Essay & podcast · Agentic Wealth

Designed around agents

By Bill Woodruff, with WilliamFrank

AI lowered the cost of building software. It also changed what the software should be, and what the company around it should look like.

A system designed around agents is a different object from the software this industry runs today. The firm built on it has a different shape from today's firm. Both have to be designed at the same time, and that is only possible from scratch.

What changes when the agent is a user

Software today is a set of screens a person operates. Every screen assumes the person is the only actor. The person reads, the person decides, the person clicks, and the system records the click. If you stop to think about it, no human wants to learn how to hunt around and find what they are looking for in a digital user experience, especially when they have multiple applications to interface with.

Give an agent credentials, a record it can read and write, and a log of what it did, and the person stops being the only actor. Most of the reading moves. Most of the clicking goes away. The person's job becomes deciding, and in order for an expert to decide, agents need to organize and effectively display the context.

Five decisions follow. I made them in the prototype and I now think they are general.

One record. The agent and the person work from the same household, so there is one household. Every copy is a place two actors can disagree.

One interface. A person says what they want in the same place every time, and routing is the system's job. Routing is shown on the way out so it can be redirected.

Agents provide context and propose, people adjust and accept. Every write carries a named actor. What you see is a queue of proposed work, not a set of forms.

Nothing is deleted. A correction supersedes the record it corrects. The record is the audit trail, so supervision is reading rather than reconstruction. This serves both a compliance record as well as an opportunity for the AI to recursively improve.

Simplicity over borrowed features. Keep what makes mail and messaging work and drop the customization surface. An agent does not need a settings page.

Those decisions only make sense when the agent is a user. Software that treats the agent as a feature cannot make them, because every screen still assumes a person is the only one there.

What changes in the company

Software sets roles. Every role in an advisory firm exists because at some point a person had to do something no system could do. Someone keyed the transfer. Someone opened the statement, found the cost basis, and typed it into the plan. Someone sat between the planner and the custodian and got the two to agree. Buy a system and you inherit the roles it assumes.

Design the system around agents and the roles become open questions. What is this seat for once the administrative work is done autonomously? Which decisions require a licensed person, and which only ever required one because a person was the only thing available? Where does a person review an agent's work, and how is that review recorded? Who owns whether the agents are any good?

The answers change the firm in three places.

The expert's job narrows to judgment. Planning, investment decisions, and the conversation about what the money is for. Thirty minutes or less at a time, then on to a personal relationship oriented conversation or on to the next household.

A middle seat appears. Between the people who sit with clients and the engineers who build, someone has to design the agents and workflows and make sure they work. Most firms have both ends and nothing between.

Hiring changes. The question stops being years in a role that will not exist in that form. It becomes whether the person has built things, and whether they can say what an agent should not be trusted to do. Supervision becomes review of work that shows what it did, so the cost of a hiring mistake falls. That is what lets a practice grow past the first hire that if the owner gets wrong causes a whole lot of pain.

Why from scratch

An incumbent carries two things it cannot put down: an old core and an old org.

Two of them showed it in one week in August. LPL spent close to $2 billion over three years and produced Latitude, four of whose five pieces were already running.¹ Vanguard bought Altruist and said it will run standalone, with its leadership, brand and operating model intact.² One built on top of the old core. The other acquired a new one and left it beside the old one. That is an annex, and an annex has a seam where the batch record and the real-time record have to agree.

Every large custodian and self-clearing firm is built on batch, and none are focused on fixing it.

The constraint on redesign was never capital. A firm with ten thousand people on one set of rails cannot change what a role is for without changing what the rails carry, and cannot change the rails without stopping the firm. Starting from scratch is what lets the system and the roles be designed together.

That is what one integrated set of agentic rails for a wealth team means. The record, the agents, the surfaces and the roles, designed at the same time.

What we rent

Not everything is rebuilt. Custody stays with a custodian. Reconciled performance carries a substantiation obligation, and that is a good reason to rent it, at least early on.

What gets built is the layer that decides what a household is. The fact ledger. The schema. The timing. The agents. The surfaces a client and an advisor touch. A rented system feeds that ledger through an adapter and does not hold the record.

The test for a vendor fits in one conversation. 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.

Three counters

Retrofits may get close. A market of connectors is forming over the old systems, and if it matures quickly the rails become reachable without a rebuild. Jump, Zocks and PreciseFP already write into planning software. The answer is the schema test above. A connector gives an agent someone else's model of the household.

People may not want to be supervisors, while agents are new they will need to learn and be trained just like a junior hire. An advisor who built a practice on doing the work may not want a week spent reviewing an agent's work. Some will not, and that is why the consortium is by application. It is for people who want the administration off their desk and also understand what is possible.

Determinism. Advice at scale needs outcomes that clear every time, and old systems hold decades of exception handling nobody remembers writing down. A firm that rebuilds the layer underneath has to earn that again. This is why this should get built by and with advisors instead of for them by those with Fintech or enterprise software backgrounds.

The number

McKinsey surveyed 1,993 people across 105 countries this year. Thirty-nine percent report any enterprise EBIT impact from AI, and for most of those it runs under five percent. The firms with the most impact were close to three times as likely to have redesigned how the work itself gets done.³

Agentic wealth

Agentic wealth is where this profession is going.

The best advisory firms of the next decade will run on one integrated set of agentic rails. One record of the household. Agents that carry the administration and organize the context. Advisors who decide.

An advisor at one of those firms will spend the week on judgment and on relationships. Less of it will go to tracking facts, or to managing the people who track them. The firm will know every household as well as it knows its best one, and pay less to prove the work, on a record that makes the agents better every time a person corrects one.

A firm that bolts an agent onto the systems it already runs will get a faster version of what a person could already do. A firm that designs the system and the roles together will set the standard the rest are measured against.

This is going to happen faster than most believe. I am building for it in the open, with advisors, and anyone who wants to build it with us can apply.

By Bill Woodruff, with WilliamFrank.