AI is writing nearly all the code at many companies now, but there are still domains where experts believe it’s coming up short.
Bjarne Stroustrup, the creator of C++, one of the most widely used programming languages in history, has a clear-eyed view of where AI-generated code falls flat. In his field — programming language design — he believes humans will remain essential, and his reasons go well beyond a general skepticism of AI.
“I think that in the field I’m mostly interested in (programming language design) — code will still be written by humans, and they will use abstraction,” Stroustrup said on a podcast. “The examples I’ve seen of attempts for AI to generate code in this domain have not been successful.”
The problems he identifies are specific and compounding. “They generate more bugs, more security holes. They have bloated code, which pessimizes — because you use more memory — and it’s hard to validate.”
This last point has a particularly serious implication: the very engineers capable of catching these problems are walking away. “The senior developers that would be needed to validate it — I’ve seen some of them starting to retire, because they don’t want to deal with the validation of something that changes every time you make a change in your prompts.”
Stroustrup also points to the regulatory dimension, which is especially demanding in the kinds of systems C++ is used to build — aerospace, automotive, medical devices, financial infrastructure. “There are regulatory bodies, there’s validation. You have to be able to validate what you changed when you make a change.”
And this is where AI-generated code creates a structural problem. “Even if you make a slight difference in a prompt, a lot of the code will change, and you have to check it again — all of the code that was generated. And there was more code generated than if it was written by humans.”
The contrast with human-written code is stark. “When a human makes a change, it’ll make a change that’s localized, and you can look for the effects of that localized change. If an AI writes it, you don’t actually know where it’s changed. You have to try and figure that out.”
Stroustrup’s concerns point to a real gap between AI’s growing capability in general software tasks and its fitness for precision-critical, expert-level domains. The gap isn’t just about code quality in the abstract — it’s about auditability, traceability, and accountability. In regulated industries, those properties aren’t optional.
The broader trend, however, is unmistakably moving in the other direction for most software work. Anthropic CEO Dario Amodei has predicted that AI could be writing 90% of all code within months, and former Google CEO Eric Schmidt has claimed the vast majority of programmers will be replaced by AI within a year. Even Y Combinator’s Tom Blomfield has said the software engineering job of today will not exist in five to ten years.
But Stroustrup’s critique suggests that bold timelines may be papering over genuine limitations. Speed and volume of code generation are not the same as correctness, security, or verifiability — and in domains where all three are non-negotiable, the calculus looks very different. The retirement of senior engineers who refuse to spend their careers validating unpredictable AI output may be the most underappreciated risk of all: when the people capable of catching the problems leave, the problems don’t disappear with them.