Gary Marcus on how AI has changed software engineers' job; says: We used to call it an engineer, now it is…

The job of a software engineer isn't what it used to be — and not in a good way, according to AI researcher Gary Marcus. In a conversation with Business Insider, Marcus said the rise of AI coding tools has quietly turned programmers into full-time code reviewers, and the shift is wearing them down. "Some people in coding, in particular, probably feel like constant pressure, and now they feel like what they're doing is debugging somebody else's code, instead of writing code," he said. "Debugging somebody else's code is not particularly fun."Marcus isn't alone in noticing this. Siddhant Khare, a developer who builds AI agent infrastructure, recently wrote about hitting a wall with AI fatigue. "We used to call it an engineer, now it is like a reviewer," Khare said. "Every time it feels like you are a judge at an assembly line and that assembly line is never-ending." Steve Yegge, a veteran engineer, thinks companies need to wake up to this reality fast. He told The Pragmatic Engineer newsletter that AI-assisted coding has a "vampiric effect" and recommended capping vibe-coding sessions at three hours a day."Do you let them work for three hours a day? The answer is yes, or your company's going to break," Yegge said.It's a striking warning from someone deep inside the industry — and it suggests the productivity gains of AI tools may come with a ceiling most companies haven't accounted for.Marcus, however, noted that AI fatigue won't land equally across every profession. For people in creative roles who don't have formal artistic training, AI might actually make work feel more rewarding. "If somebody needs to do some artistic work and they don't really have artistic talent, it might be fun to get the system to make them feel like they have a superpower," he said.That said, Marcus has long argued that current AI models are nowhere close to replacing humans. He's pointed to companies like Klarna, which once boasted its AI could match the output of 700 workers — only to later admit that quality suffered and human support was the smarter investment. For now, the software engineers stuck judging an endless assembly line of AI-generated code may be the canary in the coal mine.