The client record is the expensive part of plain-language AI.

AdvisorCRM and Zeplyn have each rolled out platforms that let an advisory firm describe the tool, workflow or agent it wants in plain language and get it, no developers required, and the two launches are contesting something less glamorous than the pitch: where the firm's own record lives, and who gets to route it.

Zeplyn, which describes itself as an AI operating system for wealth management, paired its Agent Studio with a companion model context protocol server, giving AI applications that support the protocol — Claude, ChatGPT, Copilot and Slack among them — governed access to agents built in the studio, along with Zeplyn's client intelligence and more than 60 wealth-specific tools. AdvisorCRM put Studio into limited release at this week's Future Proof Festival, with the first 50 participating firms working alongside its own team, and says each firm operates in its own dedicated database, structurally separated so that one firm's problems stay inside its own environment.

Demand is not the constraint, since MSCI's Wealth Trends 2026 report found roughly 95% of financial services and wealth management professionals planning to increase AI investment over the next three years, even as the same research found many firms unable to convert that appetite into deployment. The usual reading of that second finding — that the firms lack software — is the least likely explanation.

The recent pattern says as much: AI has kept arriving through systems advisors already open, as with Practifi's Sentir folding into the CRM's core in August, and through the model vendors themselves, whose workflow specificity drew analyst attention this month. Each wave has settled the build question and left the deployment question standing.

Ryan Borer, managing partner at AdvisorCRM, framed his own launch around that leftover: when a firm can build almost anything, he observed, the harder problem becomes choosing what to build first. That is a candid line to ship alongside a product, conceding that taking the developer out of the loop does not take the executive out of the room. Firms that have spent two years buying notetakers and summarizers have the software; the missing decision is which workflow to own, who maintains it after the pilot ends, and which system of record it writes to.

That last question is where the two launches actually compete.

AdvisorCRM's more consequential claim is about data rather than generation: one environment that centralizes records across whatever a firm builds on it, aimed at the duplicate entry and fragmented files that come from stitching disconnected systems together. Building tools is getting cheap; a governed, firm-specific copy of the client record is not.

Zeplyn's protocol server points the same way from the other side. Exporting agents into Claude, ChatGPT, Copilot and Slack is a decision to supply the interfaces rather than fight them, and it puts Zeplyn where this publication has argued the value in AI-for-advisors accrues: with whoever owns the connector. Practifi's Sentir took the other bet, keeping the interface inside the CRM. One of these two is selling plumbing.

The outputs Zeplyn lists are revealing for their ordinariness: an annual review brief, a meeting recap, an email drafted in the firm's own voice. Every one of those is assembled from material the firm already owns — the meeting, the account, the relationship — which makes the firm's ability to hand a model a clean, permissioned view of its own client the scarce input; model capability is the cheap part. That is a data-governance problem wearing an AI costume, and it looks the same whether the tool is described in plain language or bought off a shelf.

There is a version of the plain-language thesis that cuts against the house argument: if a firm builds its agents on its own dedicated database and wires its own connections, the routing layer loses pricing power and the connector premium gets narrower. Zeplyn's embrace of the protocol suggests that even a company whose product is agent-building is not betting everything on the walled interface holding. Read the launches together and the tool layer keeps cheapening while the contest stays where it has been all year — over the record and the route to it.

For advisory firms, what separates pilots from deployment is auditability. A dedicated database and a governed protocol endpoint are claims a compliance officer can put questions to and get documents back for. An agent that drafts client email in the firm's voice has to survive a books-and-records review before it travels past the pilot. The RIA that deploys furthest over the next year will be the one whose compliance team can say where the client record sits and who touched it.

The go-to-market is the tell. A limited release that seats 50 firms next to the vendor's own team is a services engagement as much as a software release — a sensible way to learn which workflows advisors will actually commission, and a slow way to reach the much larger population of RIAs that will eventually want them. Closing that distance requires the custom builds to harden into a catalog and the catalog to be reachable from the systems firms already run, which points the next version of this story at distribution through the custodian, the CRM and the planning suite. Whether that routing happens through the platforms or around them is an open question, and nothing in either launch settles it.

For now the plain-language builders are answering the easier half of the deployment gap first. The evidence that would move this from launch to category is specific and visible from outside: named workflows, data models that compliance has signed off on, and firms still building a year from now with the vendor out of the room. Fifty firms is a cohort, with the market still ahead.