Language, the Oldest Brain-Computer Interface: Seven Thoughts on the AI Native Family Office
Founder's Notes
Lately, as I build my company, I've been feeling more and more strongly that AI's impact on family offices won't just show up in who's better at asset allocation or who writes better research reports. It will show up at a more fundamental level: rebuilding understanding, and redefining delivery.
A few scattered thoughts — writing them down, writing them down.
Thought One
Feeding AI your past articles, diaries, meeting transcripts and talks is an early, rough kind of "brain-computer interface."
By "brain-computer interface" I don't mean the narrow sense of implanting a device in the brain, but something broader: how the outside world gets into a person's cognition, and how that cognition is in turn read and rewritten from outside.
Language is exactly this kind of ancient interface.
Its influence on people is often underestimated precisely because it's so commonplace.
Of course, language can't capture the whole of a person.
A person's experience, intuition, bodily sense, shame, fear, need for control and sense of responsibility mostly can never be fully put into words. Text has never equalled the person, and records have never equalled the full picture of consciousness.
Even so, language is still today's most widespread, lowest-cost and highest-leverage cognitive interface. Because it is constantly "reading and writing" our brains:
words of encouragement can rewrite a person's judgement of risk;
a new name can change how a group understands something;
a good question can compress vague anxiety into something that can be discussed and decided.
Neuroplasticity means the nervous system changes its structure, function and connections in response to internal and external stimuli. In other words, language is not just a tool of expression — it is also a tool for shaping cognition. People speak at about 150 words a minute; the bandwidth isn't high, but the penetration runs very deep.
A person's full consciousness can't be exhausted by language, yet language remains humanity's most widespread, lowest-cost and most powerful brain-computer interface.
Thought Two
The family office has never been an industry that simply handles wealth.
On the surface it manages assets; underneath, it is dealing with "people problems."
The hardest things in a family usually aren't in the financial statements or the legal clauses. They're in a "let's wait and see" at a meeting, in a silence, a change of subject, a complaint tossed off lightly.
What a family office really manages is, on the surface, money — and underneath, always, cognition organised by language:
how to define "safety"
how to define "fairness"
how to define "is the next generation ready"
how to define "conservative"
and even, how to define "why do we still want to hold on to this wealth"
None of these are purely financial questions.
They are first people questions, then structural questions, and only last product questions.
And because of this, the traditional family office
is, at heart, a high-priced handicraft.
Thought Three
The three great dilemmas of the family office:
-
Over-reliance on a handful of "decagon partners" who understand law, tax, investment and human nature all at once; knowledge sinks into individual heads and never into the system;
-
Sell-side commissions seep in like poison, quietly rewriting advice;
-
The family's most crucial unspoken needs are steadily worn away through round after round of meetings, note-taking, retelling and delivery.
I call this "the entropy of trust."
Advisers take commissions, institutions have KPIs, lawyers want to wall off risk, tax advisers only calculate tax, investment managers only watch returns — no one stands in the family's shoes and looks down on the whole picture. Every expert works faithfully on their own island, but put together, what the family gets is a plan that's neither fish nor fowl.
Worse: every few years a core adviser leaves, someone new arrives, and the family has to explain itself all over again. A decade's worth of accumulated understanding of the family — the founder's worries, the second generation's pressures, the wife's unspoken anxiety about risk — all reset to zero.
What the family really wanted to protect often ends up reduced to a set of clauses that are rigorous in form and very low-resolution in substance.
Thought Four
AI will rewrite this industry's most fundamental mode of production.
What large language models truly unlock is not the ability to answer questions, but, for the first time, the chance for machines to process humanity's most important unstructured data at scale — language.
AI absolutely should not be just "a smarter investment adviser."
It can continuously record, compare, abstract and review;
it can spot contradictions between earlier and later statements;
it can identify topics that keep getting sidestepped;
it can gradually distil "what's left unsaid" into a structured cognitive map of the family.
For the family office industry, this means that for the first time we have a chance to turn long-term understanding of a family from individual experience into an organisational asset and a system asset.
What we should build isn't a "family office ChatGPT," but a complete cognitive system for family succession:
it recognises the family's consciousness and expands the family's cognitive bandwidth;
it doesn't make decisions for the family, but preserves — over the long term, continuously and traceably — the family's true ranking of values, its language of risk, its relational tensions and its decision-making style.
Thought Five
Commercially, this is a market with enormous potential.
First, there's the tide of the times.
According to Deloitte, the number of single family offices worldwide grew from about 6,130 in 2019 to 8,030 in 2024, and may reach 10,720 by 2030; their AUM is projected to grow from US$3.1 trillion to US$5.4 trillion. McKinsey notes that from 2023 to 2030, about US$5.8 trillion in intergenerational wealth transfer is expected among HNW/UHNW households in Asia-Pacific, while the number of single family offices in Hong Kong and Singapore has grown to about 4,000 since 2020.
Second, the real big opportunity isn't in existing family offices, but in pushing the capability downmarket.
