As reported by TechCrunch.
Carla Rover has about 15 years of experience in web development and is now launching a startup with her son that develops custom machine-learning models for marketplaces.
She described vibe coding as a beautiful, boundless cocktail napkin on which you can continually sketch ideas. But using AI coding for production can be “worse than babysitting,” because such models sometimes produce unpredictable glitches or behave differently than expected.
With a drive to get her startup off the ground quickly, she turned to AI-assisted coding, encouraged by modern tools promising to help.
“Because I needed to be quick and to impress, I took the shortcut and didn’t scan the files after the automated review.”
“I handed this off as if a copilot were an employee,” she says. “That’s not the case.”
Such experienced developers, who previously wrote code themselves, are increasingly taking on the role of AI supervisors – rewriting and vetting code produced by AI.
According to Fastly’s study, nearly 95% of the roughly 800 surveyed developers spend extra time fixing AI-generated code, and the burden of verification falls mainly on senior professionals.
In the experience of these coders, problems often arise – from hallucinated package names to accidentally removed important data and security risks. If not monitored, AI-generated code can make a product significantly more buggy than people would expect.
That’s why a new corporate role has emerged – the “vibe-code cleaner” specialist.
The future of vibe coding through the eyes of experienced developers
In conversations with developers, their experiences with AI-generated code and visions for the future of vibe coding are discussed. Although opinions diverge, one thing remains constant: the technology has a long way to go to become more reliable and safe.
“Using a copilot for coding is like giving a smart six-year-old a coffee maker and asking them to bring coffee to the family in the dining room.”
“Will they be able to do it? Maybe. Will they fail? Without a doubt. And if they do fail, they will probably not admit it right away.”
“You’re absolutely right!”
Feridun Malekzade, with over 20 years of experience, also works with vibe coding through his startup and the Lovable platform. He believes vibe coding does not replace an intern or a junior developer – it’s more like “hiring your stubborn teenager to help get something done.”
“You have to ask them to do something 15 times. Eventually they’ll partially fulfill your request, partially not, and in the process they’ll break quite a few things.”
According to Malekzade, about half his time is spent formulating requirements, 10–20% on vibe coding, and 30–40% on fixing and “healing” AI errors.
He also notes that vibe coding does not grasp full systemic thinking – AI typically focuses on solving surface-level tasks; if a scalable feature for the entire product is needed, an engineer will do it with the team and place it in the appropriate parts of the system.
Meanwhile, Austin Spears of Fastly emphasizes: vibe coding usually seeks speed, but not “correctness,” which can create vulnerabilities in the code, especially for junior developers. Companies are responsible for safety through access control, peer review, and screening.
“It often happens that an engineer needs to review code, fix the agent, and say: ‘you made a mistake’.”
At NinjaOne, they are implementing the practice of “safe vibe coding” with mandatory peer review and security screening to reduce risks and preserve quality control.
Despite the debates, vibe coding has already transformed the present and future of the profession: for some, a high-speed tool truly acts as a catalyst, but it requires a responsible approach and human oversight to turn a cocktail of ideas into a stable product.
“Even as I grow older, I will keep using it.”
