You have probably noticed: an AI assistant is brilliant from the very first sentence, then it forgets everything. Every new request starts with the same ritual. Who is this client, what is our offer, what did we decide last week, how do we usually do it. In an SME, this is often where a good share of the time AI was supposed to save gets lost.
The answer has a name: the company second brain. A shared memory that your teams and your AI agents read before acting.
What a company second brain is
It is a living knowledge base where the company stores what it knows, as it learns it:
- decisions: who decided what, when, and why;
- procedures: how recurring operations are carried out (a quote, a delivery, a new client coming on board);
- clients: context, history, contacts, commitments made;
- meeting notes and reference documents.
The difference from a shared drive fits in one sentence: in a drive, you look for a document; in a second brain, you find an answer. Entries are structured, dated and given a status, in a format AI can query.
The four layers of a second brain that lasts
- Capture. What is said in meetings is recorded and summarised automatically; important emails and documents are added to the base. Nobody copies anything out by hand.
- Organise. Each piece of information becomes an entry in the right collection, with a title, a date, an owner and a status (current, to review, archived).
- Give access. Teams consult the base through a simple interface; AI agents read it through a standard connection, MCP, with defined rights: who reads what, who can write.
- Maintain. Outdated knowledge is worse than missing knowledge. Each entry has a review date, and a regular review archives what is no longer useful. This is the layer people forget, and the one that holds everything together.
What we learned building it at Codito
Before offering it to a client, we built it for ourselves. Our AI agents, which prepare quotes, track the sales pipeline or write content, share a single base: knowledge, decisions, work logs, documents. Each one reads it before acting, so nobody has to explain the context again. Three lessons came out of it:
- One single source of truth per subject. When the same information lives in two places, AI always ends up reading the wrong one. We brought what was scattered across several tools together in a single base, and closed the old access points.
- Status matters as much as content. An entry marked “to review” must not be read as a rule in force. AI needs to know what is authoritative.
- Separate what gets shared. What is internal and what can be shown to a contractor live in separate spaces, with separate rights.
Where to start in an SME
- Create three collections: decisions, procedures, clients.
- Switch on automatic meeting note-taking and add every set of minutes to the base.
- Give the company's AI assistant read access to the base, then measure how many times a day nobody has to explain the context again.
- Set a twenty-minute monthly review: anything outdated gets archived.
The second brain is also the foundation of good AI governance: an AI that reads the right rules at the right time is an AI that makes fewer mistakes. For procedures in particular, see how to automate procedure tracking with MCP.
Codito Ergo Sum builds this second brain with the business leaders it supports, as part of the AI Partnership, from €499 excl. VAT per month. The simplest way to talk it through: 30 minutes with the founder.
