At first glance, artificial intelligence seems like a software developer's dream. A recent report from McKinsey & Company found that programmers can generate code up to 45% faster with the help of generative AI.
But if not used strategically, AI can be a developer's nightmare. said Edward Anderson Jr., professor of information, risk, and operations management and the Betty and Glenn Mortimer Centennial Professor of Business at Texas McCombs.
This problem arises when AI is used to write code that interacts with so-called legacy systems with outdated software, he explains. These environments are often full of shortcuts, quick-fixes, and other bad programming techniques.
The Information & Software Quality Consortium estimates that this technical debt costs U.S. companies $1.5 trillion in lost productivity and cybercrime.
It could even lead to a real-world meltdown. In 2022, Southwest Airlines' 20-year scheduling system crashed, stranding passengers on approximately 17,000 flights.
But Anderson warns that using AI carelessly to patch such systems risks making them worse. First, the AI is trained on the existing code, including any flaws. Therefore, they tend to incur more technical debt per line of code than trained and experienced human software engineers.
How can companies avoid these problems? Anderson, along with Jeffrey Parker of Dartmouth College and Burke Tan of the University of New Mexico, interviewed dozens of programmers from a variety of industries. He offers some best practices for AI-assisted software development.
Make technical debt an engineering priority.
As companies rush to market, resolving technical debt is often a low priority, Anderson says. They're kicking cans down the street, just like Southwest Airlines did.
Rather than fixing something when it breaks, companies should build a thorough review of technical debt into developers' daily workflows, especially when using AI to make repairs.
“This is about organizational processes,” he says. “If you intend to use AI and potentially increase the rate of technical debt generation, you should allocate more time to decommissioning.”
Create clear guidelines for AI-assisted coding.
While executive policies may address the use of AI in general terms, protocols in day-to-day software development are still evolving, Anderson says. To aid development, software teams need to document when they use AI and why.
In addition to clearly defining tasks, humans need to be kept in the loop, he added. “We need to make sure we have people with enough training and experience in software engineering to catch the AI when it makes a mistake.”
Train developers on the dangers of AI coding.
As the number of experienced developers dwindles due to retirement, they can be replaced by inexperienced programmers who deploy AI tools unchecked, especially in legacy environments.
In situations like this, Anderson says, knowledge transfer is critical. Performance goals for senior programmers should include formal coaching. This means not only reviewing the code written by junior developers, but also training them to use AI effectively and responsibly.
“Let me be clear: I think AI is a productivity driver,” Anderson says. “You just have to use it judiciously, and you have to give software engineers time to do it.”
“The Hidden Cost of Coding with Generative AI” is published in MIT Sloan Management Review.
/Open to the public. This material from the original organization/author may be of a contemporary nature and has been edited for clarity, style, and length. Mirage.News does not take any institutional position or position, and all views, positions, and conclusions expressed herein are those of the authors alone. Read the full text here.
