AI Code Is a Bug-Filled Mess

The adoption rate of AI tools has skyrocketed in the programming world, enabling coders to generate vast amounts of code with simple text prompts.

Earlier this year, Google found that 90 percent of software developers across the industry are using AI tools on the job, up from a mere 14 percent last year.

But all that convenience has come with some glaring drawbacks. The tools have repeatedly been found to be unreliable and inaccurate, which can lead to mistakes falling through the cracks and even forcing some programmers to put in long hours to identify and correct them.

Adding to the reality check, a new report by AI software company CodeRabbit found that code generated by an AI was far more error-prone than the human-written stuff — and by a significant margin. Across the 470 pull requests the company analyzed, AI code produced an average 10.83 issues per request, while human-authored code produced just 6.45.

In other words, AI code produced 1.7 times more issues than human code, once again highlighting major weaknesses plaguing generative AI tools.

“The results?” CodeRabbit concluded in its report. “Clear, measurable, and consistent with what many developers have been feeling intuitively: AI accelerates output, but it also amplifies certain categories of mistakes.”

Worse yet, the company found that AI-generated code produced a higher rate of “critical” and “major” issues, in a “meaningful rise in substantive concerns that demand reviewer attention.”

AI code was also most likely to contain errors related to logic and correctness. However, the biggest weakness CodeRabbit found was in code quality and readability, which are issues that can “slow teams down and compound into long-term technical debt.”

Then there are serious cybersecurity concerns, with generated code introducing issues related to improper password handling that could lead to protected information being exposed, among other insecure practices.

On the upside, CodeRabbit found that AI code was adept at keeping spelling errors at a minimum. Humans were twice as likely to introduce misspellings.

It’s far from the first time we’ve heard of flaws plaguing AI-generated code. In a September report, management consultants Bain & Company concluded that despite being “one of the first areas to deploy generative AI,” the “savings have been unremarkable” in programming and “results that haven’t lived up to the hype.”

Security firm Apiiro also found in its research that developers who used AI produce ten times more security problems than their counterparts who don’t use the tech.

As a result, programmers are forced to pick over the generated code to ensure no glaring issues fall through the cracks. According to a July study from the nonprofit Model Evaluation and Threat Research, programmers were actively being slowed down by AI assistance tools compared to when they made do without them.

In short, while companies made sky-high promises about the tech making programmers’ lives much easier, reality looks far more nuanced. CodeRabbit’s report suggests a shift in the kinds of tasks human developers could soon be required to do — like solving issues being introduced by error-prone AI coding tools.

“These findings reinforce what many engineering teams have sensed throughout 2025,” said CodeRabbit AI Director David Loker in a statement. “AI coding tools dramatically increase output, but they also introduce predictable, measurable weaknesses that organizations must actively mitigate.”

More on AI programming: AI Coding Is Massively Overhyped, Report Finds