When senior coders ‘vibe’ more than their junior peers

Sometimes it’s those young people who don’t know technology…

When Insight Enterprises surveyed its developers on efficiency impacts from the company’s AI-powered coding options, Parker Johnston, agentic field CTO at the systems integrator, was surprised to learn that the older generation had a better handle on the emerging tech.

According to Johnston’s estimates, senior engineers reported an approximate 30% time savings thanks to AI coding help, compared to between 12% and 15% for their more junior colleagues.

Setting the stage. About a year and a half ago, Insight made GitHub Copilot in VS Code an internal agentic coding option for its team of around 600 developers. The company collected usage data through self-reported metrics, along with telemetry from GitHub and Azure DevOps.

“I was thinking senior folks might not have as much trust and confidence in AI, so they were just not going to use it,” Johnston said.

And what about the freshmen folks and their ability to vibe code, or create an application purely by prompting?

“This next wave and next generation that grew up with Siri, they grew up with all these capabilities. I was expecting them to know how to do a one-shot prompt.”

But the actual results flipped those assumptions, compelling Johnston to help figure out how his company could bring the two teams (and their respective time savings) closer together.

Why senior developers benefited. After talking with team members, one thing was clear to Johnston: “As you become more tenured in your career, you know how things should look and how things should operate.” In other words, veterans can spot when AI-generated code is risky—maybe an app shouldn’t be calling a certain API, for example—and can more easily refine prompts to achieve a desired output.

Johnston said he knows exactly how to implement a logging connector into Azure in “one shot,” using a go-to logging framework that a junior pro might not know based on their comparatively limited knowledge of options and specific company preferences.

Why juniors did not. A younger developer, drawing upon less experience, might have a “blind trust” in AI, according to Johnston, and might use a simplistic prompt like, “I want to set up logging within this application” without specific parameters like frameworks. Less robust prompts create a potential for lots of rework, Johnston told us.

What next? Insight has since built GitHub-hosted agents to ensure both senior and junior engineers and their projects adhere to architecture and security standards, Johnston said. An “architectural review board” agent, for example, pre-checks code changes, a process that used to require a weeks-long wait for an official inspection from the DevOps team.

“I’m now beholden to my organization standards and the security elements put in place and the trust put in place by those agents and those folks that are solely responsible for owning that,” he said.

The saved minutes have provided some time for mentoring and following up with younger coders, according to Johnston: “It’s all those battle scars, the arrows you’ve taken over your career, that we’re able to give back and have stories around and share with them.”

This year’s giant annual developer survey from Stack Overflow study noted mainstream AI adoption in 2025, but not a clear divide between fresh-faced and more seasoned developers.

• While the study noted that 84% of survey respondents “are using or planning to use AI tools in their development process,” the report also found that 47% of experienced developers use AI daily, with almost 56% of early-career devs using the tool with similar frequency.
• According to the data from just over 49,000 global responses, 63% of early-career coders (and 60% of coding vets) view AI tools in either a favorable or very favorable light.

But it might be the experts who can best use AI, according to Wall Street Journal tech journalist Christopher Mims.

“Experts, whether they’re coders or MDs, are able to ask AIs better questions and then continue following up with more prompts, pushing the AI ever deeper into their databases of knowledge. They’re also—and this is critical—better able to evaluate an AI’s responses and recognize when the AI gets something wrong,” Mims writes in his book How to AI.