Quick answer (TL;DR): Most people get little value from AI because every conversation starts with no memory of the last one — so they spend the first ten minutes re-explaining who they are and what they’re working on. The fix is to give your AI assistant a persistent, external memory: a small set of files you own that the assistant reads at the start of every session and updates at the end. This single change turns AI from a disposable tool into a colleague that remembers your projects, your people, your preferences, and your decisions. The conversation is disposable; the memory is what compounds.
If you’ve ever felt that AI is “impressive but kind of useless for real work,” this is almost certainly why — and it’s fixable in an afternoon.
The vending machine pattern (and why it caps your results)
Here is how the overwhelming majority of people use tools like ChatGPT, Claude, Gemini, or Copilot:
- Open a new chat.
- Ask a question.
- Get an answer.
- Close the tab.
- Tomorrow, open another blank chat that remembers nothing — and ask something that assumes context the machine no longer has.
This is the vending machine pattern: insert query, receive snack. It works fine for trivia, quick drafts, and one-off questions. It is also the reason most professionals plateau at “AI is a faster search engine” and never get to “AI is the most useful colleague I have.”
The bottleneck isn’t the model’s intelligence. Modern assistants are extraordinarily capable. The bottleneck is that every session starts from zero context, which means you are the model’s short-term memory, manually reloaded every single morning. You re-explain your role. You re-paste the project background. You remind it who the stakeholders are and what you decided last week. By the time the assistant is actually useful, you’ve burned the energy you were trying to save.
Why “AI has memory now” doesn’t solve it
Many tools advertise built-in memory features, and they help a little. But built-in memory is usually shallow, opaque (you can’t easily see or edit what it “remembers”), and not portable across tools or accounts. It’s a nice convenience layer. It is not a system you control. For real work, you want memory that is external, human-readable, and owned by you — something you can inspect, correct, restructure, and carry with you.
The shift: treat AI like a colleague, not a vending machine
Think about what makes a good human colleague valuable. It isn’t raw IQ. It’s that they don’t need re-onboarding every morning. They remember the project, the politics, the decisions, your quirks, and the thing you asked them never to do again. They pick up where you left off.
The entire difference between a tool and a colleague is persistent, structured memory — plus the discipline to maintain it.
So the shift is simple to state and powerful in practice:
Stop treating each conversation as the precious thing. The conversation is cheap and disposable. The memory is what’s precious. Sessions are water; your second brain is the riverbed that shapes where the water goes.
Once you internalize that, the workflow inverts. Instead of cramming context into each chat, you build context once, store it in files, and have the assistant rehydrate from those files at the start of every session.
What “persistent memory” actually looks like
This is not abstract. It’s a folder. Specifically, a small set of living documents stored somewhere both you and your assistant can reach — a cloud drive folder is ideal. At minimum:
- A standing handoff document — the single source of truth. The first thing read at the start of any session. It holds current state, what shipped recently, what’s pending, and the protocol for how you work together.
- A profile / preferences file — who you are, how you like things done, your communication style, and your hard rules (“never do X,” “always format Y this way”).
- A people file — the humans in your orbit: names, roles, who reviews what, who needs careful handling.
- A glossary — your domain’s jargon, acronyms, tool IDs, and links. Everything a new hire would have to ask about twice.
- A private notes file — honest observations meant only for you.
Then a folder per recurring unit of work (clients, projects, products, cases — whatever your job is made of), with one file per item.
That’s it. That’s the second brain. It’s not a product you buy; it’s a structure you create and the assistant maintains alongside you.
The two rituals that make it work
A second brain is inert without a loop around it. Two rituals close that loop:
1. The opening ritual. At the start of every session, the assistant reads the standing handoff and the relevant context files before doing anything else. This is the move that collapses the ten-minute re-explanation into nothing. The assistant rebuilds context from the files, not from you.
2. The closing ritual. Before the session ends, the assistant writes back what changed — what shipped, what’s now pending, what was decided — and timestamps it. If you skip this, the next session starts from a stale picture and you’re back to manual reloading.
Read the brain. Feed the brain. Everything good comes from keeping that loop intact.
A concrete before-and-after
Before (vending machine):
“Hi, I’m a marketing ops person at a SaaS company. I run several campaigns. We use these tools. My manager is X. Last week we decided Y. Now, about the deck I’m working on…” (Repeat every session, forever.)
After (colleague):
“Pick up where we left off on the Q3 campaign.” Assistant reads the brain, already knows the campaign is mid-flight, who’s waiting on what, the manager’s preferences, and last week’s decision — and starts working.
The second version isn’t science fiction. It’s just a folder and two habits.
Common objections (answered)
“Isn’t maintaining files just more work?” It’s front-loaded work that pays compounding dividends. You write the profile and glossary once. After that, the assistant maintains most of it for you as part of the closing ritual. The time you save by never re-explaining context dwarfs the maintenance cost within days.
“Can’t I just keep one really long chat?” You can, and it’s better than nothing, but it breaks down fast: long chats get slow, lose the thread, can’t be searched well, and are invisible to any new conversation (including ones on other devices). Files are durable, structured, and readable by every session.
“Which AI tool do I need for this?” Any capable assistant that can read and write files, or that you can paste context into, works. The method is tool-agnostic. What matters is the structure and the rituals, not the brand.
The payoff (and a note on why this matters more than it seems)
When you make this shift, three things happen fast: your sessions get dramatically more useful, you stop dreading the setup tax, and — this is the part people underestimate — you build something portable. The way you’ve structured your thinking and your work becomes an asset that isn’t tied to any one employer, tool, or project. It’s yours. It comes with you wherever you go next.
That last point turns out to matter a great deal. But that’s a topic for another post.
Frequently asked questions
What does it mean to use AI as a “second brain”? It means giving your AI assistant a persistent external memory — a set of files you control that the assistant reads at the start of each session and updates at the end — so it remembers your projects, people, preferences, and decisions instead of starting from zero every time.
Why does AI feel useless for real work? Because each conversation starts with no memory of the last, forcing you to re-explain context every session. The capability is there; the missing piece is persistent, structured memory plus the rituals to maintain it.
Do I need a special tool to build an AI second brain? No. Any capable AI assistant plus a cloud folder of plain documents works. The value is in the file structure and the opening/closing rituals, not in a specific product.
How long does it take to set up? The initial files take an afternoon. After that, the assistant maintains most of the memory for you as part of each session’s closing ritual.
This is Part 1 of a series on building a working method with AI. Next: why every conversation starting from zero is a memory problem — and the exact file architecture that solves it.