Will Generative AI Replace Developers?

Developers continue to work against shrinking time to market demands, and the latest tool in the toolbox is generative AI (GenAI). From code generation to documentation and product marketing, the technology is making positive productivity impacts, albeit less than the current hype suggests. For example, global agricultural and construction company CNH Industrial is achieving a net developer productivity gain of 5% after debugging and security scans, though the benefits don’t end there. The company is using generative AI in several ways.

At present, CNH has about 2,000 developers, 10% of which are actively using GenAI. The company launched its first customer-facing application for dealer technicians that allows them to query how to fix vehicles using a cell phone and app as opposed to sifting through 500-page manuals and PDF files. 

“We were able to go from talking to a dealer to a prototype in 30 days, from prototype to pilot in 75 days and from pilot to launch in 60 days, so all in all it was five and a half months from idea to the first iteration of the product,” says Marc Kermisch, global chief digital and innovation officer at CNH. 

Farmers are now able to track their vehicle fleet and the agronomy data produced by those vehicles. 

Kermisch’s team has been experimenting with GitHub Copilot since summer 2023. So far, the highest utilization has been around traditional web-based technologies, including C#, Java, HTML and SQL. More specifically, they’re using it for test case development, code commenting and repeatable procedure development. There’s less adoption and efficacy when it comes to embedded systems utilizing C and C++ code.

The team is also using Microsoft Copilot and testing Google Gemini for things like writing job descriptions, press releases and employee-facing information.

Code Understanding Versus Code Generation

Aleksey Didik, head of the engineering excellence program at EPAM Systems a global provider of digital engineering, cloud and AI-enabled transformation services, used generative AI to create a system in Python without knowing Python. (He’s actually a Java developer.)

“I wouldn't be able to do it without generative AI in the short term. It all depends on applicability, it all depends on the system,” says Didik. “When there is a lot of legacy code, the code base itself is very specific to the company, so the results are not as magical or immediate, but still, generative AI is a very powerful technology. If the latest updates to Gemini Pro, introduces 1 million tokens, it will increase the context the LLM can use to understand how to generate proper code.”

EPAM has been using generative AI since it was first introduced and they’re seeing productivity improvements at the individual level, but not so much at the team level. In fact, the company recently published a white paper, based on a client engagement, that explains what organizations should do to succeed with generative AI.

“If you're talking to engineers, the individual productivity and individual performance, yes, there is a boost, but how it converted to the team performance, product performance, company performance, and business performance is much more difficult,” Didik says. “Generative AI can generate elements of code, but you still need to have a whole SDLC in place to verify the correctness of the solution. We still need to verify it against the bigger scope of requirements against a more comprehensive set of user stories, for example.”

Read the full article here.

Download the white paper ‘The Complex Path of Generative AI Adoption in Software Development.’