Essay · Agentic wealth
Wealth management after superintelligence
Most of what this industry has built should be scrapped and started again, around agents, with superintelligence at the core. What becomes obsolete, what should exist, and why the companies holding the data are choosing opposite directions.
AI lowered the cost of building software. This month it showed what it does to expert work.
On October 6, OpenAI released about 700 mathematical papers with roughly 370 results that solve or advance open problems. Each took around three hours of compute, and software checked the proofs.[1] Work that would have taken people years took an afternoon of machine time.
Every firm in this industry, and every company that sells to it, was designed before that was true. I think most of what has been built should be scrapped and started again, around agents, with superintelligence at the core. That includes the software, the roles, the way the firm is organized and the way experts work with their team, their clients and their agents.
The first thing to go is the work no one wants. Tracking facts and checking them. Managing the people who do it. Keeping a stack of software in agreement with itself.
Work that runs in a loop
A mathematician proposes an idea, tests it, learns why it failed and tries again. Software can now run that loop thousands of times at once, check each attempt itself and isolate what’s useful.[2] Code went first because a program either runs or it doesn’t. Math went next because a proof either checks or it doesn’t.
A wealth firm’s operations are mostly loops with a check at the end. Did the transfer settle? Does the plan match the accounts? Did the signed form come back? Is the allocation still inside its bands after this week’s market? Who still owes the client a call?
Today a person runs each of those loops, one at a time, on a schedule. The check is whether anyone noticed.
Agents will run all of them, all the time, against a record that shows when an answer changed. That is the first call: the checkable work in a wealth firm is going to agents. This is going to happen faster than most believe.
The plan that runs every future
Financial planning already has one loop that runs thousands of times. A Monte Carlo simulation takes a household’s plan and runs it against thousands of possible markets to see how often the money lasts.
It varies one thing well, mostly the returns, and holds nearly everything else still. Income, spending, tax law, health, a business sale, an inheritance, when to claim Social Security, when to convert to a Roth: each is entered once as an assumption and left alone until the next review.
A family’s plan is not one variable. It is hundreds of facts that change on their own schedule, and they interact. A move changes state tax, which changes the conversion strategy, which changes the order of withdrawals, which changes what the children inherit.
Superintelligence can run the whole plan as a loop. Every fact and every open decision, tested across thousands of futures, rerun the day any fact changes, with each result traced back to the facts it used.
The output stops being a probability of success on a page. It becomes a short list of the decisions that matter most for this family now, what each one changes, and how much confidence the evidence supports.
That is what planning could be. The advisor’s job moves from producing the analysis to choosing among it with the family.
The portfolio can follow the plan the same way.
Personalized indexing today is mostly a tax tool. Owning positions directly, losses are harvested and a few exclusions and preferences are built in. It is set up once and run against rules, often apart from the plan.
An agent can run it against the whole plan instead. It knows the family’s income this year, the gains already taken, the Roth conversion planned for December, which account each holding belongs in and why, and the concentrated positions that have their own separate plan to unwind over five years.
It trades as those facts and the markets move rather than on a quarterly schedule, within the firm’s written method and with every trade on the record. Tax management and asset location stop being separate jobs.
And it can explain itself. Ask why a lot was sold or why the bonds moved to the IRA, and the agent answers with the facts it used and the part of the plan it served. The advisor and the family can see the state of every account and how it connects to their situation, whenever they ask.
With WilliamFrank, the family chooses who they ask. When they would rather not take up a person’s time, they ask the agent. When they want their advisor’s perspective, that is just as easy to get.
The work with no check at the end stays with people. Whether to sell the business. What the money is for. How much risk a couple can carry when one of them is ill. That work gets more of the advisor’s week once the loops are gone.
The screen stops being the product
In April, Salesforce opened its whole platform to agents. With Headless 360, an agent can read and act through Salesforce without opening a browser.[3] This month Elon Musk said Grok’s agent will pick whichever model is most likely to give the user the best result.[4] When a service works without its interface, an agent reaches it directly.
Wealth technology is sold as screens. A CRM, a planning tool, a portal, a rebalancer, a document vault. Each has its own login, its own copy of the household and a price per person who logs in. What the vendor protects is the interface and the data behind it.
An agent needs the record, permission to act and a log of what it did. It does not need the screen. A vendor whose product is the screen is selling the part that will be used least.
Models move the same way. The best one changes every few months. A firm built around one model inherits its limits. A firm built around its own record and its own written method can use whichever model is best this quarter, and the one after.
Open the door, or charge for it
The companies that hold the industry’s data are choosing between two responses.
Salesforce chose to open the door. Orion chose to charge for it.
This month Orion told the vendors that connect to it that it will charge a monthly Connected Ecosystem Support Fee, reported at about $2,000 to $20,000 a month per vendor, with usage-based pricing to follow in 2027.[5] Orion says the fee funds its investment in AI and data infrastructure.
The data is about advisors’ clients. Advisors and families put it there. Whoever pays the fee first, advisors and their clients will likely pay it in the end.
This is the old design defending itself. When the value of a system was its screen and its copy of the household, controlling access to that copy was the business. In a firm built around agents, a record an agent has to pay to read is a record the firm does not control.
The question for every firm is the same one: who owns the schema, and can an agent read and write across it? A vendor that owns the schema owns the model of your households.
