AI is coming for coders — the very people who created it — but not quite yet

The shiver that Gadhia and his crew experienced wasn’t the first to roll through the software industry. And AI’s promise to wipe away the mundanity from many jobs — including in coding — by taking over the routine stuff means there are sure to be more spells of anxiety for the people who make the software that runs the world.

That’s because it’s clear, long term, that AI is coming for the coders. Yet what that will look like remains uncertain.

The fallback position, for now, is that whip-smart bots aren’t likely to take on all of what coders do — because software folk do a lot more than code.

Coders don’t just code

According to GitLab, developers spend over 75% of their time doing other things. Several veteran software engineers told BI the amount of time spent coding might be closer to half. And half a job is still a decent amount of work.

But maybe not for the new guys.

That’s Gadhia’s concern. The jitters he and his friends experienced back in March were over the rollout of a tool called Devin that was meant to do the work of a coder. He said his nervousness and that of his coding buddies eased as early tests indicated the tool was impressive, yet far from foolproof — for now.

But as the technology pushes ahead, it’s likely to rejigger how newbie software developers earn their stripes, he said.

“Junior engineers,” Gadhia said, “have a little bit of a target behind their back.”

One major thing he’s concerned about is that if AI replaces the greenest engineers, it’ll be harder for coders to develop foundational skills necessary to move to the next level — like running without learning to walk.

“Are we going to stop having senior engineers because there’s just no junior engineers?” Gadhia said.

Questions about how coders’ careers might unfold feel more urgent following recent comments from Amazon Web Services’ new chief, Matt Garman. In an internal discussion in June, he predicted that AI could take on much of coders’ workloads, according to a recording of the discussion previously obtained by BI.

“If you go forward 24 months from now, or some amount of time — I can’t exactly predict where it is — it’s possible that most developers are not coding,” Garman said.

Meantime, in 2023, GitHub’s CEO said its widely used Copilot would write 80% of code “sooner than later.”

Being the boss of bots

Still, this may not be as shocking as it seems. Madars Biss, a tech writer and frontend developer, told BI in an email that coders’ jobs began a gradual change years ago as tech has evolved. He said he’s always viewed AI as a tool for boosting his productivity.

And over the next five years, developers might spend less time writing code from scratch and more time overseeing code that AI generates to make sure it meets the standards that coders set for quality and security, Biss said.

“This could lead to a workflow where AI tools handle much of the routine and repetitive tasks of the developer, and humans focus on managing, double-checking, and creativity.”

Biss said it’s hard to predict the future because the AI landscape is changing rapidly. The tech could, he noted, become much more efficient at flagging security vulnerabilities or automating some parts of the quality-assurance process.

But “for now, these areas of software development still heavily rely on human expertise,” he said.

Derek Holt is CEO of Digital.ai, which makes software that helps people build, secure, test, and deliver software. Holt is a software engineer and computer engineer by training. He said notions that software developers are an endangered species are overwrought.

Holt said coding jobs will continue to change in big ways, just as they did when widespread internet use became the norm, but he doesn’t expect the work will disappear.

“The roles will evolve, but software development is here to stay,” he said.

Holt said that, if anything, it’s growing more complex. That’s in part because more businesses see themselves as software companies, which will drive up demand.

“Productivity keeps going up, but also the needs keep going up as well,” Holt said.

The US government agrees. It forecasts demand for software developers, quality assurance analysts, and testers will jump 17% from 2023 through 2033. That’s far above the average growth rate of 4% for all occupations.

Where Holt sees AI as being useful beyond generating code is in areas like creating documentation for software projects — a task he contends few developers have ever loved. AI is also masterful at parsing the heaps of codes companies have, Holt said. Major companies might possess hundreds of millions of lines of it and, in some cases, billions.

“No human can understand all of that,” Holt said. AI can help programmers reuse code rather than create additional lines that can, over time, become too difficult for an organization to manage, even with AI acting as the ultimate librarian.

Holt said that, for now, most copilots are putting out work of a “slightly below average junior developer.” But he expects the bots will get better.

“That bar is going to get raised. And I think your copilot or your assistant is going to go from being a junior developer to being a bit more of a senior developer. And I think that’s a good thing,” he said.

Eating into the gains

Jyoti Bansal, cofounder and CEO of Harness, a company that helps software developers, is also skeptical that AI will boot coders out of work. That’s because so much of their roles involve testing for security, reliability, bugs, and scalability.

Bansal, who’s started and run several companies, told BI that while AI can speed up the coding process, it can also gum up the works. That’s because code generated by AI tends to contain more bugs, he said. That, in turn, requires more extensive testing to straighten out the kinks.

Bansal said the quality assurance testing that’s required can offset the productivity gains that can come with AI — for now.

Within three to five years, though, he expects AI could deliver productivity gains of 20% to 40%.

Plus, for AI to realize more of its potential, the technology will need to be used more broadly in areas beyond coding like QA and DevOps, security, and compliance, Bansal said. At present, the productivity gains aren’t uniformly distributed across organizations. That can create bottlenecks.

Software developers, he said, will need to shift more of their focus to areas where AI hasn’t fully taken up residence — things like quality assurance and development and operations.

Charlotte Relyea, a senior partner at McKinsey & Co., told BI that when individual engineers use Gen AI copilots, they can boost their productivity, but if the whole system for pumping out software isn’t redesigned, inefficiencies can bubble up. That was the case with one McKinsey client that deployed AI unevenly, she said.

“Their engineers were freeing up all of this time, but the product managers hadn’t actually adopted it. And so the product managers weren’t actually either giving them additional work to do or giving them additional requirements that they wanted to run,” Relyea said.

That meant engineers were just using the extra time to do what they wanted — though those things weren’t necessarily in line with the company’s strategy, she said.

You still need to know what you’re doing

Michael Solati, a software engineer for a major Silicon Valley tech company, told BI he often turns to AI when he’s working in a programming language he’s not as familiar with. He can use his deep understanding of one language to ask a chatbot how he might make a concept work in a different one.

“It makes that converting process super, super easy,” he said, comparing it to using Google Translate to go from English to French. “It speeds up my workflow a lot,” Solati added.

He’ll then read the code line by line to make sure he understands what the AI is generating. And if he doesn’t get something, he can toss the bot a follow-up question.

But — reverting to concerns raised by Thoughtful AI’s Gadhia — where there might be trouble is if junior developers become too reliant on AI and don’t know enough to spot problems in what AI kicks out, Solati said.

“If you can’t do a smell test on the code being generated, then I would worry.”

Solati predicts that in a decade, software developers might still spend 40% to 50% of their time coding.

“It’s just going to be coding more, coding faster, coding harder,” he said, adding that instead of pumping out two features a week, he might be up to five by that time.

“I don’t know if I want that or not, but that’s the future,” Solati quipped.

Gadhia said a lingering concern for the industry is whether autonomous AI agents will replace senior engineers. But what will be a lot harder for AI to take over is the very human work of gathering context on the needs of an organization, he said.

“You require a lot more skills than just writing the code,” he said. “It’s communication, collaboration — those are hard to replace.”