Today, with all the hype around AI, it's difficult to separate all the noise from what really matters. If you see a lot of AI-related news in your feed (like I do), it may look like AI is taking over the world and can solve all your problems and tasks with a few prompts. That is far from the truth (unless for now).
As someone who is actually building an investment platform and actively using agentic AI every day for work-related tasks, I can tell you that what sounds good and easy on paper, in practice turns out to be more complicated if you want a decent result and a system that works and not just a tick-box exercise.
This Week in 30 Seconds
- Who the new tools are for. Anthropic's adviser product launched on 14 September at around $70 to $120 per adviser per month. It is built around the work advisers do for clients.
- Why that matters if you run your own money. The model is the cheap, swappable part. What costs you is making it work with everything else you already use, and nobody sells that.
- Check what it tells you. AI can be right 95% or 99% of the time. The 1% or 5% is the problem, and without domain knowledge you might not spot it.
- What to do this week. Pick one workflow in your own money setup and write it out as instructions somebody else could follow. Give it an hour.
- On the Radar. Only one in 18 American centimillionaires runs a family office, family offices are buying AI faster than they are using it, the SEC cleared tokenised stock trading for 5 years, and two newsletters worth reading on capital policy and deal filing.
Who AI wealth tools are built for
Anthropic launched Claude for Financial Advisors on 14 September, at around $70 to $120 per adviser per month, plugged into 18 systems advice firms already pay for.
The seven workflows it ships with cover onboarding a new client, getting ready for a client meeting, writing up the notes afterwards. This has many integrations and is designed for advisers who advise clients, not founders managing their own wealth.
Not everyone in the industry is convinced. Alois Pirker of Pirker Partners expects it to hit the same wall Salesforce hit in wealth management, where a general tool needed so much fitting to how each firm worked that it never quite got there.
Why AI tools stall on data and integrations
It's like putting a Ferrari engine in a Prius. It's a powerful engine, but you won't get a Ferrari car out of it. All parts need to work together well, and you need an appropriate suspension system, gearbox, tyres, etc., for the car to perform at its best.
Another challenge is that while an AI tool can solve some tasks and parts of the workflow, you'll still likely have legacy systems and service providers—like custody, private markets platforms, corporate providers, and other parts of the wealth management ecosystem.
When implementing any tech solution, many face challenges with data quality, integrations, and different stages of the process that may not talk to each other or where bottlenecks can occur.
Mercer asked 131 asset managers what was holding them back. 69% said data. Models got cheap a while ago, which is what AI actually changes about running a family office at $5M–$100M. What still costs you time is making them work with everything else you already use.
To be fair, connecting things is exactly what Anthropic just sold, and Reed Colley of Orion says that is the point. But those connections only reach the systems that product already supports, and if you are running $5M–$100M without the institution, a fair amount of your setup sits outside that list.
Why you still have to check what AI tells you
When using AI for personal finances or wealth management, don't take what it says for granted. Always check things. Even if AI is right 95% or 99% of the time, there is still a 1% or 5% chance it could go wrong. And it can cost you a lot of money. Without domain knowledge, you might not notice the mistake.
So always check things, which is why it's better to work with professionals (tax, lawyer, accountant, investment manager, or CIO) who know their domain well.
You can outsource admin and operational work, but don't outsource your thinking and judgement.
How to build your own setup, one workflow at a time
Managing your own money requires your own system (an existing one or one you build). You can use existing tech, but you'll likely need to tailor it to your specific setup or situation.
So I encourage you to keep an eye on AI tools, work on agents and stay on top of things, but keep in mind that things are constantly changing, new tools and LLM models appear every month, new integrations, old integrations breaking etc.
Pick one workflow in your own money setup and write it out as instructions somebody else could follow:
- how a statement gets from your bank or custodian to your accountant
- how a deal gets looked at before you say yes
- where your documents live and who else can find them
Give it an hour. If you can't describe it clearly, no agent will run it for you either.
Building my own AI-native operating system
I have the same with my AI agents. I regularly review, update them, think of new frameworks and constantly improve. So it's a work in progress for me and not a finished job.
Don't get me wrong. AI is changing the way we operate and what we can do. But it's still early stage.
The way I approach it is that I am building an AI-native personal operating system for myself, where wealth and money management is only part of it. When it comes to managing wealth, there are also different areas - corporate structures, document vault, information sharing, data feeds, planning, investments, reporting and everything else around it. Each of those has its own tools, logins, and way of holding data.
I also regularly review my set-up and can completely change what I had a few months ago. That sounds wasteful, and sometimes it is.
So I treat this as work in progress. I can try splitting workflows and optimise parts of it. Once it's working, I move to the next workflow and work on that.
On the Radar
Only one in 18 US centimillionaires runs a family office.
Michael Thrasher put two numbers side by side. Owen Zidar and Eric Zwick, in The Everywhere Millionaire, count roughly 65,000 Americans worth $100m or more. According to Deloitte's 2024 figures, there are 3,550 family offices in North America. That is 5.5%. Almost all of these people are business owners, so the work is already being done inside companies they own, by people hired to do something else. Worth checking which parts of your own setup sit there.
Family offices are buying AI faster than they are using it.
65% of family offices plan to prioritise AI investments, on a February report from J.P. Morgan Private Bank. Alternatives now make up 42% of the average family office portfolio, according to UBS's 2026 survey of 307 offices. Mercer, meanwhile, found 6% of asset managers letting AI make an investment decision. Buying AI and using AI are two different budgets, and the first is moving faster. Next time a manager pitches you AI exposure, ask what they have automated in their own shop and what changed.
SEC cleared tokenised US stock trading for five years, with conditions.
The SEC's "innovation exemption", issued on 17 September, lets token-trading venues handle tokenised US shares without registering as a stock exchange. It runs for 5 years. The conditions are specific: caps on how many symbols and how much volume, tokens that carry the same rights as the real shares, written notice to the company whose shares they are, smart contracts anyone can read and audit on a public ledger, and trading halts that follow the main exchange. Comment is open on what comes next. Anyone pitching you a tokenised structure now has a list to be measured against.
One-page brief that does the job of a 40-page IPS.
Amin Naj wrote a replacement for the investment policy statement nobody reads. One or two pages, five questions: what the money has to do, with a number and a date on it; how big a fall you can live with; where your limits on concentration and borrowing sit; how cash gets raised and what sells first; and which decisions happen without you. His test takes an afternoon. Ask each professional you work with, separately, what your money is for and which calls they can make without you. Where the answers differ is where the drift already is.
Why private deals stall after they arrive.
Johnny Kao says the bottleneck in alternative assets is not finding deals. Deals arrive. The hard part is that each one turns up in a different shape, so nothing is comparable and every deal gets judged from scratch. Nothing adds up over time. This is a filing problem before it is an investment one. One template for how a deal gets written up, decided and reviewed turns a pile of one-offs into a record you can read back.
New on the Site
Last week's piece covered the other half of this: when the coordinating work gets big enough to hand to a person, and what that person costs at $30m, $50m and $100m. This one is about the systems underneath that job, the ones you keep whether or not you ever hire.
Read it: When to Hire Someone to Run Your Family Office
This was Capital Signals — weekly briefings on what's reshaping founder strategy on wealth.
Go deeper: Playbooks · Wealth Architecture · Investment Strategy · Business Building · Life Design
Found this useful? Forward it to a founder who's thinking about this stuff. Got a question or disagree with something? Get in touch.
New here? Subscribe for one email a week.
