Discussion

Prop-trading firms may be a test bed for AI agents, says industry executive

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#4714

AI agents could take on more of the routine work at prop-trading firms, but Justin Hertzberg argues that the order of adoption should depend on how easily mistakes can be undone. His example offers a useful rule for businesses eyeing agents: start with reversible tasks, and keep tighter controls around the costly ones.

Watch Desk analysis

What happened

Hertzberg, chief executive of FPFX Tech, describes how agents might handle work across marketing, customer support, evaluations, account monitoring and payouts. Unlike a chatbot that answers and stops, an agent can use tools to pursue a goal across several systems, he writes. That makes its permissions and ability to escalate uncertainty part of the job, not optional polish.

In his Forbes Councils essay, Hertzberg says adoption is moving from support towards payouts. He argues that marketing is a natural early use: a wrongly published post can be pulled. A mistaken payout is harder to unwind, so he says higher-stakes work should wait for stronger oversight and visibility.

Why it matters

The example is about prop trading, but the test travels: what can an agent do, how quickly would anyone notice a mistake, and how difficult would it be to reverse? A system that connects to more business tools can finish more work, but it can also turn a bad decision into a real-world action.

Hertzberg’s essay is an industry executive’s argument, not evidence that these deployments are widespread or that agents are already handling the listed jobs at scale. Still, it puts the right operational question ahead of the usual demo-room marvel: not just whether an agent can act, but what happens when it acts wrongly.

Our read

Start with the task you can safely undo, not the one that makes the slickest presentation. Give agents narrow permissions, make escalation explicit and keep a person close to decisions with real financial consequences. Autonomy is less impressive when the clean-up costs more than the work it saved.

What to watch

  • Whether prop-trading firms disclose which tasks agents actually handle.
  • What approval and escalation rules apply to evaluations, account monitoring and payouts.
  • Whether the systems connecting an agent to business tools make its actions easy to review and reverse.

Discussion spark: Should businesses let AI agents handle low-stakes tasks first and earn wider permissions, or is that a false sense of safety when the same systems may later control consequential decisions?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.