What happens when genAI meets low code?

When the programming language Cobol was invented in 1959, the US Department of Defense was trying to make code simpler. The Common Business-Oriented Language, like its immediate predecessor Flow-Matic, used English language prose to make the code more accessible for everyday users.  

Nearly 70 years later, every subsequent advancement in code has sought the same end: Make code simpler. 

In an industry full of hype cycles, generative AI, large language models (LLMs), and copilots are currently in the spotlight. But it wasn’t long ago that everyone was talking about low-code and no-code development practices, which promised to simplify software development, increase efficiency, and enable an army of citizen developers.

The buzz surrounding these forms of programming has died down, even as they’ve found various use cases. One could wonder, though, if the advent of generative AI—and its unique ability to translate old code into more popular modern languages, like C++, Java, and Python—will hurt the low-code movement. 

I think the two—low code and AI together—are potentially bigger than the sum of their parts

Jon Butler, Velox

Jon Butler, cofounder and CEO of financial technology development firm Velox, believes, to the contrary, commingling these technologies could finally bring the sought-after efficiency for which programmers have been waiting nearly 70 years. 

“I think the two—low code and AI together—are potentially bigger than the sum of their parts,” says Butler, who spent his career managing developers at Goldman Sachs before leaving the bank to found his startup in 2018. The aim of Velox is to stitch together different trading systems by piggybacking off of the desktop app interoperability movement.

Talent eats up a large portion of a bank’s budget. But new technology can help to either reduce staff, or at least make workers more efficient so they can prioritize higher-value endeavors. Goldman Sachs, for example, has a tech staff of around 48,000 people, more than 10,000 of whom are developers. In a recent interview with CNBC, Goldman chief information officer Marco Argenti said as much as 40% of the firm’s code has been able to be written automatically by developers using generative AI.

When Butler began writing code in school, he learned a low-level assembly programming language intended to communicate directly with a computer’s hardware. But when coding with assembly, one had to write far more lines of code than with   Python and Java. “When we say low-code, we’re really just talking about developer acceleration and simplification,” Butler says.

Low-code has yet to reach the type of mainstream awareness that genAI has garnered over the last year. Yet the mission of the technology, which uses premade, Lego-style building blocks to allow developers to build custom applications, has been supercharged by developments in generative AI, creating opportunities to mitigate risk and eliminate technical debt challenges. 

Recently, the addition of copilots—chatbot-like tools that combine LLMs and a user’s own data to augment their workflows—into low-code products has demonstrated the ability of generative AI to solve decades-long problems in programming, such as translating legacy code or streamlining semi-automated systems created across pages of Excel spreadsheets. 

Microsoft, one of the principal backers of OpenAI, the owner of ChatGPT, and the creator of Microsoft and GitHub copilots, has its own low-code offering, Microsoft Power Platform. Demonstrates that the tech behemoth still sees value in allocating resources to its low-code offering amid the genAI craze.

“AI is probably the number one topic I get asked about, and I imagine many of my colleagues at Microsoft would say the same,” says Rob Smithson, UK business applications lead at Microsoft. “What you are already seeing—and what you see much more of—is copilot functionality being added to a variety of our products.” In December, Microsoft announced a number of role-specific copilots, including one for Power Platform. “The wonderful thing about low-code is it increases the accessibility of those technologies to a much broader sector of a business,” he says. 

At a product demo in December, Smithson—who does not have a software development background—was able to design an app for a supermarket by asking Power Platform to build an app that could manage spill cleanups in a grocery store. Using natural language prompts, Smithson was able to bypass typical coding directions generated by a copilot and allow the low-code platform to design him a suitable app using genAI informed by his prompts.

Leaving a legacy

Bank IT departments have been under stress from increasing pressure on margins and regulatory compliance since the 2008 financial crisis. To manage cost, the problem of efficiency in coding is even more critical, says Tej Sidhu, chief technology officer at Genesis, a low-code tech provider backed by BNY Mellon, Citi, and Bank of America, among others, that focuses specifically on capital markets software. Genesis is also the main technology partner behind Octaura, a multi-bank platform for collateralized loan obligation trading.

