Vibe Coding Lands in FX as ING Dials-up AI Use

For the less technology-savvy, the ban on indoor smoking was probably the most significant vibe coding event in FX markets in the past two decades. Until now.

Dutch Bank ING has expanded its use of AI and started to experiment with so-called vibe coding, or the practice of prompting machines to write code and create complex systems following prompts from humans using natural language instead of using specialist knowledge. A bit like Star Trek, but for trading.

The rapidly-spreading practice is far from just hype, it has the potential to disrupt established technology providers and their relationships with banks, who might soon be able to do much of what third-party firms offer for a lot less. “I think this is a big risk to providers. Banks are always looking to reduce costs and the savings add up to quite a lot,” Simon Bevan, global head of e-trading at ING tells The Full FX.

ING has so far found two main use cases for vibe coding: on the trading side, where the quant desk can whip up a code and create a testing environment for it within hours, instead of days or weeks, without putting a demand on the time of specialist, senior and hugely expensive developers. The second, is in visualisation. 

“At this point, our quant desk is pretty much using vibe coding on a day-to-day basis,” Bevan says, adding that so far deployment has only happened under strict supervision with senior team members supervising output. But progress is rapid and “vibing” is already promising to save big bucks in costs for users.

Bevan notes that FX needs the best of the best developers because their ability to produce incredibly low latency, super-reliable code that can handle huge amounts of data can make or break a bank: it’s making big decisions, taking on big risks for the bank so it needs to be extremely reliable. This is not cheap. As more junior staff are able to match this level, senior developers are able to focus on more creative efforts.

“I was surprised by how good the results were,” he says. “One of the first use cases we found was creating the code and then it would create the testing framework for it. Instead of taking multiple days or weeks suddenly you can do it in hours. Suddenly, all these amazing developers we have are freed up to do more valuable work. 

“It’s similar to how AI is used to free up time for humans to do what humans are good at, the creative stuff,” he adds.

Another real use case is in visualisation. Turning market data into a visually compelling product has required UX specialists and third-party systems, but vibing has made it possible to do this easily and quickly in-house. And while hallucinations and other flaws can be risky on the trading side, the risk-reward with visualisation is very attractive, Bevan says.

“We are going into an experiment where we are thinking about whether we can replace our expensive visualisation tools with our own in-house builds,” he explains. “That is a real potential cost saving but it’s more than that: because it’s in-house we can make it very bespoke so it’s not just cost saving, we can improve what we do and we don’t need to hire very expensive UX guys, we can just do it with more junior people. It’s not perfect yet, but it won’t take much tweaking.” 

Using an in-house anthropic model means that no data leaves ING, alleviating concerns about privacy. And applications are rapidly emerging: ING financial markets started experimenting with vibe coding only in February and after a short period of trials its potential quickly crystallised, with use cases snowballing. 

Progress is so rapid that Bevan has warned that rules and regulations need to catch-up: while there are frameworks for client- facing AI use, this doesn’t yet exist for trading and other areas. And supervision is still key, especially on the trading side. 

But given its potential to save on costs and improve the bank’s offering, vibing is unlikely to go away. Bevan said applications on the sales and trading side could see teams using a tool that can generate bespoke reports about clients, their trading patterns and hit ratios that are fit for client presentations with just a type prompt. 

“The speed it’s going at is crazy. Week on week we see progress. I think within a year it will be embedded everywhere and I think every bank will have this but they’ll work at different speeds. Any bank that doesn’t get on top of this will get left behind,” Bevan says.