Second brain for companies: from one person's system to a company brain

A personal system assumes you saw the thing and saved it. A company runs on things that arrived in somebody else's thread, which is where the method quietly stops scaling.

If you already keep a personal system and you want the version that holds the company rather than you, the useful question is not which app. It is which of the assumptions underneath a personal system still hold when the subject is a business, because two of the three do not, and the one that breaks hardest is the one nobody notices until month four.

The idea itself travels well. Keep the record outside your head, keep it in files you own, make it answerable rather than merely stored. All three of those are exactly as true for a company as for a person. What does not travel is the operating model, which is that a single motivated human notices what matters and puts it somewhere on purpose.

What breaks on the way up

That last row is the one that reframes the whole exercise. A personal system is mostly a library: things worth keeping, findable later. A company brain is mostly a ledger: who owes what to whom, by when, and what changed since the last time anyone looked. Libraries tolerate being out of date. Ledgers do not. The full version of that distinction, and why it decides what you should buy rather than build, is AI company brain vs a wiki.

Second brain for business: what has to change

Three changes turn the idea into something that survives a real operating week.

• Capture stops being a decision. If filing is a choice somebody makes, it competes with the work and loses. The intake has to run off the mail, the meetings and the documents that are already moving, without anyone electing to save them.
• One canonical home per fact. Not one folder per person's preference. The date, the number and the status live in exactly one file, and everything else reads from it, so two versions cannot quietly disagree.
• Maintenance moves to a schedule rather than a habit. A pass runs on its own, files what arrived, closes what was resolved, opens what was created, and puts the next step in front of you. Hourly at best, on a machine that is awake, which is a modest claim and the only honest one.

Those three are the working parts, and the assembly instructions are their own page: how to build a company brain. The wider discipline they belong to, including what changes when the writing is done by an agent rather than a person, is AI knowledge management for a drug development team.

The honest shape of the installable version

Worth saying plainly, because the phrase invites a bigger picture than the product supports. What gets installed is one leader's seat: one owner, one mail identity, one set of files. It holds the company's threads, promises and decisions as they pass through that person, and it is genuinely a company brain in the sense that its subject is the business rather than the individual. It is not a multi-user knowledge platform, nobody else logs into it, and it never assigns a task to a colleague. A commitment somebody else made is captured as the owner's own follow-up, which is to check whether they did it.

For a small company that limit is smaller than it sounds, because the person asking for this is usually the integration layer already: the one who knows what the partner agreed to, what the vendor promised and what the board was told. That framing has its own page in an AI chief of staff for a biotech founder. For a larger organization it is a real constraint, and it is better known before an install than after one. The record also decays wherever its source material never arrives, and anything regulated stays in the systems built for it.

Where a drug development company feels it first

In a drug development company the ledger fills up from outside the building, and a co-development partnership is the version of it that bites hardest. The collaboration agreement carries obligations with notice periods attached. The joint steering committee decides things out loud and minutes them weeks later, if at all: a bridging study is approved one year, then deferred the next because the partner's read on the regulatory route changed, and a year after that nobody in the room can reconstruct which of those is current. A cost-share invoice arrives that does not match the split anyone remembers agreeing. A milestone payment triggers on an event you have to prove happened, using evidence sitting in three separate threads.

A personal method cannot hold that, and not because the method is weak. It is because the raw material never passes through the one person doing the capturing at the moment it matters. What is needed is something reading the same inbox on a schedule and treating an obligation buried in a paragraph as an item with a date, not as correspondence. That is the promise underneath all of this, and it has its own page in AI institutional memory.