The Links Your Tools Forgot

You know the scenario. You're trying to reconstruct why you made a specific decision six months ago. The email that prompted it is in Gmail. The meeting notes are somewhere in Notion. The task you created afterward is in Asana. The resulting conversation is in Slack.

You find all four. And you still don't understand what happened.

What you're missing isn't any of those pieces. It's the relationship between them. That link existed only in your head, and it's gone.

McKinsey measured this problem as 1.8 hours per day spent searching for information — a finding from a 2012 study that's been cited so often it's now wallpaper. The real number is probably worse; researchers tracking knowledge workers during COVID found some spending closer to a day and a half per week on information retrieval. Every citation of this stat treats it as evidence that we need better search. I don't think search is the problem.

Search is reactive by design. You type a query because you already know roughly what you're looking for. But the 1.8 hours isn't really about failing to find things you know exist. It's about trying to reconstruct relationships you once understood. The email is findable. The decision record isn't anywhere, because that's not where you wrote it. And what you cannot search for is the link between them.

Nodes and edges

In graph terms, your knowledge is the nodes. The relationships between pieces of knowledge are the edges. Your tools are very good at storing nodes. Nobody thought to build the edges.

Your contacts app has a record for the investor. Your email has her messages. Your notes have what you learned about her. None of these systems know about each other. The relationship, the one that says this person connected to this conversation connected to this outcome, lived only in your head. When it faded, it became irretrievable.

This is obvious once you name it, and somehow nobody names it. The productivity advice you'll find is almost uniformly about capture: take better notes, write things down sooner, build better habits. But captured nodes without edges are just a bigger haystack. You haven't made the information more findable. You've just stored more of it.

Vannevar Bush saw this in 1945. His hypothetical memex was built explicitly around what he called "associative trails" — the idea that the human mind doesn't operate by classification but by association, and that a good knowledge tool should let you tie two items together and follow that trail later. His design assumed the link was the primary unit, not the document.

The web gave us hyperlinks, which is the associative trail applied to public documents. What it didn't give us was the personal, private equivalent. No one built a personal memex. We got filing cabinets instead, just digital ones, organized by which app they happened to land in.

What AI connected to your tools actually gets

The pitch for AI agents connected to all your tools is that they can search everywhere at once. True. An agent with access to Gmail, Notion, Asana, and Slack can retrieve from all four simultaneously. It can surface the email and the meeting notes and the task and the thread in the same response.

But it cannot reconstruct the edges. It doesn't know that the email from the investor is related to the decision you made in Notion three days later unless you connected them somewhere. You didn't, because there was no good place to do that. So the agent presents you with a collection of nodes, well-organized, retrieved in seconds, and still missing the thing you needed: the path through them.

This is a structural gap, not an engineering one. No amount of better retrieval closes it, because the edges were never stored anywhere. They existed only as working memory, and working memory clears.

What it looks like when edges exist

Software teams figured this out. Architecture Decision Records, popularized by Michael Nygard in 2011, are documents that capture not just what was decided but why, what alternatives were considered, and what consequences were expected. The value isn't in the decision text. It's in the link from the decision back to its context. Future engineers don't have to reconstruct the reasoning from scattered commits and old Slack threads; it's right there, attached.

The same principle applies to personal knowledge. When a person record exists as a typed entity, other things can point at it. A meeting links to that person. A decision links to that meeting. A task links to that decision. The chain becomes traversable. Future-you can ask "what was my history with this person?" and actually get an answer, because the structure was built to hold edges, not just nodes.

None of this happens automatically. The linking moment, the act of saying "this goes with that," is deliberate work. But it's the work that makes knowledge compound rather than accumulate. An archive full of nodes just grows. A knowledge base with edges lets you navigate.

Eighty years ago, Bush described a device where you could tie two items together with a single gesture, name the trail, and hand it to someone else. Most of our tools today still make you copy and paste a URL.

---

Asgeir Albretsen is the founder of Harbor.