In May, Microsoft CEO Satya Nadella gave his keynote for the company’s Build developer conference virtually. He wore a Jobsian black shirt and stood in front of thematically apt books. Accelerate, read one. Prosperity, another. After discussing the rapidity at which tech is becoming “embedded in every aspect of our lives,” Nadella unveiled a new component of Power Apps, a program that allows people to somewhat easily build apps for businesses (somewhat, because: Microsoft). Previously users had to know some basic code. Now they can write out what they want the app to do in plain language, the AI generates a few options, and they pick the best. Or as Nadella said, “The code writes itself.” 

His tone was buoyant, but others among us might proceed with caution. Coding comprises more than roughly 100,000 jobs in the Seattle area, and it’s supposed to be a sort of economic refuge as other vocations are automated away. A few years ago in Wired, tech writer Clive Thompson positioned it as “the next big blue-collar job.” Yet IBM recently showed that its own model could reduce the time needed to update code for a car company from a year to a month. Could such efficiency gains mean developers are writing many coding jobs out of the future?

Last year, you may have seen a story or two with some variation of the headline, “A Robot Wrote This Article.” These published after a test run of GPT-3, an artificial-intelligence system that has basically read everything on the internet and uses it to generate language (a safety feature tries to stop it from repeating awful things). Amid the excitement some took to Twitter, showing that GPT-3 could code, too. “This changes everything,” wrote one designer, along with a video in which he typed in “an app that has a navigation bar with a camera icon.” The computer spat out an eerily Instagram-like interface. Soon after, Microsoft announced its exclusive license to put GPT-3 in its products, following a $1 billion deal with the model’s creator, OpenAI, the year before. The Power Apps addition is the first fruit of this new tree. 

Charles Lamanna, a Microsoft vice president, says the company has been making demos of similar systems for years, but they’ve “never been good enough to release.” GPT-3 helped change that. Right now, the program is mostly useful for, say, building an in-
company financial data app—not whipping up a TikTok competitor. 

As it advances, Lamanna thinks it’ll actually create more jobs. The idea, he says, is “democratizing development as much as possible.” Coding languages, after all, are already increasingly user-friendly translations, from binary’s ones and zeros to Python’s semi-English. As AI-written code improves, rather than eradicating jobs, it might lessen “monotonous syntax and expressions” so developers can focus on creative work. 

Bill Howe, founding chair of the University of Washington’s data science master’s program, agrees on that point—less monotony, more creativity—and isn’t concerned about AI code wiping out careers. Yet he’s also not convinced that this Power Apps model will do much to democratize coding. Yes, we can now describe an app and get a rough draft of what we envision. But, he says, “the first version isn’t what makes software good or bad, usually.” A program’s success is iterative: the debugging, the new features. That rough cut might save experienced developers time, but for companies to progress, to accelerate, to prosper as Nadella’s bookshelf has it—we still need that human touch. 

Filed under
Share
Show Comments