Carla Lover spent 30 minutes sobing after she had to resume her vibe-coded project.
Rover has been in the industry for 15 years and mainly works as a web developer. She is currently building a startup with her son and creating custom machine learning models for the market.
She called for Vibe, who codes beautiful, endless cocktail napkins that allow you to sketch ideas forever. However, as these AI models can ruin your work in ways that are difficult to predict, she said it is “worst than a babysitter” to treat AI generated codes that she hopes to use in production as “worst than a babysitter.”
She was turning to AI coding to need speed at a startup, as well as the promise of AI tools.
“I took shortcuts and didn’t scan those files after automated reviews because I need to be quick and impressive,” she said. “When I did it manually, I found out I was so wrong. When I used third-party tools, I found more. And I learned my lesson.”
She and her son resumed their entire project. “I handed it over like an employee,” she said. “it’s not.”
Rover appears to be a lot of experienced programmers turning to AI for coding help. But such programmers also find themselves acting like AI Babysitters – rewrite the code that AI spits out to fact-check.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
A recent report by content distribution platform companies found that at least 95% of the nearly 800 developers surveyed spent extra time fixing AI generated code, and that such a validation load is falling the most on the shoulders of senior developers.
These experienced coders have discovered issues with AI generated code, ranging from hallucinations of package names to removing important information and security risks. Without checking, the AI code can leave the product a lot more buggy than what humans produce.
Working with AI-generated code has become a huge problem, and a new corporate coding job has been born, known as “Vibe Code Cleanup Specialist.”
TechCrunch spoke to experienced coders about what is considered the future of atmospheric coding, using AI-generated code. Though the ideas were varied, one thing remained certain. Technology still has a long way to go.
“Using a coding co-pilot is like giving a smart six-year-old a coffee pot and saying, ‘Take this to the dining room and pour coffee for your family,” Rover said.
Can they do that? probably. Will they fail? absolutely. And perhaps, if they fail, they’re not going to tell you. “I’m not smart enough to be a child,” she continued. “That means you can’t completely delegate (tasks) like that.”
“You’re absolutely right!”
Feridoon Malekzadeh compared the atmosphere coding with children.
He has worked in the industry for over 20 years and has played a variety of roles in product development, software and design. He has built his own startup and uses heavily a lovable platform that codes the atmosphere, he said. For fun, he also codes an app-like vibe that generates Gen Alpha Slang for boomers.
He likes how he can work alone on projects and save time and money, but he agrees that vibe coding is not like hiring an intern or junior coder. Instead, cosmosizing coding is similar to “hiring a stubborn, fragmented teenager to help do something.”
“You have to ask them 15 times to do something,” he said. “In the end, they do some of the things you asked, some of the things you didn’t ask, and they break a lot along the way.”
Malekzadeh spends around 50% of his time writing requirements, 10% to 20% of his time in coding atmospheric coding, and estimates 30% to 40% of his time in Vibe Fixing.
He also doesn’t think Vibe Coding is best for systems thinking. It is the process of looking at how complex problems affect the overall outcome. The AI-generated code is trying to solve more surface-level problems, he said.
“If you’re creating a feature that’s widely available in your product, then a great engineer will create it once and make it available wherever you need it,” says Malekzadeh. “Vibe coding creates five different ways, five different ways, if needed in five different places. It leads to a lot of confusion not only for the user but also for the model.”
Meanwhile, the rover discovers that when data conflicts with hardcoded ones, AI “runs to the wall.” “We can provide misleading advice, filter out essential important elements, or insert them into the thought pathways we are developing,” she said.
She also discovered that she produces results rather than admitting to making errors.
She shared another example with TechCrunch, where she questioned the outcome that the AI model initially gave her. The model began to provide detailed explanations, pretending to use the data it uploaded. Only when she called it the AI model confessed.
“It surprised me because it sounded like a toxic colleague,” she said.
In addition to this, there are security concerns.
Austin Spiers is Senior Director of Fastly’s Developer Enablement, and has been coding since the early 2000s.
He likes to build something faster than the “right” through his own experience, along with chatting with customers. This could create vulnerabilities in the kind of code that very newer programmers tend to create, he said.
“What happens often is that engineers need to check the code, fix the agent, and tell the agent that they’ve made a mistake,” Spiers told TechCrunch. “This pattern is why I saw the tactics on social media that are “You’re absolutely right.” ”
He mentions how, like Claude of Mankind, AI models tend to respond with “You are absolutely right” when they are called out for their mistakes.
Mike Arrowsmith, Chief Technology Officer of IT Management Software Company Ninjaone, has been involved in software engineering and security for nearly 20 years. He said that coding in the atmosphere creates a new generation of IT and security blind spots, especially those young startups.
“Vibe coding is crucial to often bypassing the rigorous review process that underlies traditional coding and catching vulnerabilities,” he told TechCrunch.
He said Ninjaone responded by encouraging “secure atmosphere coding” in which approved AI tools have access control, along with peer reviews and of course security scans, which are essential for approved AI tools.
New normal
Almost everyone we spoke to agree that AI-generated code and vibe coding platforms are useful in many situations, such as laughing at ideas, but they all agree that human reviews are essential before building a business.
“That cocktail napkin is not a business model,” Rover said. “You need to balance the simple things with insights.”
But because of all the lamentations of that error, coding of the atmosphere has changed the present and future of work.
Rover said vibe coding is extremely helpful in creating a better user interface. Malekzadeh simply said that despite the time he spends on correction codes, he’s doing more with AI coders than without them.
“Every technology has its own negativity that was invented simultaneously with technological advances,” Marekzadeh quoted French theorist Paul Bililio, who spoke with the ship about the invention of a shipwreck.
Pros far outweigh their weaknesses.
A rapidly researched study found that senior developers are twice as likely to put AI generated code into production compared to junior developers, and said the technology would help them work faster.
Vibe coding is also part of Spiers’ coding routine. He uses AI coding agents on several platforms for both front-end and back-end personal projects. He called the technology a complex experience, but said it was good for supporting prototyping, building boiler plates, or scaffolding for testing. Remove maniacal tasks so engineers can focus on building, shipping and scaling products.
The extra time spent through the atmosphere weeds would simply be a tax tolerated for using innovation.
Young engineer Elvis Kimara is learning it now. He just graduated with a Masters degree in AI and is building an AI-powered market.
Like many coders, he found coding to atmospherics made his work more difficult and often made the atmosphere a joyless experience.
“We don’t have dopamine anymore because we solve the problem ourselves. AI just understands it,” he said. In one of his final jobs, he said that the senior developers aren’t trying to help the younger coders much. Some people don’t understand the new atmosphere code model, while others have delegated mentorship tasks to the AI model.
But he says, “Pros are far outweighing their weaknesses,” and he is ready to pay the innovation tax.
“We’re not just writing code. We guide our AI systems, take accountability when things break, and act like a machine, like a consultant,” Kimara said of the new normal he’s preparing for.
“Even if I grow into a senior role, I will continue to use it,” he continued. “It was really an accelerator for me. I’d like to check every line of code generated in AI and then make sure I learn even faster.”