McKinsey's analysis shows that only about 5% of UHNW households in Asia-Pacific have a single family office, compared with over 15% in Europe and the US; in Asia-Pacific, a single family office with AUM above US$100 million typically has operating costs of 1–3%, and below that size the operating cost ratio can even reach 4–6%. At the same time, staff costs are often the largest expense, accounting for 45–65% of operating expenses. This means that once AI can replace most junior, repetitive, standardisable labour, what it changes isn't a single process but the economics of the entire industry: capabilities that could once only be supplied by extremely high-cost organisations can, for the first time, potentially be productised, platformised and made continuous.
One of McKinsey's views on the future of wealth management: as AI, data integration and platform automation advance, the integrated capabilities once available only to UHNW family offices will gradually extend to broader HNW and even affluent client segments — "when every client is a family office." Services will also move from single-point investment advice towards integrating family governance, tax and estate structuring, trust administration, insurance, philanthropy, medical concierge and lifestyle services.
Another McKinsey view is also worth chewing on: in wealth management, task-based efficiency tools can already save advisers 20–30% of their time; and as AI moves further towards intelligent orchestration and team-level complex workflow applications, what changes isn't just a few minutes saved, but the very logic of growth and delivery. What's really worth building isn't a note-taker, but a system that links understanding, execution and review into a closed loop.
Thought Six
An AI Native family office isn't about "removing the human."
Quite the opposite — it raises the bar for the "human" even higher.
AI first eats up a large amount of junior, repetitive, standardisable labour.
What gets freed up isn't the "human" as such, but people's time, judgement and empathy.
It lets people deliver, at lower marginal cost, service that is more stable, more private and more continuous than the traditional model.
But language can never exhaust a person, so I don't believe a system can "copy a person" either.
And that's precisely not bad news.
Because the most promising thing about AI has never been replacing a person's full consciousness, but becoming a high-quality, sustainable, correctable interface for cognitive interaction:
it helps the family express itself more accurately,
understand one another more clearly,
build up judgement over the longer term,
and, in decision after critical decision, ask the right questions.
A truly high-calibre system shouldn't pretend it fully understands a family: it should know that language conceals, know that it can only reach the expressible layer, know that silence is also information, and know where it must probe further, where it must not overstep, and where only a human can judge and bear the consequences.
AI handles the process,
humans handle the key points.
AI handles memory, alignment, retrieval and scenario modelling,
humans handle judgement, ethics, boundaries and taste.
To truly achieve —
High Tech, High Touch.
Not a bunch of programs serving a family,
but a group of people who can own the results, using an AI brain to deliver high-density service that was impossible before.
Thought Seven
Open up the "black box of trust."
Values are the soul of an organisation.
In large traditional institutions, people are both the process and the nerve endings; values often lose out to KPIs, and people's actions always get distorted. But in an AI Native organisation, values have, for the first time, a chance to be written into processes, permissions, blocking mechanisms and review standards — becoming something executable, auditable, even testable.
A buy-side stance no longer has to be merely a professional virtue; it can gradually become a system constraint.
For the family office industry, this will be a very deep restructuring.
In a sense, it's closer to the platform I want to build.
Our Technical Path
We're working to build two things:
Single Source of Truth
In plain terms: a tamper-proof master ledger of the family's will
Within a family, the preferences and decisions shown by different people, in different settings and with different needs, aren't just different — they can contradict each other: over dinner with the lawyer, the lawyer, for compliance's sake, notes down only: "set up a trust in the Cayman Islands." On the golf course with the private banker, the banker, wanting to sell wealth products, notes down only: "the family needs 8% annual cash-flow dividends." During a fight with his son, he shouts: "If you carry on like this, forget about ever getting on the board!" The son remembers that line.
This is what you call "multiple sources of truth." The family's true intentions are torn into fragments, scattered across three different people's heads. Once conflict breaks out, the documents and testimony these three produce all contradict one another, and the family is instantly plunged into an inheritance fight that tears it apart from within.
Arcus42's Single Source of Truth (SSOT): we use AI to take in all these fragments, reconcile and clean them, and form a single, complete, tamper-proof "family digital master ledger." This ledger contains not only trust clauses and cash-flow requirements, but also "the boundaries of people's power."
The point of a "Single Source of Truth" is this: whether 50 or 100 years have passed, however many rounds of lawyers and bankers have come and gone, when the family faces a major decision, Arcus42's system holds one single "panoramic map of true intent," undistorted by interested parties. That is irreplaceable trust.
"Super Control Tower"
In plain terms: an "air traffic control centre" hovering above the entire family
Buying services for family succession today is like building a plane with no control tower. The tax adviser says: "To avoid tax, let's move all the assets to Singapore." The lawyer says: "No — under Singapore's marriage law, your son could risk walking away with nothing." The investment manager says: "All the money is stuck in the structure — I can't invest in that fantastic unicorn I saw yesterday!"
Everyone gives the "optimal solution" from inside their own deep professional well, but added together, the whole family's operations end up in a mid-air collision.
Super Control Tower: the control tower is originally a concept from aviation and global supply chains. The airport tower doesn't fly the planes (it doesn't directly sell wealth products or draft legal documents), but it can see every plane's flight path and is responsible for coordinating and dispatching them to prevent collisions.
Answers will eventually go out of date.
A system that can remember, reflect, evolve — and knows its boundaries — will become the most important new infrastructure for family succession in the age of AI.
I'll put this out there for now,
and I'd love to hear from anyone who shares this view or is interested — come and talk it through with me!