Artificial intelligence carries the volume so the people can carry the judgment. Where we trust the system — and the line it never crosses.
A fair question, asked more often lately: if the firm runs on software its own partner built, how much of my advice is coming from a machine? It deserves a direct answer, because the honest one is also the reassuring one. The machine does an enormous amount of the work and none of the deciding. Knowing exactly where that line sits is, we think, the most important thing you can understand about how we practice.
Start with the arithmetic that makes it necessary. Keeping the exam-binder standard — every figure reconciled, every decision reasoned in writing, every estate chapter audited annually — is enormous clerical labor, and it scales badly with people alone. A three-person firm cannot do it by hand for every household and still have time to think. Most firms resolve that tension by quietly doing less of the work. We resolved it by building BAFAos and handing it the volume.
Here is what the machine is trusted with. It drafts — the first cut of a folio chapter, a statement letter, a planning memo — from the live record, so we start from a complete draft rather than a blank page. It reconciles — tying every figure back to its custodial source and flagging, rather than hiding, anything that will not agree. And on consequential questions it deliberates, running the problem several independent ways and surfacing where those passes agree and, just as usefully, where they don't.
Here is the line it never crosses. No recommendation, statement, or folio reaches you on the strength of a model alone. Every output crosses a named, licensed human — one of us — who reviews it and owns it. When the system is uncertain, it is built to say so; a figure it cannot reconcile is shown as unknown, never dressed up as a confident default. The machine is the assistant. The accountability stays human, with a name attached.
The machine proposes. A person decides.
We are aware of the moment we are writing in — of regulators watching for firms that dress ordinary software in the language of intelligence, and of clients rightly wary of being handed to a chatbot. So we will make you a plain promise. We will not claim a capability BAFAos does not actually have, and we will not let it make a decision a fiduciary is supposed to make. Both partners review the work that carries our names; the software's job is to make sure neither of us is ever the bottleneck on doing that work right.
The point of the intelligence was never to replace the fiduciary. It was to let a deliberately small firm hold every household to a standard that would otherwise be affordable only to the very largest — and to keep the people free for the part that was always theirs. That is the whole trade, stated plainly. As always, this letter carries no market call; only the thinking behind the way we work.
