AI is changing coding—and adding friction

As enterprises continue to implement AI solutions for coding, the growing pains are leading to an increase in AI tech friction.

A new study from agentic platform Harness breaks down concerns over AI deployment, including how employee use of vibe coding is leading to extra work. Majorities of developers said reviewing AI-generated code for accuracy (53%) and fixing small bugs (52%) are top sources of tech friction.

Trevor Stuart, Harness SVP, told IT Brew that the ROI for AI is tempered by the increase in friction and rework. That “invisible tax” often presents itself in added tasks that are difficult to measure, such as fixing vibe coding.

“You have agents working on the code, refining the code, presenting you back their body of work—you’re refining it, you’re more of an architect,” Stuart said. “A lot of time is being spent on the spec and working with the agents to write the spec to work through [the process and] understand the problem.”

IT pros and cons. Even for seasoned developers, AI coding can play a role in the software development life cycle. IT pros have told IT Brew that this use of AI has expanded the capabilities of the tech stack, but also comes with complications that make human oversight important.

One of the problems with measuring AI coding is that the metrics for development success are outdated. Matt Graney, CPO at Celigo, told IT Brew that standards must shift.

“There’s a longstanding truism that you can’t measure developer productivity in terms of ‘number of lines of code produced’—it may be some indicator, but it certainly can’t be the only one, and I think there’s a similar thing going on here,” Graney said.

How to best implement this new technology with long-standing processes is the question, said David McJannet, co-founder and CEO of Dome Systems.

“There is a software development life cycle that needs to be followed for any active software development,” McJannet said. “We’re all relearning that…if you want to build professional software, there is a process that needs to be followed.”

Human element. Code development will likely require some degree of human input and oversight for the foreseeable future, and developers and engineers are trying to find ways to streamline the process. Alex Levin, co-founder and CEO of agentic AI management platform Regal, told IT Brew that IT pros are still tripping up over how long things take: “There’s time in review, there’s time in maintenance, there’s time in bug fixing, there’s time in new feature development.”

Simply implementing an AI solution won’t reduce those tasks to zero, but IT teams that accept the constraints and opportunities, and work with them, are in the best position.

“There’s a new opportunity where the best companies are still using this to focus on the things that are making them differentiated and focusing on reaching the [individual contributors] in their company to be ready to be managers of agents,” Levin said.

Stuart believes that the code review problem will work itself out in a matter of months. But the challenge remains; AI was supposed to make things easier for software development, not harder, and agents are wrangling with increasingly complicated code.

“When humans used to write code…we would write code, we would merge it, someone would review it, we all try to get the [pull requests] small so we can get our review times down,” Stuart said. “Agents are writing complex bodies of work now, and depending on how you’re setting up that process they might be coming back with a pretty significant body of work with all their test cases, all their scenarios that the human now has to review, and so that just introduces a layer of complexity.”