helping determine what deserves attention now.

Let the Machine Score It. Then Keep Your Vote.

I also don't want AI relevance scoring to become another invisible authority.

Let the machine score.

Then preserve the human's ability to score too.

Sometimes the AI says:

"This looks like a 92."

And I say:

"That's a 40."

Good.

Now we have something interesting to discuss.

  • Why did the machine think it mattered?
  • Did it catch something I missed?
  • Did I give it bad instructions?
  • Have my goals changed without my system being updated?
  • Am I emotionally overvaluing or undervaluing something?
  • Is the system wrong?
  • Am I?

That disagreement can improve both the human and the system.

This is not about outsourcing judgment.

It is about creating better material for judgment.

Preserve More Than Facts

This also means I don't want my system immediately compressing my entire history into a tidy list of "things Nicholas believes."

I want records.

I want conversations.

I want artifacts.

I want the stream of thought.

I want the ability to revisit what I actually said rather than merely trusting an AI-generated summary of what somebody decided I meant.

That doesn't mean every raw conversation deserves expensive analysis forever.

Store broadly. Process proportionally.

Important things earn deeper analysis.

Less relevant material can remain available without constantly demanding attention.

Old information can decay in relevance without being destroyed.

A tourniquet doesn't get thrown out of your EDC kit because you haven't needed it lately.

It simply doesn't need to be in your hand while you're looking for your keys.

Don't Confuse Centralization With Portability

This is another important distinction. The ultimate goal isn't to cram your entire digital existence into Notion. That would merely create another dependency.

The system can be distributed.

  • Some information may live in Notion.
  • Some may live in Google Drive.
  • Some may live locally.
  • Some may live as Markdown.

Some may live in specialized repositories with different access requirements.

What matters is that the system has an index and roadmap.

  • What exists?
  • Where is it?
  • What does it represent?
  • Why does it matter?
  • How is it used?
  • How authoritative is it?
  • How does an agent get access when it needs to go deeper?

That is different from centralizing everything.

The backpack can contain the map without containing the entire territory.

Keep the Human Able to Open the Bag

One reason this matters to me is inspectability.

If an AI vendor remembers something about me internally, I may not know exactly what representation it is carrying.

Maybe I can surface some of it.

Maybe I can't.

Maybe it has compressed ten interactions into an inference I would disagree with.

Maybe something is stale.

Maybe something important never made the cut.

Out of sight is out of mind.

And if I can't readily surface something, inspecting and correcting it becomes harder.

In an environment I control, I can open the bag.

I can inspect the contents.

I can revise them.

I can see what is fact, what is interpretation, what is old, what is current, what is important, and what the system thinks is important.

That visibility matters because this information increasingly influences what AI tells me.

Don't Start Over Every Time the Intelligence Changes

This is the practical test.

If my AI environment is working properly, I should be able to encounter a brand-new AI tomorrow and not start from zero.

I shouldn't need six months of conversations before it understands how I work.

I shouldn't have to manually reconstruct years of useful context from another vendor's chat history.

I should be able to point it toward a governed environment and say:

  • Here is who I am.
  • Here is what matters.
  • Here is what I'm doing.
  • Here is what I've learned.
  • Here are the current projects.
  • Here are the rules.
  • Here is where everything else lives.

Read the map. Get oriented. Let's work.

That is the AI EDC backpack.

And once you have one, something else starts becoming possible.

The backpack doesn't merely let you carry your stuff between tools.

Over time, those memories, experiences, decisions, rules, failures, artifacts, relationships, and accumulated context can become part of something much larger.

Something increasingly individualized. Something that doesn't merely help an AI understand you. Something that expands what you are capable of doing.

But that's the next piece.

For now, start with the backpack. Because your AI is replaceable. Your accumulated context shouldn't be.

 

Share this

Disclaimer

Thinking In Motion

Field Notes represent the authors current stand -thinking while in motion. This is not permanent, polished, Heroik doctrine. Counter claims, better evidence, experience, and argument may change where we land. So, show grace, engage in good faith in the comments, if we're getting it wrong or missing something - say something.

In This Article

What this article helps the reader see

You May Also Like...

Capability Does Not Create Sovereignty

Capability Does Not Create Sovereignty

AI can outperform humans at analysis, prediction, coordination, and execution without earning the right to define what matters. Capability can justify delegation and authority. It does not create sovereignty, erase accountability, or establish human purpose itself.

read more

Operating Values

Now You Know Who We Are.
Let's See Where You Stand.

You have seen how Heroik thinks, what we have built, who we have served, and what we stand on. The question is whether this is the right fit for where your organization stands right now. That is exactly what the Cup of Clarity is designed to answer. No theater. No long sales process. No BS. One conversation that shows you what is real, what is missing, and what the next move needs to be.

The Cup of Clarity is a complimentary, structured diagnostic conversation led personally by Nicholas McGill. You walk in with pressure, questions, or the feeling that something structural needs to move. You leave with language for what you are experiencing, a clearer picture of where you stand, and an honest assessment of what the right next step looks like.

Make your strategy a priority. We certainly will.
From Thought to Profit.™

0 Comments

Submit a Comment