Retail-brokerage
Retail brokerages converge on agent trading but diverge sharply on guardrails: confirmation-required, sub-account budgets, and advisory-only sit side by side
PYMNTS' August 31 survey of Scalable Capital, Robinhood, Interactive Brokers and Public shows four brokerages answering the same product question, let an LLM trade for the customer, with four different answers on where the human stays in the loop.
The headline fact is not that retail brokerages are wiring LLMs into trading accounts, that has been underway across the industry for over a year, it is how differently the four firms PYMNTS surveyed have drawn the line on autonomy. Scalable Capital's Agentic Investing, built on the Model Context Protocol so a customer can point ChatGPT, Claude or Grok at a live account, still requires explicit confirmation on every trade; the firm is explicit that this is a first step, not the end state. Robinhood's Agentic Trading goes further, an agent operates inside a dedicated sub-account with its own budget and can view the customer's full portfolio across accounts before acting, but the trading authority itself is scoped to that sub-account rather than the whole book. Interactive Brokers' Ask IBKR stops short of execution entirely, it answers portfolio questions and does not connect to outside models at all. Public sits at the permissive end, an in-house agent with Claude Desktop connectivity via MCP that trades stocks, ETFs, options and crypto directly.
None of these four is obviously wrong, and that is the point worth sitting with. A confirmation-required design like Scalable Capital's trades a slower, more supervised workflow for lower tail risk from a misreading agent; a sub-account-budget design like Robinhood's caps blast radius by construction rather than by requiring a human to catch every bad decision in real time; an advisory-only design like IBKR's removes execution risk altogether at the cost of the convenience that is presumably driving demand for these products in the first place. What the industry has not converged on, a year into agentic brokerage rollouts, is which of these tradeoffs regulators or customers will eventually treat as the acceptable baseline and which will look reckless in hindsight.
For readers following the regulatory-boundary and risk-governance threads on this site, the practical question these four designs raise is not which one is safest in the abstract, it is what happens at the seams: a sub-account budget that is technically capped but large enough to matter if the agent burns through it on a single bad read of the order book; a confirmation-required flow where the customer rubber-stamps agent recommendations without real scrutiny, which functionally converges on full autonomy without the governance that was supposed to come with the extra step; an advisory-only product whose answers a customer then manually executes, moving the accountability question from the brokerage to the customer's own judgment. None of the four firms in this survey has published incident data or audited outcomes for their agent products, so for now every one of these tradeoffs is a design choice being tested on live customer money rather than a conclusion backed by a track record.
The comparison also underlines something this site has flagged before: agent-native trading tooling is arriving faster than a shared vocabulary for describing what any given implementation actually permits. "Agentic trading" currently describes products ranging from an LLM that can only answer questions to one with direct order-execution rights inside a ring-fenced sub-account to one with essentially full account access mediated only by a confirmation click. Customers comparing these products by marketing copy alone have very little basis for judging which one is actually the more conservative choice.
Read the original: PYMNTS - Retail brokerages converge on agent trading but diverge sharply on guardrails: confirmation-required, sub-account budgets, and advisory-only sit side by side. Commentary is the independent editorial view of Share Trading; the original article is credited to its publisher.