Low-code technology, Sidhu says, is in a unique position to help banks utilize generative AI in compliance with security regulations. Much of the basic plumbing of financial markets applications—such as data streaming, authentication, entitlements, and audit tools—are pre-built into the Genesis platform and were designed in compliance with existing regulations. 

“Banks first had to ask whether they could allow AI at all. They had to consider whether their information could leave their bank and go to OpenAI servers and the open internet. That took a long time to overcome. We’re now seeing that moving pretty rapidly,” Sidhu says.  

With technology, there is the classic curve: there is huge promise, then a huge hype, and then the business of making it work. That’s where we are now.

Tej Sidhu, Genesis

Genesis, which has built primary bond issuance platforms for Tier 1 banks, has embedded AI functions that allow users to extract bond market data from disparate sources, often in unstructured formats like emails and instant messages. Not only can low-code provide regulatory guardrails, but the ability of generative AI to synthesize large amounts of language—either natural language or programming languages—can help address the problems of legacy tech.  

Thierry Bonfante, chief product officer at low-code provider Unqork, says the biggest business opportunity for companies wishing to use generative AI is finally tackling the legacy migration issue. At banks, more than 40 years of code from Cobol, Fortran, Assembler, MS Access, MS-DOS—“you name it”—exist within capital markets systems. “GenAI, for the first time, provides a path to change that,” Bonfante says.

Often, banks’ IT departments do not know what function an application may perform, making it risky to intervene in the underlying software, Bonfante says. 

Unqork has built a solution where users can copy a legacy application’s code into the platform’s genAI module. In less than a minute, the platform replicates not only an application’s original logic, or the action it was first meant to perform, but also generates a task to modernize it based on Unqork’s pre-built modules.

“Every CIO was looking at their legacy systems, knowing it was a time bomb, quite frankly. They were basically hoping—and doing a bit of a gamble—that it would not explode under their watch,” Bonfante says, and it has been going on since Y2K. 

Talent crunch

In finance, recruiting top tech talent is harder, says Genesis’s Sidhu, as banks and other financial firms compete with social media companies, tech giants, and start-ups.  

Sidhu began his career as a programmer after graduating college, though he studied biology and considers himself a product of a liberal arts education. Over years of building financial technology, he rose to managing teams of developers himself. He says that for those in the capital markets, a domain knowledge of finance is what sets apart the most promising young developers from the pack. 

When low-code was earning headlines, the conversation often revolved around the technology replacing the developer, Velox’s Butler says. Now, the conversation has switched to fretting over AI replacing developers. “Neither of which are true,” he says. 

The recent attention on AI copilots, such Microsoft’s offering, have brought further attention to the problem of efficiency in software development, where it is famously difficult to predict how long a piece of software will take to build or how much the project will cost. This is because programming is a deeply human undertaking, Butler says.  

Generative AI is now being implemented by a growing number of institutions as the technology passes security scrutiny. “It’s like children doing homework. Someone comes in and says, ‘I typed this line of code into ChatGPT, and look what it produced.’ Your first reaction is—especially if you are a lawyer at a bank—‘Oh, my God, you can’t do that!’ But you quickly realize that people are going to do it. You cannot stop that kind of trend,” says Sidhu, who notes that Genesis spent the back half of 2023 seeing how it could best integrate genAI into its product. 

One area where Genesis has incorporated AI is for developer training. The company integrated a chatbot that is built on top of OpenAI and is trained on how to use Genesis. A developer can ask the chatbot using natural language to write a trade screen or stream a certain set of data and the bot will show the developer how that can be done in Genesis. 

Tools that make coding easier for developers at banks are also a useful way to limit third-party risk. Sidhu says many clients use Genesis for end-user control of operational risk. “Regulators have identified the risks associated with end-user computing, and they are saying to either fix them, mitigate them, or exit that part of the business,” he says.

While some new technologies fail to live up to the hype, genAI and low-code can demonstrate the significance of simplifying programming and tackling the decades of old code each new development has rendered obsolete. “With technology, there is the classic curve: there is huge promise, then a huge hype, and then the business of making it work. That’s where we are now,” Sidhu says.