Pontera reframes itself as AI-agent gatekeeper as Schwab readies Charley agents

Pontera, the New York City software company that manages 401(k) accounts, spent late September arguing that its real product is permission — the layer that decides which AI agent may touch an account and what it is allowed to do once inside. The case arrived in a Sept. 24 blog post by Ben White, the firm's senior director of public policy, titled "Agentic AI reframes the debate over financial access," drawn from takeaways at the 2026 FDATA summit in Toronto earlier that month. RIABiz framed the post as a possible bargaining chip with the financial institutions Pontera has spent years trying to win over, on the argument that agentic AI is arriving ahead of those institutions' ability to manage it.

In White's telling, the industry is racing toward hyper-automation without the plumbing to make it safe: much of the financial ecosystem is not yet built for agent-to-agent communication, many firms lack the infrastructure to distinguish a person from an authorized agent, and not every institution can govern automated activity on its own. Pontera's answer is to serve two roles at once — a trust layer for the consumer and a protection layer for the institution — validating users, scoping permissions, minimizing unnecessary data sharing and keeping records of activity.

Joel Bruckenstein, who sponsors the T3 events, was careful not to endorse the framing, telling RIABiz the field is still emerging, that any assumptions made today might be upended within months, and that he had no specific read on whether Pontera is being clever or desperate; on Fidelity's stance, he hoped the firm would continue to monitor the situation and make adjustments as technological advances permit.

That year-long fight over Fidelity's ban on credential sharing, which RIABiz has covered, escalated when Pontera's chief executive published an open letter pressing the plan provider, and it has cast Pontera as a privacy threat that could run afoul of the DOL rules plan sponsors live by. The two firms have not found common ground to date; Pontera has proposed various ways it might work with Fidelity, and the RIABiz account does not say whether any of them have moved.

Schwab's announcement lands on the same territory: as RIABiz reported, the custodian said it is prepared to unleash "Charley" AI agents to serve its millions of clients, coming on the heels of Meta's agent, Muse. But the account does not say whether Charley reaches retirement plan accounts or data held outside Schwab — exactly where Pontera's argument lives — nor how those agents would be authorized.

For a firm that owns both the account and the client record, validating its own agents is cheaper, but Pontera is betting that institutions will not want to referee their own agent traffic, or will not be able to at the speed agents multiply, and that a company already sitting between a client and a third party's plan record is a plausible place to put the gate. That is a bet on the limits of the incumbents' engineering, placed while a custodian with millions of clients moves toward the same ground.

This publication has argued that model access is not where the AI money sits: Schwab's exclusive with Anthropic bought queue position rather than a better model, and when a billion-dollar RIA left the custodian for Altruist in September, the read was that the startup's data lead of two to five years is not a gap a partnership can purchase. White's line about maintaining records of activity is the same argument from the other side — whoever holds the log of which agent acted, under whose permission and for how long, holds the governed client record.

The platform fight has been drifting that way all year, moving off custody basis points and onto the client record and the cash spread; an agent trust layer is a bid for the authoritative version of that record — who the client is, what they hold, and who may act on it. A 401(k) software company's blog post, read in that light, is a platform argument.

The policy layer is where the reframing does real work: if a human advisor logging in with client credentials was the privacy problem, an agent acting on that account without a human in the loop is a harder version of the same question, and one that a credential-sharing ban does not answer. White's proposal is that a validator absorb that risk on the institution's behalf, but independent validation is a job that has to be granted before it can be performed.

For an RIA, the practical version of the question is narrower than the theory: if a custodian's agents become the surface a client touches, the validation layer sits either inside that custodian or in a third party the custodian blesses, and Pontera is applying for the second job. Nothing in the posted argument settles whether it gets it. Bruckenstein's caution, that assumptions made today may be upended within months, is a fair warning against drawing the org chart before the products exist.

Fidelity is where this becomes testable, because a trust-layer argument gives the plan provider a reason to revisit a dispute it has held through a year of public pressure, and RIABiz does not report that any such conversation is under way. Until one side moves, Pontera is describing a job nobody has hired it to do while Schwab prepares to serve millions of clients with agents of its own.