A second brain that knows what it is allowed to decide.
My network lived in a LinkedIn export, a phone and sixty days of email: 2,264 relationships I could not hold in my head. In eight days, with AI as the build partner, I built Alice. It remembers every conversation, notices who has gone quiet, and each morning tells me where the next hour is best spent and why. Building it was the easy part. The real design work was deciding what the AI may decide.
Four decisions behind it
A second brain fails when it depends on someone typing into it. This one works because of four decisions made on purpose.
- Capture what already happens. Alice reads mail, calendar and meeting notes. Nobody types into it, because a second brain that depends on data entry goes stale within a month.
- Write the win down as structure. The objective lives in the system as fields, so Alice can find what the network is missing. It surfaced 57 people I would not have thought to look for.
- Draw the judgment line. The model sorts information into fixed categories. Arithmetic in the database ranks. I approve every change on a card showing before, after and the evidence.
Illustration with sample data.
- Earn every control. Each check exists because of a specific failure and retires when that failure can no longer happen. Changes to the AI's own instructions are tested against a fixed case set and ship only if they perform no worse than what is live.
- Per change
- tests, schema versus every consumer, adversarial review
- Every 48 hours, unattended
- drift, access, controlled values, job health
- Monthly
- the whole system
AI classifies. Arithmetic decides. A person approves.
What transfers to a company
Most AI rollouts automate before anyone has decided who decides, and the gap between growth and infrastructure widens. The sequence that worked here is the one I would run inside a finance organization: start from the decision that is not getting made, write the objective down, decide where judgment lives, route every change through one queue, and add controls only when a failure earns them. Alice is the smallest system I have applied it to, which is why the whole thing is visible at once.
What it looks like
The dashboard is the only surface I touch. Names are blurred; everything else is the live system.
Alice has been hardened and extended every day since launch. Next, it extends from one search to every relationship across the businesses I help run. The full design, including the architecture, what I refused to build and the correction that mattered most, is a conversation I am glad to have.
If you are building a second brain for yourself or your team, let's talk →Or tell me which decision in your company is being made badly: msandfordboone@gmail.com