Every time a major AI coding tool is updated, social media erupts with discussions about "programmers losing their jobs." But what's the actual situation?
METR's Research Is Telling
AI safety research organization METR conducted a rigorous experiment: they had experienced developers complete programming tasks with and without AI assistance.
The result: Developers using AI tools were actually 19% slower.
Yes, you read that correctly. Experienced developers saw their efficiency decrease with AI assistance, rather than increase. The main reason is that the time spent reviewing and correcting AI output exceeded the time saved by AI.
However, this study has some limitations — the experimental design, task types, and specific tools used all affect the conclusions. It doesn't mean "AI coding tools are useless," but rather that in certain scenarios, the return on investment for AI tools isn't as high as imagined.
More Data Points
GitHub's own data shows that developers using Copilot did see efficiency gains in code completion scenarios, especially when writing boilerplate code and repetitive tasks.
But in areas that require architecture design, requirement understanding, and bug debugging — the truly brain-intensive parts — AI currently can't help much, and may even hinder progress by providing incorrect suggestions.
Real Impact
A more accurate assessment at this point is:
- AI won't "replace" programmers, but it will change what programmers do
- Time spent writing code will decrease, while time spent reviewing and validating will increase
- Demand for junior developers may decrease, while demand for senior developers who can leverage AI tools may increase
- "Knowing how to use AI tools" is becoming a basic skill, much like "knowing Git"
Neither panic nor blind optimism is necessary. This field changes too quickly, and the most reliable strategy is continuous learning, treating AI tools as tools rather than replacements.
Sources: METR research report, CocoLoop, GitHub annual survey