Everyone says AI makes developers more productive.
Here's what a new study found. Experienced developers working on their own repositories take 19% longer when using AI tools compared to working without them.
That's right. Slower, not faster.
The Research Nobody Expected
METR ran a randomized controlled trial in 2025. They studied experienced open-source developers working on repositories they already knew well.
When developers used AI tools, they took 19% longer to complete tasks.
This contradicts everything we've been told. GitHub reports 55% faster task completion. Surveys show perceived 10-30% productivity gains. 92% of developers now use AI tools daily.
But when you measure experienced developers on familiar codebases? They slow down.
Why AI Makes You Slower
Think about what happens when you use AI to write code.
You write a prompt. You review the output. You test it. You find issues. You prompt again. You review again. You integrate it with your existing code. You refactor to make it fit your patterns.
That's a lot of context switching.
When you already know the codebase, you could have written the code faster yourself. You know the patterns. You know the edge cases. You know what will break.
AI gives you code that looks right but doesn't account for your specific context. So you spend time translating between what AI generates and what actually needs to exist in your codebase.
The Trust Problem
Here's the deeper issue. Only 33% of developers trust AI results. That's down from over 70% in 2023.
Just 3% highly trust AI-generated outputs.
You're using a tool you don't trust. That means every line of AI code requires careful review. Every suggestion needs validation.
When you write code yourself, you trust your own judgment. You move faster because you're not constantly second-guessing the output.
When AI Actually Helps
AI isn't useless. It's just not universally faster.
AI helps when you're learning a new framework. Or exploring unfamiliar codebases. Or writing boilerplate in languages you don't use often.
But when you're deep in your own codebase, working on complex problems you understand well? The cognitive overhead of managing AI often costs more than it saves.
The 55% speedup GitHub measures? That's for routine coding tasks. Not the hard problems experienced developers spend most of their time on.
The Cognitive Load Problem
Your brain has limited bandwidth. Every time you context switch between writing code and reviewing AI output, you lose focus.
The research on context switching is clear. It takes 23 minutes to fully recover from an interruption. That includes interrupting yourself to check what AI generated.
You think you're being productive because you're generating more code. But you're actually fragmenting your attention across multiple cognitive tasks.
Writing. Reviewing. Debugging. Integrating. Refactoring. That's five separate mental modes instead of one focused flow state.
Movement Resets Your Focus
When you're caught in this cycle of AI-assisted context switching, your brain gets fried.
You need something to interrupt the pattern. Not another AI tool. A complete reset.
Movement breaks force your brain out of the scattered state. Two minutes of squats or pushups shifts you from fragmented attention to clear focus.
When you come back, you can make a clean decision. Use AI for this task, or just write it yourself without the overhead.
The Movedoro Approach
I built Movedoro because I was drowning in AI-generated code that needed constant review.
Every 25 minutes, your screen locks. You move for two minutes. When you unlock, you're not just physically refreshed.
Your brain has reset. You can see clearly whether AI is actually helping or just creating more work.
AI makes you slower when you're too mentally fatigued to evaluate whether using it is the right call in the first place.
Movement breaks give you the clarity to use AI strategically instead of reflexively.
That's it.
