- by x32x01 ||
AI Coding Agents Write 30% More Code, But Are Developers Getting More Done?
AI coding agents are helping developers write more code, but that doesn't necessarily mean they're getting more work done.
A study by Harvard researchers, analyzing data from 718 companies that started using AI coding agents, found an interesting gap between code production and actual software delivery.
Here are the key findings:
Every additional change still needs to be reviewed, tested, and approved before it can reach production.
When AI generates code faster than developers can review it, the bottleneck simply moves to the next stage.
Code reviews become slower, testing workloads increase, and more pull requests wait for approval.
This explains why producing more code doesn't automatically translate into higher productivity.
It can also use one model to write code and another to review it.
The goal is to get performance closer to leading AI models while reducing costs.
But even with better AI models, the bigger challenge remains: making sure the resulting code is correct, maintainable, and ready for production.
Instead, develop the skills that help you evaluate and deliver reliable software:
As one Microsoft employee put it:
Completed work, not more code, is the output that matters.
And that's probably one of the most important lessons developers should take from the rise of AI coding agents.
💬 What about you? Since you started using AI coding tools, are you actually finishing projects faster, or are you just writing more code without getting more work done?
AI coding agents are helping developers write more code, but that doesn't necessarily mean they're getting more work done.
A study by Harvard researchers, analyzing data from 718 companies that started using AI coding agents, found an interesting gap between code production and actual software delivery.
Here are the key findings:
- 📈 30% more code: The amount of code written increased by around 30%.
- 🔄 20% more commits: Developers committed code more frequently.
- 🔀 23% more pull requests: The number of pull requests also increased.
- ⏳ 49% longer merge times: Pull requests took longer to get merged.
- 🚫 No meaningful increase in completed deployments: Despite producing more code, the number of tasks actually completed and deployed barely changed.
🚧 Why Is AI-Generated Code Creating a Bottleneck?
The problem is that writing code is only one part of software development.Every additional change still needs to be reviewed, tested, and approved before it can reach production.
When AI generates code faster than developers can review it, the bottleneck simply moves to the next stage.
Code reviews become slower, testing workloads increase, and more pull requests wait for approval.
This explains why producing more code doesn't automatically translate into higher productivity.
🤖 What's Changing in AI Coding Tools?
At GitHub Universe, GitHub introduced Copilot Fusion, a system designed to select the most suitable AI model for a given task.It can also use one model to write code and another to review it.
The goal is to get performance closer to leading AI models while reducing costs.
But even with better AI models, the bigger challenge remains: making sure the resulting code is correct, maintainable, and ready for production.
💡 What Should Developers Focus on Now?
If you're learning programming today, don't focus only on writing code faster. AI is already getting very good at that.Instead, develop the skills that help you evaluate and deliver reliable software:
- 🔍 Code review: Learn to read code written by other developers or AI and identify bugs, security issues, and poor design decisions.
- 🧪 Testing: Write tests that verify whether the code actually works as expected.
- 🔀 Smaller pull requests: Break changes into smaller pieces that are easier and faster to review.
- ⚙️ CI/CD: Use automated checks to catch problems before code reaches human reviewers.
🎯 The Bottom Line
AI is making code cheaper and faster to produce, but reviewing, testing, and trusting that code are becoming more important.As one Microsoft employee put it:
Completed work, not more code, is the output that matters.
And that's probably one of the most important lessons developers should take from the rise of AI coding agents.
💬 What about you? Since you started using AI coding tools, are you actually finishing projects faster, or are you just writing more code without getting more work done?