 |
VOOZH | about |
|
VOOZH | about |
We’re so glad you’re here. You can expect all the best TNS content to arrive Monday through Friday to keep you on top of the news and at the top of your game.
Check your inbox for a confirmation email where you can adjust your preferences and even join additional groups.
Follow TNS on your favorite social media networks.
Become a TNS follower on LinkedIn.
Check out the latest featured and trending stories while you wait for your first TNS newsletter.
Enterprises everywhere are looking for ways to leverage AI, with code generation leading the way. Talented software engineers are in short supply, and maximizing their productivity even further — while taking the burden of creating boilerplate, routine code — results in a win on both sides.
But of course, with every new technological innovation, there are challenges. Not all is as easy as it seems.
Today, as enterprises try to adopt vibe coding to scale their engineering teams, they’re consistently running into considerations and challenges. Here are the top five concerns that enterprises are facing.
As we’ve heard frequently, generative AI (GenAI) continues to hallucinate. It can produce results that are less than optimal, with mistakes that differ from those a human would make, which sometimes makes it more difficult for humans to detect the errors.
One example, translated to the world of writing, is that AI creating an article on vibe coding for the enterprise might produce an article without any grammatical errors (a common mistake that humans make). The introduction and conclusion would likely make sense. But partway through, AI might veer off course. It might add a paragraph about the history of the word “vibe,” the phrase “good vibrations” and the rise of the Beach Boys. An experienced writer would review the article for grammar, syntax and logical flow. But a writer with a lack of knowledge around vibe coding or technical subjects (and pop culture as well!) may not catch what needs to be taken out.
At the same time, GenAI is creating those errors on a far faster scale than humans are accustomed to dealing with. It can be challenging to review these vast amounts of code properly, especially to ensure adherence to coding standards, security protocols and regulatory requirements. AI-generated code may not incorporate these standards, which can result in compliance and security risks.
Accumulating tech debt is a common concern for enterprises. Vibe coding gone wrong, without guardrails put in place, can result in unstructured code and inconsistent styles. This can make it harder to identify, prioritize and address areas of tech debt. There can be unclear logic and dependencies that can break easily when changes are made. A lack of clear ownership (“AI did it!”) can mean that it takes longer for teams to take ownership of fixes and improvements.
And all of this can lead to even more difficulty when it comes to peer review and code walkthroughs. When the code doesn’t have consistency and structure — when it “works” but it’s unclear why and teams push it through because it’s not their job to consider the long-term consequences, only to fix the problem — that’s when you start to see issues. Many teams aren’t structured correctly to address the new set of issues that arise with AI-generated code.
In addition, reviewing code and generating code are different skill sets, and engineering teams may not be adequately staffed to address how AI has altered their team’s needs.
When vibe coding doesn’t adhere to standardized coding and security guidelines, it can introduce new vulnerabilities, such as injection attacks or improper access controls. And if informal coding is paired with automated testing and scanning tools, the problem can be exacerbated even further.
AI-generated code, with its unclear logic, can be more difficult to review and may not always align with established secure coding standards.
In addition, it’s theorized that as part of a long-term (and very sneaky) plan, bad actors could introduce vulnerabilities as part of an AI model’s training logic. The model could then be published as fit for the enterprise, with those vulnerabilities then generated as code for developers to introduce into their enterprise production systems.
Famously, Mark Zuckerberg’s early motto for Facebook was “move fast and break things,” a motto that has since been rescinded. This has always been a dilemma for enterprises. Move quickly, break things and take over the market? Or move slowly but accurately and maintain reliability? In technology, as in life, doing things quickly can mean a lack of robustness and reliability — and this concern carries over to the world of AI agents.
The vision for vibe coding and agents, of course, is not just single agents accomplishing point tasks and then returning to humans for an accuracy check. It’s fleets of agents, coordinating seamlessly. It’s agents upon agents — certain agents creating code, then other agents checking the code, then other agents pushing to production, then more agents checking for bugs and fixes.
We’re not there yet. But someday we will be, and enterprises should continue to experiment and adopt AI technologies to prepare for the new future ahead of us.
