Isabelle Bousquette
Efforts to automate parts of the coding process, which can be tedious and time-consuming, have been under way for years. They received a boost thanks to the growing scale and accuracy of generative AI models.
Generative AI coding tools promise huge efficiency gains for developers, but some tech leaders fear the consequences of spawning too much code too fast.
IT leaders at United Airlines, Johnson & Johnson, Visa, Cardinal Health, Goldman Sachs and other companies say they are excited about generative AI’s potential to automate certain parts of the code-writing process and expect it to result in significant productivity gains.
However, some IT executives say that lowering the barrier for code creation could also result in growing levels of complexity, technical debt and confusion as they try to manage a ballooning pile of software. “Technical debt” is a broad term describing the expected future costs for applying quick-fix solutions.
“The potential for increased technical debt and orphan code is always a concern when delivery can be accelerated,” said Tracy Daniels, chief data officer at financial-services company Truist.
“People have talked about technical debt for a long time, and now we have a brand new credit card here that is going to allow us to accumulate technical debt in ways we were never able to do before,” said Armando Solar-Lezama, a professor at the Massachusetts Institute of Technology’s Computer Science & Artificial Intelligence Laboratory. “I think there is a risk of accumulating lots of very shoddy code written by a machine,” he said, adding that companies will have to rethink methodologies around how they can work in tandem with the new tools’ capabilities to avoid that.
Efforts to automate parts of the coding process, which can be tedious and time-consuming, have been under way for years, said Solar-Lezama. They received a boost thanks to the growing scale and accuracy of generative AI models, which also contributed to the popularity boom of ChatGPT, he said.
A shortage in the talent pool of available developers is also pushing companies to invest more in tools that can aid the process, said Amanda Silver, corporate vice president and head of product for Microsoft’s developer division.
Different enterprises are at different points of evaluating and deploying tools such as Github’s Copilot, owned by Microsoft, and as well as other tools from Amazon, International Business Machines and startups such as Tabnine and Magic AI. These tools typically work by suggesting new code snippets and tests, and providing technical recommendations inside of the code-writing programs developers already use. But there are also risks, IT leaders say.
“I think it makes the CIO’s task a lot more complex, even as it makes the programmer’s task easier,” said Vivek Jetley, executive vice president and head of analytics at EXL, a data analytics and digital operations and solutions company.
These tools have the capacity to democratize code-writing, meaning more and more employees could start writing code for several new use cases. As the amount of code explodes, the CIO will need to work to control and govern that code and prioritize what to keep, what to junk and how to run the system, he said.
“There’s going to be more confusion for sure,” said Jetley.
To be sure, technical debt and orphan code have long been challenges that have plagued CIOs, according to OutSystems Chief Executive Paulo Rosado. As more and more code is built, there’s naturally confusion that comes with understanding what certain code does and how it was created, he said.
As developers leave companies, that confusion intensifies and as time goes on a growing pile of code becomes more and more difficult to keep up to date, he added. Rosado said he does expect these issues to be aggravated by generative AI coding tools.
Technology leaders should be careful not to equate accelerated delivery of code with productivity, said Sanjay Srivastava, chief digital strategist for professional services firm Genpact. Businesses should be thinking more about return on investment than actual amount of code being written, and should weigh the economic cost of running power-intensive generative AI tools.
United Airlines chief information officer Jason Birnbaum said these risks are valid, which will make it important to engineer cloud environments for security and resiliency and make it harder to release software that hasn’t been properly vetted and tested.
Despite risks, CIOs are pushing forward. Birnbaum said United is testing several generative AI applications, including code generation. Cardinal Health recently established a cross-functional task force to evaluate use cases and risks. Truist is exploring net-new code generation and code annotation with a vendor. And Goldman Sachs is already seeing double digit efficiency gains in early pilots.
No comments:
Post a Comment