When Intuit chief technology officer Alex Balazs was getting his undergraduate mechanical engineering degree at Kettering University more than three decades ago, he recalls the Michigan school’s professors being split on whether they’d let students use calculators in class.
“And now, when you think about it, of course you use a calculator,” says Balazs.
Similarly, he believes today’s AI coding assistants that help developers write software will gain broad acceptance in schools and the workplace. In response, he said Intuit is reevaluating how it tests potential engineering hires during the interviews.
One expected change is that the coding exercises will be more complicated, requiring that candidates solve bigger problems, with the expectation that those candidates will use AI tools to complete some of those tasks. “Because when they arrive inside Intuit, that’s how we expect them to work,” says Balazs.
CTOs and chief information officers frequently laud the big productivity gains the coding assistants provide and the help they give employees in getting off to a faster start at their new jobs as they learn company-specific programming languages. Usage rates, which many CIOs and CTOs have been closely monitoring, have increased steadily over time.
With new tools comes a rethinking of the skills required for an AI-enabled developer workforce, says Deborah Golden, chief innovation officer at accounting and consulting giant Deloitte. It will be less important for engineers to memorize application programming interfaces (APIs), the rules that let software applications communicate with each other, and more critical for them to show good judgement on the job, including determining if there are any risks or bias in AI-written code.
“AI doesn’t just level the playing field, it tilts towards those that can adapt quickly,” says Golden. For both new college graduates and more established working professionals, embracing AI means “anybody can be left behind the same way that anybody can leap forward,” she adds.
Several CEOs of major corporations have said that 20% to 30% of code written within their companies is being done by AI tools. But those claims should be taken with a grain of salt, according to Andrew Rabinovich, the head of AI and machine learning at online freelancer marketplace Upwork. “The numbers can be highly inflated because it’s verbose,” says Rabinovich, referring to AI coding assistants regularly churning out unnecessary lines of code.
He also says coding assistants aren’t good at personalization, or gearing what they write to the tastes of more senior software engineers. Some of those managers may reject AI-written code if not presented the way they like it.
“The older or the more experienced of a software engineer you are, the more habits and rules you impose on the LLM in order to be satisfactory,” says Rabinovich. “But if you’re a junior software engineer, it’s kind of an open playing field, and you’re like, ‘I’m okay with everything, as long as it gets the job done.’”
Brendan Humphreys, the CTO of Australian software maker Canva, says some in the industry have expressed concern about “cheating” during the interview process, with candidates using AI tools to mask how well they can write code. “We think that’s the wrong framing,” says Humphreys. “Software engineering as a job has fundamentally changed. And you need to now demonstrate that you can have competency in using these tools to accelerate your output.”
With that in mind, Humphreys has changed Canva’s assessment criteria to make it tougher, yet more ambiguous—meaning job prospects cannot just feed inputs into LLMs to get a response that would satisfy Canva’s expectations. “You’re going to have to work with an AI intelligently and we want to see that competency,” adds Humphreys.
That’s led Autodesk to embrace more adaptability, actively encouraging software developers to be less siloed on specific projects. “We’re creating an environment within our company where it’s okay for you to disrupt another team’s work,” says Arasu.
Nikhil Krishnan, senior vice president and CTO of data science for the enterprise AI software company C3 AI, says his business almost always conducts in-person interviews, so there’s little risk that candidates are cheating with AI tools unless C3 wants to test them on how to solve a problem with an AI coding assistant.
He prioritizes problem solving and reasoning skills and is on the hunt for candidates who have curiosity, a passion to learn, can show their ability to absorb new information, and work well on a team. With those skills in mind, Krishnan says C3 AI has a bias toward more senior candidates.
“I do see a world where it becomes harder and harder, as a junior entry-level software engineer, to land that first opportunity,” says Krishnan. “We certainly find that we are much more careful about who we’re hiring.”
John Kell