December 4, 2023
Screen Shot 2023-01-25 at 9

Screenshot from OpenAI Website

AI-powered chatbot ChatGPT is, according to its boosters, poised to disrupt or even make obsolete everything writerly—from the college essay to legal arguments. But large language models are already transforming another human pursuit: writing code. 

Coding, essentially the process of humans communicating with computers, didn’t always refer to using programming languages like Python and Javascript to develop an application. During the early days of computer programming, before keyboards were invented, coding meant flipping switches. Then, coding meant punching digits into cards. As high-level coding languages were introduced, beginning with C in 1972, coding started to become not only more efficient but also closer and closer to our natural human language. Along the way, coding became a highly-valued, and high-paying, skill, to the point that coders have long told laid-off writers, “Learn to code.”

Now, with the development of large language models such as ChatGPT, many computer scientists see a near future in which everyone will be able to code simply by using natural language. And in a climate of mass layoff and austerity, the technology is already being looked at as a way to juice more from workers. The evolution in coding is happening so quickly that maybe the advice should be flipped. Perhaps some people working in software development should be told: “Learn to write.”

While many people have been worried about ChatGPT’s English language-writing capabilities being able to replace writers and students, another important feature of the AI model is its ability to translate natural language into code, as well as auto-complete, debug, and suggest code. In August 2021, OpenAI released Codex as a private beta tool that can translate natural language commands to code. According to OpenAI, Codex understands natural language because it is a descendant of GPT-3 and is, in part, trained on the same data as the chatbot. 

“The release of OpenAI Codex, a new Al system that translates natural language to code, marks the beginning of a shift in how computer software is written,” wrote OpenAI co-founder and CTO Greg Brockman and CEO of Hadi Partovi in an article on TechCrunch. This article, titled “No-code is code” brought attention to a growing movement in computer science called “no-code,” in which people can develop software without technical knowledge of coding languages, no computer science degree required. 

Amjad Masad, the CEO of Replit, an online platform for collaborative coding, wrote in a 2020 blog post that the advent of AI-powered coding tools is going to transform programming, making it easier to learn and more efficient. 

“At some point—probably not in the near future—the word ‘coding’ will disappear from our lexicon because programming will stop requiring code and instead be about the pure act of solving problems using computers making it accessible to more and more people,” Masad wrote. 

“We’re only scratching the surface of what’s possible in this new technology. I think ChatGPT brings it to another level,” Masad told Motherboard in an interview. “We’re now at the start of another big jump in developer productivity. I think it’s going to be anywhere between 10x to 100x improvement in productivity.”

Masad said that having knowledge of coding will still be a good skill to have, as AI will simply help expedite programming processes and break down some of the barriers to access to computer science. “There really aren’t enough programmers in the world, they’re very expensive. The more efficiency we have in programming, the more software we’re able to create. An AI assistant can help you debug your code, can help you make it better, and refactor it, and that will just make every aspect of the software development lifecycle better. On the accessibility side of things, I think it will make software more accessible for people.”

The advent of “no-code” coding using AI isn’t all sunshine and roses, however. Already, it’s running into the same ethical problems as image generation models trained on people’s art without consent, and it has potentially troubling implications for workers who code. After all, technology that boosts workers’ productivity has historically come hand-in-hand with a reduction in the total number of workers a company employs.

Currently, Microsoft’s GitHub has its own AI auto-completion coding tool called CoPilot, which launched in June and suggests code to users in real time. However, CoPilot, which is powered by Codex, is facing a class-action lawsuit from GitHub programmers who claim that Codex used their code without proper licenses. 

Tech companies have recently laid off masses of workers—Google, Microsoft, Meta, and Amazon have collectively laid off hundreds of thousands of their employees in the last few months. It’s not clear that this had much to do with AI replacing coders, but companies are already hoping AI can allow workers to do more with less.  

Google has already implemented AI for its internal developers, in which 10,000 Google software developers used a code completion language model, in their programming tasks. According to a blog by CEO Sundar Pichai, AI reduced “coding iteration time for these developers by 6%.”

Many billionaire investors are putting pressure on companies like Google to reduce employee headcounts and be more stringent on budget as stock prices decrease. Even if AI is not leading the charge to layoffs, it’s part of an emerging framework of marginally increased productivity at a time of austerity.    

Masad told Motherboard that he does expect to see one part of software engineering disappear: back-end and full-stack engineering, or people who build the structure of applications and software. “You’ll see the product and the front-end engineer be able to do a lot of the work that maybe the back-end or full-stack engineer was doing in the past. I think that’ll create pressure from both sides, which could have an impact on the employability of those engineers in the middle, and they’re gonna have to specialize. Or they’re gonna have to either go and build products or become low-level platform engineers.” 

With the inclusion of AI in coding, Masad said that the value of software engineers will be more about the ability to build something new and to supervise and manage the code, rather than just write it. 

“While the AI could support you in some way, it would not be able to write the entire program in itself. So I think the novelty of the code that you’re going to create will dictate the AI’s effects on your job,” he said. 

Masad is hopeful about the future of AI-generated coding processes. “It’s important to understand that despite the layoffs and everything there are still a lot of unfilled software roles in the world today,” he said, adding that he thinks Silicon Valley has been “hoarding” talented workers.

Either way, coding is soon probably going to look different than it does today, or a year ago, just like coding in C++ is different from coding using assembly language. That will have effects on the cohort of people who code for a living—essentially, talking to computers—when the computers start talking back.