Why most team wikis die
Almost every company has tried it: a Notion workspace or a shared Drive folder that was going to hold "everything we know". A few months later, half the pages are out of date, nobody trusts the search and people go back to asking on Slack.
The reasons are always the same:
- Capture depends on willpower. Writing things down is extra work on top of the real work, so it stops when the team gets busy.
- The structure is too big. Forty nested pages designed on day one, and nobody knows where anything goes.
- Nobody owns it. When everyone is responsible for the wiki, nobody is.
- Finding is harder than asking. If a question takes three minutes to search and thirty seconds to ask a colleague, people ask.
A second brain fixes these four problems by design, not by asking people to be more disciplined.
What a team second brain is (and isn't)
A team second brain is a shared memory: what you decided and why, what was said in meetings, how things are done, and what you know about your customers. It is not a dump of every file you have ever created. The test is simple: can a new hire answer "why do we do it this way?" without interrupting someone?
The four spaces you actually need
- Decisions log. One line per decision: date, decision, why, owner, link to the source. This is the most valuable and most neglected space.
- Meeting memory. A summary, decisions and action items for every important meeting, linked to the project or customer it concerns.
- Playbooks. How recurring work is done: onboarding a client, closing the month, publishing a post. Short, step by step, with an owner.
- Customer and market knowledge. One page per key account with context, history and open topics, plus what you learn about your market.
Four spaces is enough to start. You can always add more once these are alive.
Make capture automatic
This is where AI changes the game. Instead of asking people to write, you capture what already happens:
- Meetings: a recorder transcribes the call, an AI agent writes the summary, extracts decisions and action items, files them in the right space and updates the CRM.
- Chat: a reaction such as ✅ on a Slack message sends it to the decisions log with a link back to the thread.
- Client email: key threads are summarised onto the customer page.
Keep a human checkpoint: the owner of each space reviews what was added once a week. Five minutes of review beats an hour of writing.
Add an AI layer people can ask
The second brain becomes useful when people stop searching and start asking: "What did we decide about pricing for annual plans?" An assistant such as Claude or Notion AI, connected to your spaces, answers in plain language and links to the source page. The link matters: it keeps answers checkable and builds trust in the system.
Respect permissions. The assistant should only see what the person asking is allowed to see.
Keep it fresh
- Every space has a named owner.
- Every playbook has a "review by" date.
- Once a month, a 30-minute cleanup: archive what is stale, merge duplicates.
- An agent can flag pages nobody has opened or updated in six months.
A 30-day rollout
- Week 1: list where knowledge lives today (Drive, Notion, Slack, email, heads) and pick one home.
- Week 2: build the four spaces with simple templates. Move only what is still useful.
- Week 3: automate meeting capture and the decisions log.
- Week 4: connect the AI assistant, show the team how to ask it questions, and start measuring.
How to know it works
- Share of important meetings with notes filed automatically.
- Questions answered by the assistant without pinging a colleague.
- Time for a new hire to become autonomous.
- Number of stale pages, going down.
Common questions
Notion or Google Drive?
Use the one your team already opens every day. Notion is easier for structured spaces and databases; Google Drive works if your team lives in Docs. The automation and AI layer work with both.
Do we need a vector database or a custom AI?
Not to start. Built-in AI in Notion, or an assistant connected to your workspace, covers most teams under 50 people. Custom retrieval only makes sense with large volumes or strict data constraints.
How much time does it take to maintain?
With automatic capture, about five minutes per owner per week, plus a 30-minute cleanup each month.