'AI stalls at 80%': Why AI isn’t coming for your software development job just yet, but is changing work

A viral Reddit post is sparking conversation in the developer community over whether AI code assistants like GitHub Copilot will eventually replace software engineers—or simply redefine their roles.

The original post, shared in a tech-focused subreddit, likens AI-assisted programming to the decades-long struggle of building self-driving cars. “I have historically been a real doomer on this front,” the poster wrote, “but more and more I think AI code assists are going to become self-driving cars in that they will get 95% of the way there and then get stuck at 95% for 15 years—and that last 5% really matters.”

The user predicted a shift in the role of developers: “Our jobs are just going to turn into reviewing small chunks of AI-written code all day and fixing them if needed.” While this could reduce the need for engineers in some organizations, they argued it could simultaneously create new roles due to “a bunch of non-technical people [trying] to write software with AI that will be buggy.”

One top-rated reply from a self-identified software engineer echoed these sentiments, confirming the tools’ usefulness while emphasizing their limitations. “AI/Copilot can do 80% of the tasks at work. The last 20% is much harder for it to get done,” they wrote. “Even with rewriting my prompts 3-5x I’m forced to take over.”

The commenter argued that the dream of replacing engineers wholesale is still far off—particularly because AI would need to interpret vague or complex instructions from non-engineers. “Getting AI to replace software engineers would require it to do the most difficult 20% of tasks consistently, correctly and as communicated by people who aren’t software engineers. Fat chance.”

Instead, AI tools are becoming "force multipliers," they said—making experienced developers faster and more productive. “When I run into a niche issue I used to comb through 10–20 StackOverflow threads… now with AI I can identify issues much faster.”

They also drew a parallel with another digital revolution: “Excel was a force amplifier for accountants, it did not replace accountants. We have more accounting jobs out there right now than at any point in the past.”

Perhaps most critically, they pointed to the massive backlog of technical debt across enterprise software stacks. “Major corporations have trillions of man hours worth of technical debt. If their employees become more efficient, they would be better served putting those more efficient employees to use than getting rid of them.”

The conversation reflects a growing consensus in the tech world: while AI is rapidly transforming the way software is written, reviewed, and debugged, it’s unlikely to fully replace the engineers behind the keyboard—at least not anytime soon. Instead, it’s ushering in a new era of augmented development, where speed and problem-solving skills may matter more than ever.