How vibe coding is impacting business in the long term

When you’re flying a commercial plane, the autopilot is on almost all of the time. But that doesn’t mean there aren’t two human pilots in the cockpit at all times, making sure that things are going right.

That’s an analogy Quickbase CTO Jon Kennedy uses to describe the role of engineers and other IT professionals in vibe coding.

“You have to know what the airplane’s doing,” Kennedy said. “At any moment, you might need to take control and hand-fly.”

Vibe-coding is having a moment. Utilizing AI to help software development is swiftly becoming normalized across organizations, leaving IT pros to manage any potential problems.

Companies need to consider their level of liability when choosing to deploy vibe coding, Mike Wehrs, TieTechnology COO, told IT Brew. Issues can vary, and depend on the relative newness of vibe coding to the tech stack, as well as the talent of those using it.

Wehrs noted that users who employ a diversity of tactics, like using multiple models, are in a much better position.

“The people who use multiple instances of AIs and don’t rely on one [model] in parallel dramatically cut down the failure rate and dramatically cut down on the amount of errors that are allowed in,” Wehrs said.

Drop in a bucket. As Kennedy put it, there are roughly two buckets of users: in bucket A, the professional developer who uses AI to accentuate their work; in bucket B, the inexperienced staffer using the technology with a weaker understanding of its capabilities. Both groups need awareness of what they’re dealing with.

“You still need to understand the stuff you’re building, the code you’re writing—you need to review the code, it does make mistakes, it lies to you, it does dumb things,” Kennedy said. “You, as a professional software developer who are developing for a platform that businesses rely on to keep their business running, you need to make sure that everything you’re checking in, you understand, and has been tested and it is sound.”

Taking care. While more experienced users are putting AI to work developing infrastructure and managing busywork, those with a less developer-heavy background are running into challenges related to security.

Security is a particular concern for Eran Kinsbruner, VP of product marketing at Checkmarx. With a lack of consistently used and applied guardrails in production, the technology can generate vulnerabilities in code.

“The AI won’t recommend security practices while [you’re] building software; it’s not baked into the LLM models today,” Kinsbruner told IT Brew. “If you are not instructing AI, you won’t get secure code time after time.”

Like having two pilots in the cockpit, coding still needs humans in the loop—securing infrastructure as needed.