Generative AI will augment role of software developers, says IBM's Ritika Gunnar

Generative artificial intelligence (GenAI) has become the top priority for enterprises across the globe as they seek to harness this technology not just for productivity gains, but to also create newer models for businesses to operate. This would not mean replacing the existing technology systems or people, but augment through GenAI.

IBM has been at the forefront of this, and believes, for example, in the case of several older computer languages which are still operational but skills are fewer, GenAI is playing a key role in refactoring these legacy systems into more modern applications.

At the same time, IBM says GenAI will only augment the skills of software developers and will not replace them as the human element is critical for any further development.

In an interview with Enterprise Story, Ritika Gunnar, General Manager, Data and AI, IBM; and Vishal Chahal, VP, IBM India Software Lab, highlighted the broad changes that GenAI will usher in.

Edited excerpts from the interview:

Enterprise Story (ES): What kind of changes is GenAI bringing into the world of software development?

Ritika Gunnar (RG): If one looks at the software development area, I think generative AI (GenAI) is going to be used to improve productivity in multiple ways. GenAI is being used to understand code as a starting point. There are billions and trillions of lines of code that exist in different languages, and sometimes those skills don’t exist so often in the space of code. So, the first step is understanding. The second is, even as you’re writing code, what can you do to document. These legacy languages are so important and critical that they've been there for decades.

The ability to understand those languages (older computer languages) as skills has become a lot more scarce. However, the ability to use generative AI to modernise these languages is really important and these are some of the capabilities that IBM possesses.

The last area is, what do you do to actually debug and use generative AI for debugging. So, when you look at the whole life cycle of development--what it means to actually understand, document, to create, to convert, and modernise. In this whole life cycle, we are starting to see the impact of GenAI at every single point.

ES: What will be the role of software developers now given the advancements made by GenAI?

RG: I think you always need to look at it as augmenting the human and not replacing the human. From our perspective, the way we want our developers to use the technology is as an assistant, because even in itself, like the generative AI that’s driving code, the accuracy rates are still not as you can optimise your code as well, but you want to ensure that you still have a human in the loop to really understand what’s going on. So, a lot of the technology that we are using for code generation is really augmenting the human themselves. And that’s very critical for us.

ES: What are the benefits that enterprises are deriving from the implementation of GenAI?

RG: The first use cases are of productivity gains and efficiency. I think we are actually at this cusp. I call it the second wave. And this second wave is really transferring from what it means to have a number of flowers blooming of generative AI projects, to what does it really mean to get into production. In addition to that, we’ve also applied the other kind of components, which I think are essential to cross the chasm from pilots to production. And that includes, how do you have trust and transparency into how AI is operating? How do you make sure the AI itself is as efficient as smaller models being sometimes better than the larger models. Bigger is not always better. So, all of these things become really essential to cross that chasm from pilots to production.

Vishal Chahal (VC): When you take it from pilot to production, there’s going to be a change in the cycle the way it runs today. For developers, DevOps and other things are going to be automated.

ES: Are there any particular industries that are early adopters of GenAI?

RG: I think almost every industry is adopting GenAI. There’s not an industry that isn't experimenting. If you look at it, what's probably most common is the horizontals in which they're actually applying. The customer care piece is probably one of the first, easiest, and low-risk ones that we see across every organisation and every industry. The second, when it comes to employee productivity, the digital labour piece is probably the second area where we see it being most adopted, and the third being for the developers themselves. The productivity that developers are going to receive is really important. So, these are the three primary areas, and I actually see them across every industry.

VC: As a developer, you’re not just coding but essentially augmenting your code. You’re giving instructions that this is what I want to write. It gives you some piece of thing, but you still have to integrate one piece of code that's given with the other. That's not an easy thing to do, like you need to really be able to do that. That's where the documentation and understanding of that comes into picture.

ES: What future trends do you see in GenAI?

RG: One, we’re working on research right now that I think is closer to the use of agents. Today, when you ask a large language model to generate a piece of code as an example, it just gives you a straight in line response of how it generates code. But as you see agents being used, you can now use a model to not only plan for what kind of code you want to write, it can generate the code, and you can critique that piece of code and optimise it. You can now have a model even debug that piece of code. This multi-step piece of using agents in code is what is really imminent.

VC: Agents will become the middleware of GenAI. You still need somebody to do the plumbing, but we are in a low code era, so agents will do the plumbing for you. But the value of your data pipeline doesn't go away if you do not have a way to have a governed, explainable data pipeline itself, with traceability all the way to the source, be it the model or the code that you're going to use.