Nobody mentioned a portfolio

On 14 September, Charles Schwab announced that the independent advisors on its platform could start their day in Claude. Three days later, Zeplyn announced that a wealth firm could define an agent once and make it available through Claude, ChatGPT, Copilot, Slack or the firm’s own applications, with the permissions, audit trail and human review defined in one place. Neither announcement mentioned a portfolio.

• Managed account assets reached $16.4 trillion at the end of March 2026, with $388 billion of net flows in the first quarter, according to MMI-Cerulli data. Every one of those accounts is implemented by someone: the asset manager or the program sponsor.
• AI is changing how an advisor enters a wealth platform. The advisor works through one interface that coordinates the applications underneath. The applications are still there, but they are no longer the way in.
• Interfaces are easy to replace. Investment workflows are not. The next generation of wealth platforms will be judged by how their applications work together as one investment workflow, closer to an operating system than a product suite.
• An operating system does not mean a single vendor. The test is whether a component built by someone else can work inside the same governed workflow, with the same entitlements and the same audit trail.
• AI that reaches the investment layer has to meet four requirements: it sees only what the user can see, it answers from retrieved data, it refuses rather than guesses, and it does not trade.

Managed account assets reached $16.4 trillion at the end of March, with $388 billion of net flows in the first quarter alone. UMA programs drew $159 billion of that, separate account programs $115 billion, and rep-as-portfolio-manager programs $76 billion.

A decade ago much of that money would have sat in pooled vehicles: one portfolio, one set of holdings, one reconciliation, one set of restrictions. Now it sits in thousands of legally distinct accounts, each with its own holdings, many with their own restrictions and tax lots, and at least one whose owner will not sell the shares her grandfather bought, whatever the model says. And every one of them is implemented by someone. In a manager-traded SMA, the asset manager. In a UMA, a model-delivered SMA or in a rep-as-PM program, the sponsor.

Whether you manage the strategy or run the platform, the work of running each account is yours. The front door to those accounts is moving. The work is not.

The old model of a wealth platform was an advisor moving between applications: planning, CRM, rebalancing, compliance, trading, reporting. Each had its own login and its own idea of the client. The advisor was the connective tissue, which is one reason Capgemini finds 41% of advisor time going to operational tasks.

The model the September announcements describe is different. The advisor works through an interface that coordinates the applications underneath: CRM, custodian, planning, portfolio, compliance. The applications are still there. They’re just no longer the way in.

When the interface sits above the applications, what a platform is changes. It stops being the set of screens it ships and becomes the set of workflows that can be entered from anywhere. A planning tool that can only be used from its own screen is a tool. A rebalance that can be requested from a prompt, checked against the account’s restrictions, and held for a portfolio manager’s approval is a workflow. Interfaces are easy to replace. Investment workflows are not.

So the next generation of wealth platforms will be judged less by the applications they include and more by how those applications work together as one investment workflow. Closer to an operating system than a product suite.

An operating system does not mean a single vendor. The capabilities clients now expect, tax management, direct indexing, alternatives, planning, often come from specialists, and firms will keep adding them. Nobody builds all of that. Nobody should try. Direct indexing alone is now $1.2 trillion, the largest manager-traded SMA strategy by some distance.

So the question for a platform is not how many components it owns. It is whether a component built by someone else can behave as part of the same governed workflow: the account the advisor is looking at, the instrument the trader selected, the order awaiting approval, carried into the partner’s component and back out again, with the same entitlements and the same audit trail.

I wrote about this theme in the context of investment operations recently. The layer you standardize on and the components you buy underneath it are two different decisions, and only one of them should be hard to reverse. The wealth version is the same argument with a faster clock. Partners will change. Interfaces will change. The workflow layer, and the governance inside it, is the part a firm must be able to keep.

The first wave of AI for advisors reaches research, meeting preparation and documentation. That is useful work, and 76% of advisors say they want it automated. But the cost and the risk of a managed-account program sit one layer down, in the investment layer: a model change reaching every account that follows it, a rebalance inside each account’s restrictions and tolerances, every order tested against compliance rules before release, and a named person accountable for the release. A team managing 50 models is really managing every account attached to them.

An AI interface that reaches the meeting notes and stops there does not change which accounts a firm can run, or how many. An AI interface that reaches the investment layer changes both. And it has to be built differently, because at that layer a wrong answer is not an awkward paragraph. It is a trade.
Four requirements follow, and I would hold any vendor, including us, to them.
 

1. It sees only what the user can see. The user’s own identity travels with every request.
2. It answers from the book, not from memory. Every factual claim rests on retrieved data, cited to the account, the instrument and the as-of time.
3. It refuses rather than guesses. Missing data, or a question the user is not entitled to ask, produces a plain refusal, not a plausible sentence.
4. It does not trade. It can prioritise the alert queue, explain a compliance result, propose where to start. The order is placed by a person with the authority to place it.

None of this is a limitation on AI. It is what makes AI permissible and useful next to a portfolio.

Longview Horizon is Linedata’s front-office platform: portfolio management, model management, trading and compliance as domains inside Mosaic, a workspace that uses the FDC3 standard to carry account, instrument and order context across Linedata, partner and client-built components.5 Firms add capabilities as modular components and replace them independently. BondCliQ already connects this way.

The AI capability we are bringing into that workspace reads through the platform’s own APIs and is being built to the four requirements above.
The advisor’s front door has moved before. From the branch to the browser, from the browser to the phone, and now to a prompt. It will move again, and nobody in Boston this October knows where.
What sits behind it has to be built to be entered from anywhere

Request a demo

Thomas Dadmun leads Linedata’s global product management team, developing portfolio management and investment operations systems for asset managers. Previously, he founded Point Focal, where the company explored natural language processing (NLP) and alternative data for trade signaling and risk management. He has worked in data strategy and product analytics at State Street and began his career as an equities trader. Thomas also taught in Northeastern University’s Master’s in Analytics program and enjoys working at the intersection of markets, data, and technology.