Artificial Intelligence and Machine Learning

Will AI Replace Computer Programmers?

Artist's representation of an AI that writes code.
Not only do programmers work faster with AI assistance, it also frees them up to focus on more complex (and usually more rewarding and higher-value) tasks.

Since ChatGPT took the world by storm late last year, white-collar professionals have been forced to reckon with the fact that artificial intelligence (AI) might soon do parts of their jobs better than they do.

So far, an explosion of so-called "generative AI" tools—machines that generate text and imagery—has writers and designers equal parts excited and apprehensive.

On one hand, these tools give some creative types unprecedented abilities to tell compelling new stories, create outstanding content, and produce innovative art at scale.

On the other, they have some looking over their shoulders as they see AI increasingly invade an area of knowledge work typically reserved for human beings—threatening their skillsets and livelihoods in the process.

Soon, computer programmers could be equally divided with the release of a clutch of increasingly powerful generative AI tools that produce code automatically.

AI-powered coding assistants like OpenAI's Codex model, GitHub Copilot, and Replit Ghostwriter are changing how computer programmers do their jobs. That's because, thanks to advancements in the large language models that power them, these tools can now, in some instances, automatically generate reliable code.

In the process, they are having a significant impact on programmer productivity and causing some to ask bigger questions about how AI coding copilots will affect coding work and jobs.

But just how good are AI tools that can code? And what do they mean for the industry at large?

If you rated today's AI coding tools on a 10-point scale, they're at a three, says Shanea Leven, founder and CEO of CodeSee, which builds solutions that help companies understand their codebases. There's no question they can speed up basic coding tasks, Leven says.

They're also useful for generating ideas and boilerplate code, according to Ilkka Turunen, Field CTO at Sonatype, a software supply chain management company.

As a result, these tools are having an immediate impact on novice and expert programmers alike. For novices, models like Codex can help them immediately solve basic problems, says Paul Denny, associate professor of computer science at The University of Auckland, New Zealand.

"For professional developers, already some good evidence is emerging for improvements in productivity," Denny says. A recent study by Github found that 88% of programmers said they were more productive when using Copilot. That was thanks to benefits like less time spent searching for code, getting stuck on or bored with repetitive tasks, and staying in flow longer.

Not only do programmers do their work faster with AI assistance, but they also free themselves up to focus on more complex (and usually more rewarding and higher-value) tasks.

However, AI coding copilots today still have serious limitations. Active codebases are highly customized, says Leven; that gives AI tools very little usable data to learn from quickly, making their usefulness limited. Even with enough data, today's tools are unable to handle the levels of complexity that serious programming challenges require.

"Many times when a project is complex, AI also needs to consider the needs of the business when making decisions and tradeoffs, something it just can't do today," Leven says.

Not to mention there's no guarantee the outputs of these tools are correct. Large language models are good at "hallucinating" or confidently producing inaccurate outputs, so you still need to have an extremely knowledgeable user overseeing these tools, says Turunen.

As a result, don't expect an entry-level programmer to suddenly become a software development superhero just because they use AI. And, given the limitations of today's tools, coding jobs aren't going away any time soon.

"As they stand right now, AI tools do not signify the end of the software developer," says Turunen.

Leven agrees. "Anything is a possibility given enough time and resources, but engineer jobs are safe," she says. In fact, many engineers will likely become more valuable thanks to these tools as they reallocate their time to tackling bigger challenges across the organization.

Still, don't underestimate how fast AI coding tools are progressing, says Denny; it's still very early. OpenAI's Codex model has already scored in the 75th percentile on CS1 and CS2 coding problems compared to human students. And it was released only in 2021, just a year after the company's GPT-3 model that powers it. In a short time, we've gone from a zero out of 10 to a three out of 10, and the technology shows no signs of stopping.

"The ability to synthesize source code from natural language descriptions of problems is going to be hugely impactful," he says.


Logan Kugler is a freelance technology writer based in Tampa, FL, USA. He has written for over 60 major publications.

Join the Discussion (0)

Become a Member or Sign In to Post a Comment

The Latest from CACM

Shape the Future of Computing

ACM encourages its members to take a direct hand in shaping the future of the association. There are more ways than ever to get involved.

Get Involved

Communications of the ACM (CACM) is now a fully Open Access publication.

By opening CACM to the world, we hope to increase engagement among the broader computer science community and encourage non-members to discover the rich resources ACM has to offer.

Learn More