Second Brains for Teams: Organisational AI Without the Complexity
In my work advising technology companies and leading engineering teams, I repeatedly diagnose the exact same organizational disease: tribal knowledge fragmentation. The critical context that runs the business (why pricing is structured a certain way, what key customers complained about, how deployments operate) lives in people's heads or lost chat threads.
When I began building sovereign AI workflows and organizational knowledge systems, I realized that AI makes this disease suddenly visible. Bring in a capable model, ask it about your internal business logic, and it knows nothing, because your team never established a structured memory a machine or a new hire could query.
The cure is what I call an organisational second brain: a shared, living memory for the team. And the first thing to understand is how little overhead it takes to start.
A shared folder and an AI agent. That's it.
Here is a complete, working organisational second brain:
- One shared Google Drive folder, with a sane structure.
- NotebookLM pointed at it as an AI agent.
That is the whole system. Meeting notes, decisions, project briefs, customer feedback, and SOPs go into the folder. Anyone on the team, from the newest hire to the CEO, asks the agent questions in plain language and gets answers grounded in the company's actual documents, with sources cited.
No platform procurement. No vector database project. No six-month knowledge-management initiative with a steering committee. A team of five can set this up in an afternoon, and it will outperform the seven-figure intranet at most enterprises, simply because people actually use it.
The three rules that make it work
The tooling is straightforward. The discipline is what matters. Three rules:
1. Properly maintained
A second brain is a garden, not a warehouse. Someone must own it: pruning duplicates, removing obsolete files, and organizing strays. One hour a week is genuinely enough. An unmaintained knowledge base is worse than none at all, because it answers confidently from stale information.
2. Semi-structured
Notice the emphasis on semi. Rigid taxonomies fail because nobody files correctly under pressure, so people stop filing altogether. Zero structure also fails because an unstructured dump folder is write-only memory. The practical path wins: a dozen top-level folders that mirror how the business actually thinks (customers, products, operations, decisions), consistent document titles with dates, and a one-paragraph summary at the top of anything long. Structured enough for retrieval; loose enough that people actually contribute.
3. Keep the context recent
This is the rule most teams miss. An AI agent reading your second brain trusts what it finds. A 2023 pricing sheet sitting next to the current one is not harmless clutter, it is misinformation with a citation. Archive aggressively. Date everything. Make "is this still true?" part of the weekly maintenance hour. A smaller brain full of current truth beats a large one full of outdated history.
Why this matters more every month
I wrote before about AI minions as the new operating system for work. Here is the dependency nobody talks about: agents are only as good as the context you feed them. Two companies buy the same AI tools; one has a maintained second brain and gets a workforce multiplier, while the other has scattered chat logs and gets a chatbot that hallucinates about its own business.
The organisational second brain is the substrate everything else lands on. Build the habit now, while it is just a folder and an agent, and every AI capability you adopt later inherits it. Teams that treat shared memory as infrastructure (boring, unglamorous, and decisive) will quietly outrun everyone else.
A new series begins here
Start simple. Stay maintained. Keep it recent. Complicate only when the simple version is bursting at the seams, and you will be surprised how long that takes.
This post opens a series on this blog: Second Brains. We have lived this idea for years. It is the name of our AIoT architecture layer, where buildings and factories get the same treatment: a local, maintained, current memory of operations. Next up: how to structure the folder, what belongs in it, and how to grow from NotebookLM to a self-hosted brain you fully control.