News

Justin Hertzberg is the CEO of FPFX Tech, PropAccount and BullRush, leading prop firm technology and white-label solutions.

getty​Ask most business operators about AI, and they will describe a chatbot that answers questions faster than a person could. The technology gaining ground now does not just answer questions. It can build a campaign, publish it, watch what happens next and adjust before anyone asks it to.​

That shift shows up clearly in prop trading, where firms already manage a high volume of repetitive decisions across marketing, support, evaluations, risk and payouts. But the real change has little to do with trading. It is about who, or what, actually does the work.

What AI Actually MeansA chatbot generates a response and stops there. An agent pursues an outcome instead. Give it a goal, a set of tools and permission to use them and it will plan a sequence of steps, carry them out across whatever systems it needs and adjust as needed.​

That distinction matters because it changes where the technology can safely be used. A chatbot can sit at the edge of a business, answering questions, without touching anything that matters if it gets something wrong. An agent acts inside the business, and the actions it takes are real from the moment it takes them.​

That capability was mostly theoretical two years ago. It is not anymore. The tooling needed to deploy an agent reliably inside a real business has improved enough that narrowly scoped autonomous work is now a practical use case.

From AI Tools To AI WorkersThe first useful agents will handle the work around the biggest decisions. Marketing is a natural starting point across almost every industry.​

An agent can already help produce content and publish it with the right permissions and oversight. From there, it can launch a promotion, coordinate it across channels and shift effort toward whatever is actually working.​

The same pattern shows up in other areas. An e-commerce agent can watch inventory and trigger a promotion. A support agent can resolve an issue across billing and a CRM. A financial services agent can gather documents, run checks and route anything unusual to a person.

The industry changes. The structure does not: an objective, access to the right systems, boundaries on what it can do alone and a way to escalate whatever it cannot safely decide.​

Prop trading is a useful case study because the business is almost entirely operational: attracting traders, support, evaluations, account monitoring and payouts, often across several integrated systems. Agents are already moving into that work roughly in that order: support first and payouts last, tracking what a mistake actually costs.

Matching Autonomy To ConsequenceThe useful way to think about an agent is not only by what it can do but by how easily a mistake can be noticed and corrected before it spreads.​

A marketing agent that publishes the wrong headline is simple to fix: someone pulls the post. Other kinds of work call for more caution before autonomy increases, simply because the actions involved take longer to unwind.​

That is why adoption has moved slowly; it started with lower-stakes work and will only extend toward higher-stakes decisions as trust, oversight and visibility improve.​

An agent generally acts with the same confidence whether or not it has read a situation correctly, which is exactly why the boundaries and monitoring placed around it matter as much as the model behind it.​

None of this necessarily means fewer people. It changes where their attention goes: checking exceptions instead of every payout, setting objectives instead of writing every campaign. The scarce resource stops being the execution.

What To Ask Before Giving AI AutonomyThe questions become: What is the agent actually allowed to do? What happens when it is uncertain: Does it stop, escalate or guess? What requires human approval?​

A system that answers a question well but still needs five manual steps to finish the job is not useful. It is not automating the workflow.

The Architecture QuestionAn AI agent faces many of the same limitations a prop firm does today—it’s only as good as the systems it can act through.​

AI tends to work best when the technology around it is well connected. That does not mean every tool needs to live under one roof. Systems that share information easily can support an agent just as well as a single unified platform, and different businesses are likely to take different paths depending on what already works for them.​

The businesses that will benefit most from this shift are unlikely to be defined by any single product or demo. They are more likely to be the ones that think about how their technology works together as a whole. ​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?