Carla Rover once spent 30 minutes crying after having to restart a project that she vibe coded.
Rover has been in the industry for 15 years, mainly working as a web developer. She is now building a start -up with her son, creating custom machine learning models for marketplaces.
She called Vibe, who coded a beautiful, endless cocktail napkin on which to draw ideas. But dealing with AI-generated code that one hopes to use in production can be “worse than babysitting,” she said, as these AI models can ruin work in ways that are difficult to predict.
She had turned to AI coding at a need for speed with her startup, just as the promise of AI tools.
“Because I had to be quick and impressive, I took a shortcut and didn’t scan these files after the automated review,” she said. “When I did it manually, I found so much wrong. When I used a third -party tool, I found more. And I learned my lesson.”
She and her son wound up restarting their entire project – hence the tears. “I handed it over as the copilot was an employee,” she said. “It’s not.”
Rover is like many experienced programers who are aimed at AI for the coding of help. But such programmers also find themselves acting as AI-Babysite rewriting and Facts control of the code that AI spits out.
TechCrunch -event
San Francisco
|
27-29. October 2025
A recent Content Delivery Platform report quickly found that at least 95% of the nearly 800 developers surveyed said they spend extra time fixing AI-generated code, with the load of such verification that falls most on the shoulders of senior developers.
These experienced coders have discovered problems with AI-generated code ranging from hallucinating package names to deleting important information and security risks. Left uncontrolled, AI code can leave a product far more buggy than what humans would produce.
Working with AI-Generated Code has become such a problem that it is causing a new corporate coding job known as “Vibe Code Cleanup Specialist.”
Techcrunch talked with experienced coders about their time using AI-generated code about what they see as the future of vibe coding. Thoughts varied, but one thing remained safe: Technology still has a long way to go.
“Using a coding co-pilot is a bit like giving a coffee pot to a smart six-year-old and saying,” take this into the dining room and pour coffee for the family, “Rover said.
Can they do it? Possibly. Could they fail? Definitely. And probably if they fail they won’t tell you. “It doesn’t make the child less smart,” she continued. “It just means you can’t delegate [a task] Like it completely. ”
“You are absolutely right!”
Feridoon Malekzadeh also compared vibe coding with a child.
He has worked in the industry for more than 20 years and holds different roles in product development, software and design. He builds his own startup and using vibe-coding platform lovable, he said. For fun, he also lashes codes apps like someone who generates Gen Alpha Slang for Boomers.
He likes that he is able to work alone on projects, save time and money, but agrees that vibe coding is not like hiring a trainee or a junior nodes. Instead, vibe coding looks like “hiring your stubborn, insolent teenager to help you do something,” he told Techcrunch.
“You have to ask them 15 times to do something,” he said. “In the end, they do some of what you asked, some things you didn’t ask for, and they break a lot of things along the way.”
Malekzadeh estimates that he spends about 50% of his time requirements, 10% to 20% of his time on vibe coding, and 30% to 40% of his time on mood Fixing -The remedy of bugs and “unnecessary script” created by AI-written code.
He also doesn’t think that vibe coding is the best for system thinking – the process of seeing how a complex problem can affect a unified result. AI-generated code, he said, is trying to solve more problems at the surface level.
“If you create a feature that should be widely available in your product, a good engineer would create it once and make it available everywhere as needed,” Malekzadeh said. “Vibe coding will create something five different times, five different ways if needed in five different places. It leads to a lot of confusion, not only for the user but for the model.”
Meanwhile, Rover finds that AI “runs into a wall” when data is in conflict with what was hard coded to do. “It can offer misleading advice, omit key elements that are important, or insert yourself into a thought path you are developing,” she said.
She also found that it is rather than admitting to make mistakes, it will produce results.
She shared another example of Techcrunch, where she questioned the results that an AI model originally gave her. The model began to give a detailed explanation that was going to use the data she uploaded. Only when she called it did the AI model admitted.
“It ran me out because it sounded like a toxic employee,” she said.
On top of this there are security concerns.
Austin Spiers is the senior director of developer activation on fastly and has coded since the early 2000s.
He has found through his own experience – along with chatting with customers – the vibe code likes to build what is quick rather than what is “right.” This can introduce vulnerabilities to the code of the kind that very new programmers tend to do, he said.
“What often happens is that the engineer needs to review the code, correct the agent and tell the agent that they made a mistake,” Spiers told Techcrunch. “This pattern is the reason why we have seen the tropical of ‘you are absolutely right’ appearing over social media.”
He refers to how AI models, like anthropic claude, tend to respond “you are absolutely right” when you are called on their mistake.
Mike Arrowsmith, Chief Technology Officer at IT Management Software Company Ninjaone, has been in software technology and security for about 20 years. He said that vibe coding creates a new generation of it and safety -blind spots, which especially young startups are susceptible.
“Vibe coding often bypasses the strict review processes that are fundamental to traditional coding and crucial to capturing vulnerabilities,” he told Techcrunch.
Ninjaone, he said, counts this by encouraging “secure vibe coding”, where approved AI tools have access control, along with mandatory peer review and of course security scanning.
The new normal
While almost everyone we were talking to agrees that AI-generated code and vibe-coding platforms are useful in many situations-as to spot ideas-they all agree that human review is important before building a business on it.
“The cocktail napkin is not a business model,” Rover said. “You have to balance ease with insight.”
But for all the regretting of his mistakes, Vibe coding has changed the present and the future of the job.
Rover said that vibe coding helped her tremendously to create a better user interface. Malekzadeh simply said that he is still more done with AI coders than without them, despite the time he spends fixing code, and is still done with AI coders than without them.
“‘Each technology carries its own negativity invented at the same time as technical progress,” Malekzadeh said, quoting French theorist Paul Virilio, who was talking about inventing the shipwreck with the ship.
Benefits outweigh the disadvantages.
The rapid study found that senior developers were twice as likely to put AI-generated code in production compared to junior developers and said the technology helped them work faster.
Vibe coding is also part of Spiers’ coding routine. He uses AI coding agents on multiple platforms for both front-end and back-end personal projects. He called the technology a mixed experience, but said it is good to help with prototype, build boiler plate or scaffold a test; It removes menial tasks so engineers can focus on building, shipping and scaling products.
It seems that the extra hours spent fighting through the mood, weeds will simply be a tolerated treasure by using innovation.
Elvis Kimara, a young engineer, is learning it now. He is just trained with a master’s degree in AI and is building an AI-run marketplace.
Like many coders, he said that vibe coding has made his job harder and has often found that vibe codes for a joyful experience.
“There’s no more dopamine from solving a problem by myself. Ai just finds it out,” he said. On one of his last jobs, he said senior developers did not look to help young coders so much-some did not understand new vibe-coding models, while other delegated mentoring tasks for that AI models.
But he said, “The advantages outweigh the disadvantages,” and he is prepared to pay the innovation tax.
“We don’t just write code; we guide AI systems, take accountability when things break down and act more like consultants for machines,” Kimara said of the new normal for which he prepares.
“Even as I grow to a senior role, I continue to use it,” he continued. “It’s been a real accelerator for me. I make sure I review every line of AI-generated code, so I learn even faster from it.”