AI-generated code may sound like a career-killer, yet a growing number of IT leaders are now turning to AI to generate various types of routine code, freeing human coders to focus on more complex and rewarding tasks.
AI can significantly accelerate software development, says Prasad Sankaran, president, software and platform engineering, at IT consulting firm Cognizant. “Based on natural-language prompts, AI can write code and test scripts across a range of programming languages, frameworks, libraries, and more,” he states in an online interview. Sankaran notes that AI can also generate synthetic test data, limiting the need to use sensitive live data sets. “At Cognizant, 20% to 30% of our code is now written by machines — that’s productivity we can pass along to our clients.”
Multiple Benefits
AI-generated code’s benefits are clear: more output, better quality in shorter timeframes, and an improved ability to navigate talent constraints, Sankaran says. “All of that translates into business productivity, agility, and effectiveness.” He believes that AI-enabled coding allows team members to work faster and with greater focus on higher value, creative activities such as feature relevance, user experience, security, resilience, and performance.
Enterprises with large development organizations should expect substantial gains in productivity, says David Menninger, executive director at ISG Software Research. “Organizations that rely on service providers to create applications for them should expect to see rate reductions as a result of the increased efficiencies associated with AI-generated code,” he states in an online discussion.
Getting Started
As is the case with many new technologies, it’s best to start simple, Menninger recommends. He suggests beginning with a prototyping exercise. “In this scenario, the quality of the code is less critical, and it will afford an opportunity to learn about AI’s capabilities.” Another good starting point is code conversion, since the final result can be compared against existing working software.
Devansh Agarwal, a machine learning engineer at Amazon Web Services, says AI is particularly adept at writing standard boilerplate code. “For example, if I want to create an API client for calling a particular API, AI can generate the boilerplate code for it,” he explains in an email interview. “If a developer writes this same code, it will take a few hours.” AI can generate the identical code almost instantly. “Since this is a very standard piece of code, which is already present on the Internet, it will most likely be correct.”
Getting started can be surprisingly easy, Agarwal says. “My mom, who has never written a line of code, managed to do it,” he states. “The process is simple: Decide what you want to build, prompt the LLM to generate the code, and keep asking it to fix errors until it finally works.”
Despite its relative simplicity, Sankaran stresses the need to train team members on the most effective use of AI coding assistants while dealing with its inherent risks and limitations.
Menninger recommends that AI-generated code should always be reviewed and tested — just like any other code. View AI in the same light as a junior developer. “You can give it an assignment, but you certainly want to check the results and probably add some refinements or ask it to make some revisions.”
Potential Roadblocks
Sankaran says enterprises need to recognize AI technology’s inherent limitations, including the risk of team members relying too heavily on AI-produced code. Meanwhile, generated outputs can be inconsistent and error prone, so there’s still a need for human oversight. A combination of predefined plans and templates, as well as automated enforcement of quality gates, baking in organization policies and guardrails, can help tackle this challenge by ensuring predictable, consistent, and compliant results within preset boundaries.
The biggest potential drawback is fostering a growing overreliance on AI-generated code, Menninger says. “There can be a tendency to trust the code simply because it has been generated by a machine,” he explains. “The good news is that you can also use AI to test code both for quality and security purposes so, hopefully, you can end up with even higher quality code than manually-generated and manually-tested code.”
Parting Thoughts
AI code assistants have come a long way and are continuously improving and delivering a superior developer experience, with more purpose-built and holistically-integrated solutions, driving automation and lead productivity across the software development lifecycle, Sankaran says.
Looking ahead, agentic AI has the potential to make software development easier, more autonomous, and orchestrated, Sankaran states. “For example, specialized agents for understanding requirements, writing code, reviewing code, testing, deployment and more, all working together in an autonomous manner under human oversight,” he says. “We’re moving closer to the day when we can simply describe the functionality we want … and have the system do the rest.”