How generative artificial intelligence can make engineers more efficient

This new field is enormously interesting and virtually everyone I know has used it for fun and interest. But we are now shifting to the practical applications and how we might use this technology to transform the way we work and better serve our customers.

" Artificial intelligence has hit the headlines thanks to the text generation abilities of ChatGPT. But it may also help engineers free up time by producing software code"

There’s many applications of AI for knowledge workers. ChatGPT contains 175 billion parameters, making it one of the largest and most powerful models for AI processing available. It also reached 100 million monthly active users in January, just two months after it launched, making it the fastest-growing consumer application in history, according to a UBS.

But what are the tangible benefits for the technology within financial services and how are we pursuing its use inside ANZ? With an organisation of some 4,000 software engineers, there are many avenues to improve the efficiency, reliability and performance of our code assets.

While many will have used ChatGPT to generate English text, it’s also remarkably powerful at producing software code snippets. Software is written in various programming languages and, like natural language, it follows a set of grammatical rules.

And just like the written word, there’s an enormous body of example code taken from the public domain that has trained the ChatGPT model to recognise and generate software.

These examples act like ‘over-the-shoulder’ assistants, a peer programmer watching you write code and providing real-time suggestions and feedback on the code being written. This allows engineers to lay out their code and build robust outlines more quickly, use the best of common techniques and help identify bugs early. This means engineers can spend more time solving complex problems and less time on repetitive coding tasks.

Digesting complex information

ChatGPT’s conversational AI also helps software engineers deepen their comprehension. The tool can provide answers to common questions and help engineers find and understand complex technical information. ChatGPT can summarise large amounts of technical information into more digestible chunks – and it can also take an engineer’s code and create easy-to-read documentation.

We’ve formally sanctioned one team in ANZ to explore generative AI for software engineering to improve our testing. Writing test cases – normally also written in code – is an important, if somewhat tiresome, task.

Generative AI is really powerful at reading code and proposing the complete test case to properly exercise that code. This will dramatically improve unit testing and, I think, give us a step change in identifying bugs and errors earlier in the development process.

We are also looking to improve how we validate code passed strict standards and rules around software code. We have comprehensive policies at ANZ but it still requires a human to read and understand a policy and recall the key elements while completing a code review.

We think there is enormous potential to use generative AI to complete that interpretation for us – read our policies, read a recent code addition and advise where the non-compliant aspects are.  Of course, it’s extremely controlled right now.

Efficient engineers

We are building up our positions as we explore this. Primarily, we see this as an augmentation tool, not a final arbiter. All our code today relies on senior engineers to approve the completeness and compliance before it is pulled into the main repositories. While that responsibility remains unchanged, we believe we’ll still be able to make our engineers far more efficient through the use of AI technology.

All new technologies present challenges and opportunities and this will be true of using generative AI in software engineering. There will be concerns around reliability, repeatability and risks of these tools. There will need to be checks and balances for some time yet.  We are not yet ready to use this outside of centrally managed, executive approved experiments.

However, ANZ has a long history of embracing technology early in the adoption life cycle. We launched our first website in 1996; internet banking in 1999; mobile banking in 2008 and, in 2016, we were the first bank to release Apple Pay.

We have more than a third of our application estate in the cloud, we run critical workload in large container architectures and we’re embracing modern data technologies. ANZ is, and always will be, an early adopter of critical new technologies and generative AI is the next major wave.

But we will do it with rigour and discipline. 

Tim Hogarth is Chief Technology Officer at ANZ.