What your mind keeps open

A waiter at a Berlin café could recite every item from any table still running a tab. Once the bill was paid, the order vanished — forgotten cleanly and completely. The psychologist Bluma Zeigarnik noticed this in 1927 and ran experiments confirming it: incomplete tasks stayed in memory roughly twice as long as completed ones. The unfinished thing maintains a kind of cognitive grip until it's resolved.

David Allen built a whole system on this. Getting Things Done starts from the observation that your mind is bad at storage and good at processing, and that using it for storage is a waste. Every open loop — an unresolved commitment, an idea you mean to return to, something you've promised yourself you'll handle — consumes ongoing mental overhead. The prescription is to get it out of your head and into a trusted system.

The prescription works. The hard part isn't the capture. It's the "trusted" part, which does more work in that phrase than it appears to.

Why trust isn't automatic

You've written something down and still found yourself thinking about it. This is a system-trust failure from the inside. Your brain assessed the system as unreliable — specifically, as unlikely to surface this thing when you need it — and declined to release the open loop. The cognitive overhead didn't go away just because you wrote it down.

Trust in a knowledge system is not the same as trust in a person. With a person, you trust based on track record: they do what they say, over time, until they don't. With a system, the mechanism is different. What makes a system trustworthy is that when it fails, you can see how and why. Not that it never fails.

Think about the difference between a box of notes and a structured list. Both store information. The box fails silently — things get buried, become unfindable, and you have no way of knowing whether something is in there or just missing. The list fails legibly: you can scan it, notice a gap, and confirm what's present and what isn't. Its failure modes are visible. The box's are not.

This is why format matters beyond aesthetics. A knowledge system earns trust not through perfect recall but through inspectable state. When you can look at the system and understand its current condition — not just retrieve from it but assess it — your brain will more reliably release open loops to it.

A new kind of open loop

For a long time, the trust problem in personal knowledge was mostly about retrieval. Would the system surface this thing when you need it? The answer depended on search quality, on whether you'd organized things sensibly, on whether the file format had survived the years.

AI changes the trust problem. When an AI can write to your notes — add tasks, update a contact record, extract decisions from a conversation — you have a new category of question: what did it change? An AI that writes silently, without any visible record of what it touched, is a box of notes. You can retrieve from it. You can't inspect it. You don't know what got added, what got altered, what reflects you and what reflects the model's inference about you.

That's not a reason to avoid AI writes. It's a reason to care about the architecture. Patch proposals, audit logs, structured edits that show exactly what changed — these aren't security features bolted on for cautious users. They're the engineering requirement for a brain to actually offload something to a system that an AI is allowed to touch. Without them, your brain keeps the loop open. Reasonably.

What the structure gives you

The difference between a note that's a blob of text and a note that's a typed entity — a person record, a task, a preference with a source field — isn't just queryability. Structured data carries its own inspectability. You can look at a person record and see every field, notice the ones that are empty, trace who changed what and when. A prose note just sits there, looking authoritative, regardless of whether it's accurate or stale or half-understood from three years ago.

I spend a lot of time thinking about this in the context of Harbor. Not because the Zeigarnik effect is a product feature — it isn't — but because the design question underneath it is real: what does a knowledge system need before your brain is willing to hand something off to it? The answer isn't "store everything." It's "let me see what's in there and why it's there."

Zeigarnik's waiter forgot paid-off orders immediately because the cognitive system registered them as resolved. For your notes to feel resolved, the system has to be legible enough to trust — not just present, but inspectable. That's a higher bar than most tools meet. The ones that meet it tend to feel qualitatively different: less like a place where things go, more like a place where you know what's there.

Asgeir Albretsen is the founder of Harbor.