Superintelligence at the core
Most of the industry is adding AI to the systems it already runs. A note-taker in the meeting. A drafting tool in the inbox. A chatbot in the portal. Each speeds up a step a person still owns, on rails that were built for people.
A firm built for superintelligence starts from the capability. What could a system far more capable than any analyst do for a family if it had every fact, the firm’s method and permission to act? Then the firm decides where people belong.
Three things have to be true for that to work.
The record is the product. Every fact about a household carries its source and its date, and agents read and write it directly.
The method is written down. How the firm allocates, sequences withdrawals for tax, runs a review and decides when to call. An agent can follow a method only if it exists outside someone’s head.
Every action shows its work. Math moved fast because a machine can check a proof. Advice needs its own version of that: what the agent saw, what it proposed and who accepted it, on the record. Supervision becomes reading rather than reconstruction.
Designing how experts work
The screen stops being the product, but it does not disappear. It changes jobs. It stops being where the work is done and becomes where judgment happens.
Almost every interface an advisor uses today was built for a person to enter data and find it again. Few were built for an expert directing agents. That design work is ahead of this industry, and it is as large an opportunity as the agents themselves.
Three relationships need designing from scratch.
The expert and the agents. Agents provide context and propose, people adjust and accept. The expert sees what an agent saw, what it proposes and what would change, and decides in one place.
The expert and the team. The leader stops assigning tasks and chasing status. The team reviews agent work that shows its sources, and the leader sees where judgment was used and by whom.
The expert and the client. A family should not log into a portal to hunt for a statement. They should see what changed, what was done about it and what is waiting on them, in plain language. Or they ask WilliamFrank a question in plain English, and a visual or a dashboard comes back to help answer it, with a suggestion to connect with their advisor when the question calls for one. Their time with the advisor goes to the decisions that are left.
This takes people who know the work and people who know design, building together from the start.
What becomes obsolete
Overnight integrations. A copy of the household that arrives tomorrow is already out of date when an agent acts today.
Data held behind a toll. A firm should not pay a vendor to reach the facts about its own clients.
The work no one wants. Keying the transfer, chasing the form, getting the planning software to agree with the custodian, managing the people who do it and keeping the software stack connected. Those seats become the ones that design the agents and check that they work.
Software priced per user. When agents do most of the logging in, the price has to follow the work.
Minimums set by cost. Tailored financial advice has been rationed for decades by cost structure rather than by intent. When the cost of serving a family falls, a $2 million minimum becomes a choice.
Advice that is only analysis. A family can get much of a plan’s analysis from a free tool in an afternoon. That part of advice will be priced close to zero. Families will keep paying for judgment, accountability and someone who knows them.
A single probability of success as the answer. A number from a model that varies one input tells a family little about the decisions in front of them.
Portfolios managed apart from the plan. Loss harvesting that does not know about this year’s Roth conversion, and asset location set once and left alone.
Interfaces built for data entry. An expert directing agents needs a different surface from a person keying a form.
The review spent on fact-finding. When the first twenty minutes of a meeting go to finding out what changed, the loops above were not run.
The org chart that grows with every new service. Each specialty adds a person, and each person adds someone to manage. The leader’s week fills with tracking and supervision before it reaches a client.
What should exist
Firms designed around agents from the start. The record, the agents, the surfaces and the roles built at the same time, by people who know the specialized type of work.
A plan that runs as a loop. Every fact and every open decision, tested continuously, with the decisions that matter brought forward.
A portfolio that follows the plan. Personalized indexing, tax management and asset location run together by an agent that can say why it did what it did.
Interfaces designed for experts. One place where the advisor works with agents, the team and the family, and every item shows its work.
A leader whose week goes to clients and decisions. Agents keep track of what changed. The people the leader manages review agent work that shows what it did.
An advisor who walks into every meeting already knowing what changed, with the work done or drafted and the decisions laid out.
A middle seat between the advisor and the engineer. Someone who designs the agents, writes the method down and owns whether it is any good.
A far better work product for the fee. Advice that is more efficient on fees, taxes and risk, delivered as personal, timely service from an expert who knows the family.
Software the firm owns at the core. Custody and reconciled performance can be rented. The household’s record, the method and the agents belong to the firm.
Agentic wealth
The people who will feel this first are the ones running a team. Their week is where the old design costs the most: keeping track of everything that changes, and managing the people who help.
Agentic wealth moves both onto agents, and the leader’s time goes to creating value.
If you have built a firm or a product for the last era, ask what you would keep if you started today. I would keep the people, the clients and the method. I would rebuild everything else around agents, and that is what we are doing, in the open, with a founding consortium of advisors.
Sources: [1] OpenAI’s October 6, 2026 release of mathematical documents, as reported by The New York Times (October 6) and Scientific American; Terence Tao’s statement on the release (October 7) · [2] All-In Podcast, Episode 292, October 10, 2026: David Friedberg on loops that can be run and checked in software, and David Sacks on why code and math moved first · [3] Salesforce, Headless 360, announced April 15, 2026 at TrailblazerDX; discussed by Chamath Palihapitiya on All-In Episode 292 · [4] All-In Podcast, Episode 292: Jason Calacanis on Elon Musk’s announcement for Grok’s agent · [5] Davis Janowski, “Wealthtech Industry Scrambles As Orion Introduces Data Access Fee,” WealthManagement.com, October 8, 2026.