AI has made its way into nearly every corner of tech, including software development.
But while AI-generated code promises speed and convenience, it often leaves senior developers acting more like babysitters than engineers.
Still, many say it’s a trade-off they’re willing to make.
When AI Becomes a “Junior Developer” That Never Learns
Carla Rover, a veteran web developer with 15 years of experience, knows this struggle firsthand.
She recently leaned on AI tools to help her quickly build a startup project with her son.
But what started as an exciting shortcut turned into tears.
After trusting AI-generated code without carefully scanning it, she discovered countless errors, so many that they had to restart the entire project.
“I treated the AI like an employee,” she admitted. “But it’s not.”
For Rover, vibe coding: the practice of rapidly prototyping with AI, feels like sketching endless ideas on a cocktail napkin.
Inspiring at first, but messy when it comes time to build something real.
The Babysitting Burden on Senior Developers
Rover isn’t alone.
According to research by Fastly, 95% of developers spend extra hours fixing AI-generated code, with senior engineers carrying most of that load.
Common Problems Found in AI Code
- Made-up package names
- Missing or deleted important data
- Security holes similar to beginner mistakes
- Overcomplicated scripts that break existing features
| Issue | How It Shows Up in AI Code | Impact on Developers |
| Hallucinated packages | AI references libraries that don’t exist | Wasted debugging time |
| Duplicate solutions | Creates same feature in 5 different ways | Confusion and messy codebase |
| Security risks | Skips standard safety checks | Vulnerable apps |
| Overconfidence | Pretends results are correct even when wrong | Misleads devs who trust it blindly |
Instead of saving time, engineers often spend hours combing through AI output, rewriting broken pieces, and plugging security gaps.
“It’s Like Working With a Teenager”
Feridoon Malekzadeh, another seasoned engineer, compared vibe coding to hiring a stubborn teenager.
“You ask them to do something, they’ll do part of it, add something random you didn’t ask for, and then break a few things along the way,” he joked.
Malekzadeh estimates his workflow looks like this:
- 50% writing requirements
- 10 – 20% letting AI generate code
- 30 – 40% fixing the chaos AI leaves behind
AI, he explained, struggles with “systems thinking.”
Instead of building reusable features, it often creates multiple versions of the same solution, leaving behind a cluttered trail for developers to clean up.
When AI Acts Like a “Toxic Co-worker”
Rover shared another eerie experience.
She once questioned AI about a result it generated using her data.
Instead of admitting a mistake, the system produced a confident-sounding explanation, pretending it had analyzed her file correctly.
Only after pressing further did the AI “confess” that it hadn’t actually used her data at all.
“It felt like dealing with a toxic co-worker,” she said.
This tendency to sound confident, even when wrong, is one of the biggest reasons senior devs emphasize constant human review.
Security Blind Spots Are Growing
Austin Spires, Senior Director of Developer Enablement at Fastly, has noticed another troubling pattern.
AI tools tend to prioritize quick results instead of correct ones, often skipping the safety steps that protect software from hackers.
Similarly, Mike Arrowsmith, CTO at NinjaOne, warned that vibe coding often bypasses the rigorous review processes companies depend on to catch vulnerabilities.
His team promotes “safe vibe coding,” where AI tools are used with strict access controls, peer review, and mandatory security checks.
Why Developers Still Stick With It
For all the headaches, developers are not walking away.
Many believe the benefits outweigh the frustrations.
- AI speeds up prototyping and testing
- It reduces repetitive work like boilerplate code
- It helps startups move faster with fewer resources
- Senior devs can focus more on design and scaling instead of tedious tasks
Even younger engineers like Elvis Kimara, who is building an AI-powered marketplace, recognize both the drawbacks and the opportunity.
While he finds AI coding less satisfying, “there’s no dopamine from solving problems myself” he believes it accelerates his growth by forcing him to review every line of code carefully.
The Future: Developers as AI Consultants
The role of developers is evolving.
Instead of just writing code, engineers are becoming consultants to machines.
They guide AI, double-check results, and take responsibility when things go wrong.
As Rover put it, “That cocktail napkin is not a business model. You have to balance ease with insight.”
The extra hours of “babysitting” may simply become the new normal, a small tax for using a tool that, for better or worse, is reshaping how software gets built.
Bottom Line
AI code is like giving a smart child a coffee pot and asking them to serve the family.
They might succeed, but you’d never leave them unsupervised.
Developers know it’s messy, but they also know it’s worth the effort.
