Career Guidance

AI Might Be Slowing Down Some Employees’ Work, a Study Says



Sometimes innovative new tech is overhyped. Can AI really deliver on its promises of increased efficiency?


AI is often promoted as a game-changer for workplace productivity, especially when it comes to writing code. Tech CEOs love to tout the share of their company’s code that’s now generated by AI. While some critics question whether AI can truly “write” code in the creative, human sense, plenty of developers use these tools to spark ideas or get unstuck on tough problems.

But a new study from the nonprofit Metr suggests there’s a catch: for some highly skilled developers, using AI may actually slow them down.

Metr, an AI benchmarking organization based in Berkeley, California, ran a study to see if AI tools help “experienced open-source developers” complete tasks faster. Their motivation included understanding how close we really are to automating the work of high-level AI lab engineers.

Before starting a coding task, developers estimated AI would make them 24% faster. After completing the task, they revised that estimate slightly, thinking AI had given them a 20% boost. But the actual measured result was the opposite: developers who used AI took 19% longer to finish the task than those who didn’t.

This is a striking result that cuts through some of the hype. While expert developers expected AI to speed them up, the hard data showed it actually slowed them down.

As AI researcher Gary Marcus put it, “If this is a general, replicable finding,” it would be a serious blow to generative AI’s biggest promise: improving productivity. Instead of delivering time savings, AI might be adding new costs—like wasted time or distractions—without companies realizing it.

Of course, there’s important context. The study was run in early 2025, with tools that are constantly improving. The test group was also narrow: experienced developers working on complex codebases they knew well. That might explain why AI slowed them down: they already understood the code deeply, so the AI’s suggestions weren’t that helpful. Metr itself noted that AI might deliver bigger productivity gains in other settings, such as with smaller projects, less-experienced developers, or tasks with looser quality requirements.

The impact of AI on the coding profession remains hotly debated. A recent Microsoft study warned that some new graduates lean so heavily on AI that they don’t really learn the underlying principles, potentially a recipe for future failures. At the same time, OpenAI’s CFO Sarah Friar has talked about developing AI agents advanced enough to automatically “build an app for you,” suggesting an even more ambitious future for AI-generated code. Meanwhile, experienced developers like Salvatore Sanfilippo have publicly argued that humans still outperform AI at coding.

So what should you take away from this?

It’s a nuanced topic. For small businesses without dedicated developers, AI can help write code they’d otherwise have to outsource. But for companies with skilled engineers or other employees using AI tools, the lesson is clear: don’t assume these tools are automatically saving you time. It’s worth testing them carefully to see if they’re boosting productivity—or just adding new friction and slowing your team down.

Post a Comment