Word dump code Aboukhadijeh’s finger.
As a follower of open source software movement, he’s written incredibly popular web applications, peer-to-peer file exchanges, and over 100 other pieces of code he’s given away, all in the 10 years since he graduated from college. Recently, however, Aboukhadijeh has entered a new type of flow state, powered by an instrument called Copilot. It’s part of artificial intelligence software that does some typing and thinking for him.
Built by OpenAIprivate research labs, and GitHub, the site owned by Microsoft where programmers share code, this tool is basically autocomplete for software development. Many times Gmail tries to complete a sentence as you write it, Copilot suggests completing a segment of your program. The tool was released last summer to a select group of programmers.
Aboukhadijeh quickly discovered that Copilot was so good, it was almost unsettling. He’ll start typing a line of code, and within seconds the AI will figure out where he’s going—then boom, the next four or five full lines will show up as light gray text , he can accept by pressing Tab . When he saw that it produced clean code exactly what he intended, he found it a bit odd. “How did it get to these predictions?” he wondered again. “Some of them are really weird.”
For weeks, Aboukhadijeh left Copilot on while he worked. He discovered that it had other impressive tricks; it can even understand commands he writes in basic English. If he just typed in his code editor “Write a function that capitalizes every word in the document,” Copilot would pull all that code together on its own. He’ll check to make sure it’s free of errors; sometimes it did.
Furthermore, the tool has Advanced your code. For example, at one point, Aboukhadijeh needed his software to recognize several different formats of text documents, so he carefully listed all the formats, one by one. , in my code. Instead, Copilot recommends a single, neat command that elegantly scans them all together.
“I was like, how does it even…?” he said, lingering in surprise. He doesn’t think he’s going to turn off Copilot.
He’s not alone either: Nine months after Copilot launched, tens of thousands of programmers have tried the software. I’ve talked to 15 programmers who have used it, and most, like Aboukhadijeh, find that it dramatically speeds up their speed — even if they sometimes get nervous about how good it is. . (“It was amazing,” as Mike Krieger, the original Instagram coder, said.) True, they also found it to be full of bugs, from the gross to the disturbingly subtle. GitHub and OpenAI tracked Copilot’s performance through anonymized data on the number of proposed lines accepted and the number of subsequent archives on GitHub. They found that the AI writes a significant 35% of newly posted user code.
Since the advent of computers, people have been hunting for ways to make them easier to program. America’s first programmers, the women who created instructions for the ENIAC machine in 1945, had a ridiculously difficult job: They had to construct logic with rope. In the 50s, punched tape and cards made the job a little easier. Then there are programming languages with English-like syntax, some of which — such as Basic or Cobol — are explicitly designed to encourage neophytes. In the 90s, languages like Python automated some of the most frustrating and frustrating parts of coding, like memory management. In the 2000s, the open source movement spawned a generation of programmers who rarely wrote everything from scratch.
Suffice it to say, the Hollywood image of a programmer frantically typing in a bunch of their own code hasn’t been true for years. By piecing together pieces written by others, individuals can create applications that are vastly more complex than was possible 20 years ago. Copilot promises to be the next major step in this decade-long orbit.
