The May 15 piece on access-rents drew a line through every services industry. On one side: access-rents — work that consists of operating an interface the customer cannot operate themselves. On the other: integrated expertise — work that consists of telling the customer which question to ask. AI eats both, with opposite policy implications.
Wealth management is the industry that has the largest mix of both, sitting in the same company, frequently in the same advisor. The agent shock that is coming for this industry will be uneven in a way the public discussion has not started to model.
The Workflow Surface
A wealth-management workflow, observed from the operator's side, has roughly twelve recurring operations.
Account aggregation across custodians — pulling positions, balances, and transactions from Schwab, Fidelity, Janney, IBKR, Empower, smaller firms. Transaction categorization for tax and reporting. Performance attribution against benchmarks. Rebalancing signal generation against a target allocation. Tax-loss harvesting candidate identification. Cost-basis tracking across in-kind transfers and corporate actions. ACATS execution. RMD calculation. Backdoor Roth conversion mechanics. Estate-planning vehicle structure across multiple entities. Client behavioral coaching during drawdowns. Strategy revision after a life event.
Half of those are mechanical. Half of those are judgment. The mechanical half is where agents already work today. The judgment half is where they do not, and where the durable expertise lives.
Where Agents Already Work
Account aggregation has been agent-tractable for ten years. Plaid and its competitors have been doing it under different vendor names. The underlying technology — credential vaulting, OAuth-where-supported, scraping-where-not, transaction normalization — is mature. The integration cost is non-trivial but bounded.
Transaction categorization is a solved supervised-learning problem. Consumer apps have been categorizing personal transactions since Mint shipped in 2007. The same models work for advisor-facing categorization at higher quality with feedback loops.
Performance attribution is table joins and arithmetic. Benchmark selection has some judgment in it; the math against the chosen benchmark is mechanical.
Rules-based rebalancing — if allocation drift exceeds threshold, generate the trade list — is a working agent today inside every major robo platform. Wealthfront, Betterment, Schwab Intelligent Portfolios all run automated rebalancing as the core product.
The agent layer for these workflows is not a future problem. It is a present commodity. The advisors whose primary value-add is operating these systems on a client's behalf are doing access-rent work. Their position is the same position the travel agent occupied in 2002.
Where Agents Still Do Not Work
Multi-entity tax strategy across a household — when to convert, how much, against which marginal rate, against which projected future bracket, against which Roth conversion ladder, accounting for state-level interactions — is integrated expertise. Models can generate options. The judgment that selects between options under specific client constraints is not in the model. The advisor who runs this kind of optimization for a client has a moat, not a rent.
Rebalancing under tax constraints crosses the threshold. A rules-based rebalance is mechanical; a rebalance that knows not to harvest the loss in the IRA because the wash-sale rule reaches across accounts, and that schedules the gain realization across calendar years to avoid an NIIT spike, requires the kind of context that lives in the advisor's head.
Behavioral coaching during drawdowns is the part the academic literature has been studying for thirty years and that DALBAR keeps measuring as the largest single source of advisor value. The client who calls in March 2020 and asks to move everything to cash is not asking for an answer; they are asking for a counterparty. The advisor who is that counterparty has work no agent currently replaces.
Estate planning across vehicles — a revocable trust, a charitable remainder trust, a 529 with appreciated stock from an employer ESPP, a defined-benefit plan that has been in run-off mode for six years — is the kind of multi-entity, multi-rule integrated expertise that no current model has the working context to author. Generating draft documents is easy. Deciding which draft to use is not.
The Procurement Bite
The May 17 piece on substrate argued that the policy surface for AI inside organizations is decided at the procurement layer, not the policy committee. The wealth management version of this is the data-feed layer.
Plaid TTLs are short by industry standard. Bank credential reauth typically resolves in days. Brokerage credential reauth is often shorter — most retail brokerage feeds require monthly reauth at minimum, some weekly. The actual durability of the agent layer that operates against any household's accounts depends on the worst reauth cycle across all the accounts in that household. If even one custodian has an aggressive reauth policy, the whole stack inherits that policy.
Custodian feeds vary wildly. Schwab through its standard Plaid path has different rate limits, different transaction-history depth, and different reauth cycles than Fidelity through theirs. Empower's institutional feed for 401(k) data has its own rules. Janney's broker-dealer feed has a much shorter TTL than the consumer-bank standard — recurring re-auth requests are documented as expected behavior, not as a defect. The agent that operates across all of these inherits the most restrictive constraint of the bundle.
The advisor-tech firms that decide today which custodian feeds to integrate first, and which corporate actions to normalize across feeds, are deciding the policy surface for the entire household-level agent layer of the industry. By the time the official agent rollout from any individual brokerage arrives, the cross-custodian agent layer will already be sitting on top of it, and the brokerage that locked down its feed too aggressively will find itself routed around.
The Advisor Split
The advisors most exposed to substitution are the ones whose primary value-add sits between the client and a custodian API. The advisor whose deliverable is a quarterly performance report, an annual rebalance, and an occasional ACATS — that work is access-rent. The robo platforms are already running it for ten basis points. The full-service advisor's spread above the robo is paying for relationship continuity and occasional judgment. The judgment fraction of that work shrinks as agents handle more of the mechanical surround.
The advisors least exposed are the ones whose primary value-add is multi-vehicle, multi-rule, multi-stakeholder judgment under tax and estate constraints. A family with a closely-held business, a defined-benefit plan, a real-estate concentration, and three generations of heirs needs the kind of integrated reasoning that does not collapse into an agent prompt because the constraints do not fit into a prompt. The fee structure for that work survives. The fee structure for the access-rent work does not.
This is the same split the May 13 piece on AWS Cloud Support engineers described at a different industry — the L1 work falls to agents first, the integrated-expertise work falls to agents last, the badge of having operated the L1 interface for eighteen months loses signal value almost immediately. The wealth-management equivalent is happening on a parallel curve.
What This Looks Like in 2027
The robo platforms will offer richer agentic layers, with the Plaid-style feeds plumbed in by default. The full-service brokerages will offer hybrid agent-plus-advisor products at lower spreads. The independent RIA shops that built integrated-expertise practice — multi-entity tax, estate-across-vehicles, behavioral coaching, business-owner work — will continue to charge premium fees and grow market share. The independent RIA shops that built access-rent practice will lose market share at the lower-fee end of their book and try to move upmarket.
The cross-custodian agent layer that operates across multiple feeds — household-level rebalancing, tax-aware harvesting across accounts, RMD coordination — will exist as third-party software before any single custodian builds it natively. The firms that own that layer will have the same relationship to the custodians that Plaid has to the banks. Some of them will get acquired. Some will not.
The decade-long erosion that hit travel agents between 2000 and 2010 — LA Times documented the 2002 commission elimination as the inflection — is the rough analog. The mechanical work fell to interfaces the customer could operate themselves. The advice work survived where it had been integrated expertise; it died where it had been a rent on the access surface.
Wealth management is in the 2002 moment now. The five years that follow are the redistribution.
Top comments (0)