In short: At two people, the enterprise brain is a single document and a weekly habit. Its job is to stop two honest people remembering the same decision differently, and it is worth more than any tool you could buy at this stage.
Key points
- Founders remember decisions differently because memory is reconstructive, not because of a trust problem.
- Each entry records the decision, why, what was rejected, what would prove it wrong, and any disagreement.
- Recording a disagreement calmly at the time defuses it later.
- Transcribe calls with consent and have AI draft entries to keep capture to a few minutes.
- Build no knowledge base, taxonomy, tagging scheme or integrations beyond transcription.
Two founders make dozens of significant decisions in the first year, mostly in conversation: on a walk, on a call, at 11pm in a message thread. Both remember them. They remember them differently.
This is not a trust problem. Memory is reconstructive. Each founder retains the version that fits their own view of why it was decided, and over months the versions drift. The drift surfaces at the worst moments: in a disagreement about direction, in front of an investor, when a co-founder relationship is already under strain.
The ledger
One document, append-only, that both founders can see. Each material decision gets a short entry:
- the date and the decision, in one line;
- why, in a sentence or two;
- what was rejected;
- what would prove it wrong, and when to check;
- where the founders disagreed, if they did.
That last line is the one most founders skip and most need. Recording a disagreement at the time, calmly, defuses it later. "We disagreed on the price; A wanted higher; we went with B's view and agreed to review after twenty deals" is a far better artefact than two memories of the same argument.
The habit
Thirty minutes once a week. Read last week's entries. Add any that were missed. Check any falsifier whose date has passed. That is the entire system.
AI makes it cheaper. Record and transcribe the founders' calls, with consent, and have a model draft candidate ledger entries from them. The founders correct and accept. Capture drops from a chore to a few minutes of editing.
Uses you will discover
Investor updates write themselves. The ledger is a record of what was decided and why, which is most of what a good update contains.
Due diligence becomes easier. A buyer or a lead investor asking why the company pivoted can be shown the entry, written at the time.
The first hire gets context. When the third person joins, the ledger is their best onboarding document, by far.
What not to build
No knowledge base. No taxonomy. No tagging scheme. No integrations beyond transcription. At two people, every hour spent on infrastructure is an hour not spent on customers, and the structure will be wrong by the time it matters anyway.
In practice
- Start one shared, append-only decision document today.
- Record disagreements explicitly, with who held which view.
- Hold a 30-minute weekly review, and check falsifiers on their dates.
- Use transcription and AI drafting to keep capture under five minutes a week.
What would prove this wrong. Founding teams with no written decision record experiencing no more disputes about past decisions, and no slower onboarding of early hires, than those with one.
Common questions
How should co-founders keep track of decisions?
Keep one shared, append-only document with an entry per material decision: date, decision, why, what was rejected, what would prove it wrong, and where the founders disagreed.
Why do co-founders remember decisions differently?
Memory is reconstructive. Each founder retains the version that fits their own view, and over months the versions drift.
Can AI help founders keep a decision log?
Yes. Record and transcribe calls with consent and have a model draft candidate entries, which the founders correct and accept.
Should a two-person startup set up a knowledge base?
No. At two people, every hour on infrastructure is an hour not spent on customers, and the structure will be wrong by the time it matters.