Vision
Memory lives in people.
Decisions, reasoning, paths tried and abandoned — it all accumulates in employees' heads. A company's most valuable asset sits in no system at all.
And it leaves with them.
Every departing employee leaves a hole their successor spends months re-digging. Every departure is a small archive fire.
Borje today: memory lands in files.
Borje writes decisions and context into readable markdown files as you work with AI. The person leaves; the decision, the reasoning, and the context stay in git history. On your disk, in your format.
Tomorrow: memory scales to the organization.
The same model grows upward: role-based, permissioned work memory. Managers query the current state without waiting for a brief; everyone sees within their permission boundary. Not surveillance — decisions, tasks, blockers. Personal memory stays personal.
The path
- Individual beta
- Team memory
- Organizational layer
What is organizational memory?
Organizational memory is the shared record of a company's decisions, the reasoning behind them and the project context they came from, held independently of any individual employee. In most companies it sits in no system at all: it lives in people's heads, scatters across chat histories and meeting notes, and loses a piece with every departure. The cost never shows up in the code. What shows up is this: a new developer has nobody left to ask why Postgres was chosen over Redis six months ago. Nobody remembers. The decision is quietly reversed, the same path is tried a second time, the same wall is hit again. Organizational memory is exactly the record that stops that repetition. Not what was done, but why. Borje writes that record into plain markdown files inside the project rather than into a database. The decision, the reasoning and the context sit in git history. The file opens, reads, gets corrected, moves to another tool. The person leaves, the record stays.
Why does organizational memory disappear?
It disappears because it is never written down. At the moment a decision is made, the reasoning is obvious to everyone involved: you were in the room, you weighed the alternatives, you picked one. Writing it down feels redundant. Six months later that clarity is gone from every head in the room.
The second reason is that whatever does get written lands in the wrong place. The reasoning survives in a Slack message, a pull request comment, a meeting recording. Those are archives, not memory: nobody searches them, nobody reads them, and six months on nobody recalls which channel it was in.
The third is that the owner of the memory is a person. As long as a company's most valuable record lives in one employee's experience, that employee's resignation is data loss. Their replacement asks the same questions again and repeats the same experiments. The loss is not a single moment, it spreads over months and never appears on any invoice.
Why don't AI agents solve this?
They don't, because an agent's context window is not memory. The context window is the amount of text a model can see at once within a single session: a temporary workspace. When the session ends, everything in it is gone. Open the same agent the next morning and it has never heard the architecture you spent three hours explaining.
This is not a defect, it is the design. Model weights do not learn your project; on every request the model reads the text it is handed and produces an answer. What looks like learning is context that was copied into that window.
The result is a loop everyone recognises: every session opens with an introductory meeting. The agent scans the codebase, fails to find the reasoning in the code, and asks you. You explain. The session closes and the explanation is gone. Breaking that loop needs a record that lives outside the window, on disk.
Why a markdown file instead of a database?
Because the carrier of memory has to stay readable at its worst moment. A markdown file manages that: it opens in a text editor, reads by eye, gets corrected by hand and is versioned in git. When you see a wrong record you change one line and commit it. What changed is visible in the diff.
A database takes that access away by putting a layer in between. You need a query to see a record, a tool to correct it, an export to move it. Embedding vectors cannot be read at all: they are arrays of numbers, impossible to verify with your own eyes.
The real difference shows up as trust. You never have to guess what the AI knows about you, you open the file and look. If the product shuts down, the contract ends or your tooling changes, the files stay where they are and read the same way. Markdown is less a format choice than an agreement with the exit left open.
Is this a surveillance tool?
No. Organizational memory and employee monitoring are two different things, and Borje's vision is the first one. The scope is work memory: decisions, reasoning, open tasks, blockers and project context. Who worked how many hours, who asked which question, what someone discussed with an AI: all out of scope.
The separation is built in rather than promised. Personal memory stays in the person's own space and is never moved into shared memory. In the shared layer, visibility is bounded by role-based permissions: everyone sees within their own boundary, managers included.
The test is simple. If a record helps the next person taking over the same work, it is work memory. If it only describes how one individual works, it is not memory but monitoring, and it has no place in Borje's scope.
Questions about the vision
What is organizational memory?
The shared accumulation of a company's decisions, reasoning, and project context that lives independently of individual employees. In most companies it lives in people's heads — and disappears with every departure.
How does Borje approach organizational memory?
Bottom-up: first the individual developer's project memory (today's closed beta), then team memory, then an organizational layer with role-based permissions. At every stage, memory is kept in readable markdown files.
Is this an employee surveillance tool?
No. Borje's organizational vision is work memory: decisions, tasks, blockers, and project context. Personal AI conversations and personal memory are out of scope; visibility is bounded by role-based permissions.