What AI Changes for Broker Technology 

Key takeaways  

• AI coding tools have cut the cost of building software, yet the systems that hold client money and answer to regulators still belong with specialist vendors.
• License the regulated core: the trading platform, CRM, liquidity, risk, and payments. Reserve in-house building for the smaller pieces regulation does not reach.  
• Two sourcing models dominate: one all-in-one platform from a single supplier, or a modular stack of specialists connected through APIs. The right choice tracks the broker’s stage.  
• AI belongs in the layer around the licensed core, where APIs decide what an agent can read and what it can change.  
• Trustworthy AI rests on two safeguards: a model grounded in the broker’s own data and a human in the loop for any decision with financial or legal weight. Moody’s research and the EU AI Act point in the same direction.

AI coding assistants have significantly changed the capabilities of a small development team. Over the decade before AI emerged, the question of whether to buy or build from scratch had almost faded away. 

Now, a founder can describe a feature, and working code will appear within minutes. For brokers under pressure to stand out from competitors, building technology in-house looks affordable, and the temptation to try it has grown. 

But this temptation deserves a hard look, because brokerage software handles client funds and is subject to regulatory oversight. 

So the question shifts from whether to build a trading platform to which parts brokers can build in-house, the smaller pieces that regulation does not reach, and which to license from specialist financial software firms. 

The pieces a broker assembles 

A working brokerage runs on several distinct systems. Clients trade through a front-end platform. A CRM tracks onboarding and the relationships that follow. A liquidity layer connects the broker to execution for A-book or B-book flow. Risk and analytics tools monitor client activity and flag anything that looks off. Payment connections move money in and out, and they cause more day-to-day trouble than any other piece. 

Most of these carry regulatory weight, sit close to client money, or both. That puts them on the license side of the line. Brokers license these rather than build them, so the open choice is which supplier, or how many. 

One supplier or several 

Two models dominate: a broker can take an all-in-one platform from a single supplier, or assemble a stack from specialists and connect the parts through APIs. 

The all-in-one route is the fastest way from nothing to live, but its downside arrives later in the form of vendor lock-in. You get the features your supplier chooses to build, on their schedule, and switching grows expensive once the business leans on the whole bundle. 

The modular route spreads the work across firms that each do one thing well: a trading platform from one, a CRM from another, liquidity from a third, risk analytics from a fourth, and so on. When a new requirement lands, you add a component instead of replacing the system. Connecting these pieces is easier than it was ten years ago, because the vendors have built integrations with one another and expect to run side by side. 

“The industry has finally passed the debate about whether modular wins. I think it does.”—Aeby Samuel, Founder and CEO of FYNXT

Which model fits depends on the broker’s stage. A firm starting out with no engineering team and a tight budget may want a single supplier to handle everything from the website to liquidity. As volume and client types multiply, that single setup strains, and brokers reach for stronger components. We make the case for custom development over an in-house build in more detail elsewhere. 

Where AI fits, and where it doesn’t 

This is where the smaller in-house pieces come back into view. Building a competitor to a mature trading platform with an AI assistant is not on the table; the support burden and the reliability bar put it out of reach. What AI suits is the layer around the licensed core. 

“It’s very easy to make something work for 90% of the scenarios, but that’s not what you need in a brokerage or a prop firm. You need something that’s going to work 100% of the time.”—Tom Higgins, Founder and CEO of Gold-i 

What AI suits is the layer around the licensed core, and the harder part is deciding what that layer should do before building it. 

In a modular stack, APIs decide what a brokerage can connect to and what it can automate. An AI agent earns its place only when it can read from and act on the systems beneath it, and that access runs through those APIs. A risk manager can ask, in plain language, how a symbol routes to the market and get an answer drawn from the broker’s own configuration. A dealer can describe a change to spreads or liquidity, and it is applied without having to click through a settings screen. 

Jon Light said prospects now lead with a question that “nobody would have even asked” a couple of years ago: can they control the platform through AI? DXtrade answers it with a trading control API that brokers have connected to AI assistants through MCP servers. 

The limits matter as much as the uses. A model trained to please will answer confidently even when it lacks the facts, and a confident wrong answer about routing or risk costs money. Two safeguards hold this in check: ground the model in the broker’s own data and documentation, so its answers rest on the configuration in front of it rather than on whatever it absorbed from the internet. Then keep a person in the loop wherever a decision carries financial or legal weight. AI shortens the work but doesn’t move the responsibility. 

“You have to train it with your own data. You have to put in the guard rails, and wherever there are financial or legal implications, you should have a human in the loop.”—Aeby Samuel, Founder and CEO of FYNXT 

Industry research and regulation point in the same direction. Moody’s global survey on AI in risk and compliance found that 84% of respondents agree AI offers significant advantages, yet only 30% see those benefits clearly in practice. The same survey reports that 42% treat human oversight as mandatory rather than optional. A 2026 systematic review in Entropy is more pointed about finance, where the regulatory and fiduciary weight of the work demands close human oversight of AI outputs. The EU AI Act puts this into law. It requires human oversight for high-risk AI systems, a category that reaches into parts of financial services.  

Watch the full conversation 

In a recent webinar, Jon Light, Senior Director of Product Management at Devexperts, put this question to two people whose companies build core parts of broker infrastructure: Tom Higgins, Founder and CEO of Gold-i, and Aeby Samuel, Founder and CEO of FYNXT. The discussion covered the main components of a modern broker technology setup, the trade-offs between single-vendor and multi-vendor models, how brokers can move away from a closed platform without disrupting a running business, and where broker technology may be heading over the next five years.

Watch the recording below or on Devexperts’ YouTube channel and book a call with us to discuss the update to your platform setup.