Are AI Coding Tools Slowing Down the Pros

A recent study by AI research non-profit METR, challenges the prevailing belief that AI coding tools universally boost developer productivity. 
 
Conducted with experienced open-source developers in familiar codebases, the study found that AI assistance actually slowed performance—developers took 19% longer to complete tasks when using AI.

This outcome sharply contrasts with participants’ expectations. 

Before the trial, they anticipated a 24% time reduction. Even after experiencing delays, many still perceived the tools as helpful—revealing a gap between perceived and actual efficiency.

The root cause of the slowdown lies in the complexity of mature codebases. 

 
While AI-generated code was often "directionally correct," it lacked the contextual precision required for high-quality, tightly integrated systems. 
 
As a result, developers spent less time coding and more time crafting prompts, waiting for responses, and scrutinizing AI output for subtle flaws.
The study underscores a critical limitation: current AI tools may be ill-suited for seasoned developers deeply familiar with their work environment. 
 
For them, the cognitive overhead of interpreting and correcting AI suggestions often outweighs the benefits.

However, for junior developers or those navigating unfamiliar code, AI may still offer significant advantages. 
 
The findings point to a need for more context-aware AI solutions tailored to expert workflows.

Ms. S Mohini Ratna - Editor VARINDIA