How AI is "collapsing" the role of the software engineer

The continued injection of AI into the world of software engineering means that coding will take a back seat when it comes to measuring the effectiveness of software engineers, according to a Gartner VP analyst.

Manju Bhat, who has been with Gartner for over a decade and previously worked as a software engineer at Manhattan Associates and an R&D director at VMware, said that while AI is helping to compress experience in a big way, this comes at a cost

"What would probably have taken a developer five years to learn, AI can compress the same learning down to five months,” he said.

“When you compress experience exponentially, [however], it has an effect on the skills that [engineers] would have otherwise gained, because when you compress experience, you're also starving them of experience.”

One way this manifests, Bhat said, is in the loss of gaining skills that are "battle tested", meaning junior engineers of today don’t have an insight into what ‘good’ looks like in terms of coding.

“[Where] senior engineers are able to say ‘this is good code, this is bad code’, the future is more like code is irrelevant, which then begs the question of what should the next senior engineers be doing?” Bhat stated to techpartner.news.

“We believe that the engineers will not be measured based on the code they write, but measured based on all of the non-functional characteristics of software engineering, which so far has been scattered across different teams. Now, all of those non functional attributes - reliability, resilience, security, performance, architecture, design - are captured into one role.

“You [previously] had coding as the core of what a software engineer does and everything else was the adjacencies or the fringe activities; now, with that core collapsing and AI taking over, it subsumes all of those adjacencies into the role itself.”

Productivity benefits at the cost of skills

That loss of experience by way of AI use was a finding that Anthropic, the makers of the Claude family of large language models and AI chatbots, also found in a study released earlier this year.

In a randomised controlled trial, the company examined both how quickly software developers picked up a new skill with and without AI assistance, as well as whether using AI made them less likely to understand the code they’d just written.

The study found that using AI assistance led to a statistically significant decrease in mastery.

On a quiz that covered concepts they’d used just a few minutes before, participants in the AI group scored 17% lower than those who coded by hand, or the equivalent of nearly two letter grades.

Using AI sped up the task slightly, but this didn’t reach the threshold of statistical significance.

“Productivity benefits may come at the cost of skills necessary to validate AI-written code if junior engineers’ skill development has been stunted by using AI in the first place,” the study stated.

“Managers should think intentionally about how to deploy AI tools at scale, and consider systems or intentional design choices that ensure engineers continue to learn as they work—and are thus able to exercise meaningful oversight over the systems they build.”

Moving from implementer to orchestrator

Those proposed productivity benefits are also the first of three themes that Bhat identified in recent research he penned for Gartner on the impact of agentic AI in the software development lifecycle (SDLC).

He found that AI is going to augment different roles in the software team, not just developers - product owners, quality engineers, site reliability engineers, designers and more are all slated to also take advantage.

“All of these roles benefit tremendously from the use of AI and when we amplify and augment these roles, we start to see true benefits of AI for the whole team,” Bhat told techpartner.news.

“One of the big problems that we see in many organisations is that even though individual developers are becoming productive, organisations, especially leaders, find that teams are not productive.

“The reason is they're obsessed with improving developer productivity, while not [improving productivity] of other roles.”

Bhat also predicted that while a lot of software development workflows today are synchronous, this will shift to become increasingly asynchronous, meaning you can “parallelise” activities.

“You can send a command to an agent and and say ‘fix all the vulnerabilities in this repository', and at the same time, spin up another agent and command it to fix all the bugs in a previous release’,” he explained.

“You can almost become a developer superhero, and say ‘I have a team of agents working for me, all in parallel, and all I have to do is orchestrate those agents’. This means the role of a developer evolves from being like an implementer to an orchestrator of agents.

“The third theme, which we believe is the future, is what we call autonomous software delivery through frictionless SDLC. This concept is going beyond augmentation or asynchronous to where you could set a desired behaviour of an application and you don't care how the implementation is arrived at, [just] that conversion from intent to implementation happens in an autonomous manner.”