Chapter 2

Memory in the Age of AI

Chapter 1 ended on two observations. Memory outlives the individual only when the relationships between memories are preserved, and a new kind of reader has arrived: AI.

AI is changing how we make records in the first place.

Nobody has to write a document from beginning to end any more. One short question produces a report, summarizes a meeting, writes code, and reweaves scattered information into a new shape.

The cost of producing information has never been lower.

That creates a problem unlike the old one. The trouble is not too few records. It is far too many.

Documents, messages, code and AI-generated text pile up every day. It gets hard to tell what matters, and a few months later hard to recall why any of it was made.

There is a more important problem. Much of what AI produces is useful, and much of it is not our memory.

Ask an AI how to design software and it explains general principles. It lists the trade-offs of any technology. But some questions need a different kind of information.

“Why did we choose this approach?”

“What did we try last time, and why did it fail?”

“What did we know when we made this decision?”

“What has changed since?”

The answers do not come from general knowledge. They come from what we actually went through.

The experience a team gained solving one problem over months. The know-how one person built up over years. A decision reached after several failures, and the record of how it changed later. None of it exists anywhere on the internet in complete form.

That is our memory.

1. Memory and cognition are not the same thing

In the age of AI we can borrow powerful cognition. Ask and it helps us think, reads and summarizes long documents, compares information and reorganizes it from a new angle. It writes and reviews code.

But cognition and memory are not the same thing.

AI can help us think. That is no reason for it to hold our memory in our place.

Models keep changing. A better one than today's may appear tomorrow, and in a few years we may work with AI in ways nothing like the present.

So should memory built up over years be tied to one model?

It should not have to be.

Our memory should outlive the model.

2. Cognition can be borrowed

This is where Vinc's point of view begins.

Cognition is borrowed. Memory stays ours.

Vinc owns no AI model and does not try to train our memory into one. There is no particular model inside Vinc.

This is not a rejection of AI. It is the opposite. It is so that any AI can use our memory.

When a better model appears we should be able to switch. Using one model today and another tomorrow should not mean retraining or migrating the knowledge we have built each time.

So Vinc does not own cognition. It borrows it from an outside AI when needed, and lets that AI read our memory, connect it, question it, and look at it from a new angle.

We call this Borrowed Cognition.

3. What Vinc is

So what is Vinc?

Vinc is a knowledge graph that people and their AI assistants read and write together.

Vinc builds a knowledge graph from the material we leave behind. The source text is kept as the canonical record, and every concept and relation drawn from it can be traced back to the sentence it came from. Nothing asks you to trust an extraction you cannot check.

The graph holds concepts, the records that back them, and the decisions made on top of those records. What came from what, when it held, and what later replaced it remain in the graph as relations.

What Vinc keeps is not a pile of stored sentences but the whole life of a piece of knowledge: what it means, what supports it, which decisions rest on it, how it changed, how it was verified, and how an AI agent puts it to use. All of it sits in one graph, where a person and an AI see the same thing and can check it.

One note on words. This book says “memory” for the reader's sake. The product's screens and documents call the same things knowledge, records and decisions. We speak in human terms here; what the system handles is records that can be checked.

4. Three ways to reach the memory

At the center of Vinc there is one graph. It belongs to no particular AI and no particular application.

There are three ways to reach it.

The desktop. You sign in once, and it shows the same graph the web does. Turn on local mode and adding material, turning it into a graph, exploring and searching all happen on this machine, with no outside service, no API key and no server.

The web. Sign in and the same graph is there in a browser or on another machine.

An AI assistant. Through MCP, an assistant reads and writes the same graph. This is the path by which a conversational AI opens our memory directly.

All three stand on the same graph and the same contract. Whichever way in, the graph at the center must not change.

The ways in can change. The center stays one.

5. The more AI there is, the more memory matters

As AI advances it can look as though every problem of knowledge will be solved. But the more AI produces, the more there is to tell apart.

What is general knowledge. What is a single answer generated just now. What is a fact we actually lived through. What is a decision we made. What is a lesson from a failure. And what still needs remembering from here on.

AI will produce more and more. Which is why what we need may not be more generative power.

What we need is the ability not to lose what we have experienced.

Borrow cognition, keep memory ours. Vinc exists to draw that line.

And when we hand that memory back to an AI, what actually changes?

The next chapter begins with that question.

Continue with Chapter 3, Memory Changes the Answer.