.jpeg)
SMA Crown Confidential Notebook – Note: 'Autonomy Score'
Occasion: The first external instrument aimed at the sector’s own tools appeared — Ezra Group’s “AI Agents for Advisors” directory, cataloguing 51 advisor-specific agents and announcing an eventual 0–5 “Autonomy Score” for how much human input each agent needs (CONFIRMED as to the directory’s launch; the Autonomy Score described as forthcoming, not yet published — who audits the scorer, and how the score is verified against real approval workflows, unstated). The directory is written as launched and the score as forthcoming; the position turns on what the scale measures, not on the score existing yet.
Position occasioned: beginning with the person, not the assets — re-lit through a measurement occasion and turned on the axis of the measure itself: a scale that ranks instruments by their reach over the work has no coordinate for whom the instrument was formed from.
A directory of advisor AI announced this week that it will begin scoring the agents it catalogues — a scale from zero to five, measuring how much human input each agent still needs. The score is described as forthcoming rather than published; for now it is the intention that matters. Someone has set out to rank these instruments by how far each can carry the work on its own, and to publish the ranking. It is the first serious attempt we have seen to hold the sector’s tools to a common measure, and we think it is a genuinely useful idea.
What we want to sit with is what the measure measures. A zero-to-five score of how much human input an agent needs is, precisely, a reading of the instrument’s relationship to the work — how much of the doing it takes off a person’s hands, how far toward acting alone it sits. Every gradation on the scale describes the same thing at a different degree: the tool’s reach over the task, its independence in performing it. That is worth knowing. A firm weighing one of these agents against another should want exactly this.
But the scale has one axis, and it is not the only one that matters. There is nothing on it for the instrument’s relationship to the person whose financial life it touches — whom it was formed from, whom it is turned toward, whether it began at the individual or at the account. Those questions do not live anywhere on a measure of autonomy, because the measure, like the sector that produced it, begins at the assets and the work to be done on them. It asks how much a tool can do. It does not ask where the tool began.
This is the distinction we keep, and a yardstick built to rank the sector’s instruments throws it into unusual relief. An intelligence can earn the highest mark for how much it does to the holdings — reconcile, draft, analyse, act — and never once have begun with the person who holds them. It is not that such a tool scores badly on our axis. It is that our axis is not on its scale at all. The two do not sit at different points on one line; they run in different directions.
We begin at the person. What we build is formed from the individual rather than the market, and it stays turned toward them — not the breadth of the work to be done on the assets, but the depth of what it means to hold them. It would score poorly, we suspect, on a measure of how much it does on its own, because doing on its own is not what it is for. That is not a shortfall on the scale. It is a sign that the scale was built to measure something else. We hold the person, not the portfolio — and no ranking of how much an instrument acts will ever reach the question of whom it was made from.
Founder & CEO of SMA Crown Confidential
Digital Confidantes: Bespoke AI Intelligence for Private Decision-Makers
Comments