Skip to content
BSWEN
Blog

Why AI Coding Tools Don't Make You 10x Faster (According to Research)

Cowrie
Cowrie
Dev @ Bswen

The Problem

Every day, I see tweets and blog posts claiming AI coding tools make developers 10x or even 100x more productive. My manager asks why I’m not using Copilot or ChatGPT to code faster. Conference speakers promise AI will revolutionize software development.

But when I tried these tools, I didn’t feel 10x faster. I felt slower. Something didn’t add up.

What happened?

I started using AI coding assistants about a year ago. The promise was simple: write prompts, get code, ship faster. Here’s what I expected:

Manual coding: 4 hours
AI-assisted: 24 minutes (10x faster!)

Here’s what actually happened:

Manual coding approach:
┌──────────┐     ┌──────────┐     ┌──────────┐
│ Understand│ →   │ Write code│ →   │ Test     │
│ problem  │     │          │     │          │
└──────────┘     └──────────┘     └──────────┘
Time: 30 min     Time: 90 min    Time: 120 min
Total: 4 hours


AI-assisted approach:
┌──────────┐     ┌──────────┐     ┌──────────┐     ┌──────────┐
│ Understand│ →   │ Write    │ →   │ Fix AI   │ →   │ Test     │
│ problem  │     │ prompts  │     │ code     │     │          │
└──────────┘     └──────────┘     └──────────┘     └──────────┘
Time: 30 min     Time: 60 min    Time: 120 min   Time: 90 min
Total: 5 hours

I spent 60 minutes writing and refining prompts. The AI generated code that looked correct but had subtle bugs. I spent 2 hours debugging the AI-generated code, longer than it would have taken to write it myself.

So I tried a different approach. I used AI for smaller, well-defined tasks:

// Task: Write a function to validate email
// My prompt: "Create a function that validates email addresses using regex"


// AI generated this:
function validateEmail(email) {
  return /^[^\s@]+@[^\s@]+\.[^\s@]+$/.test(email);
}


// But I needed to spend time checking:
// - Does this handle edge cases? (What about + signs?)
// - Is the regex correct? (I had to test it)
// - Should I use a library instead? (More research)

Even for simple tasks, I spent time verifying the AI’s output. The time saved in typing was lost in verification.

What the research says

I found Anthropic’s research on AI-assisted coding, and it confirmed what I experienced. Their key findings:

  • No significant speedup in development
  • AI assistance impairs developers’ abilities
  • Prompt composition takes time comparable to manual coding
  • Claims of 10x-100x productivity boosts contradict research findings

The research showed that the time I spent crafting prompts wasn’t unique to me. It’s a fundamental part of working with AI tools.

Here’s the breakdown of where time goes:

Manual coding time breakdown:
├── Understanding problem: 12.5%
├── Writing code: 37.5%
├── Testing: 37.5%
└── Debugging: 12.5%


AI-assisted time breakdown:
├── Understanding problem: 12.5%
├── Writing prompts: 25%
├── Reviewing AI output: 25%
├── Fixing AI mistakes: 25%
└── Testing: 12.5%

The tasks don’t disappear. They just shift. Instead of writing code, I write prompts. Instead of debugging my own code, I debug AI-generated code.

Why AI impairs ability

This surprised me, but the research found that AI assistance actually impairs developers’ abilities. Here’s why I think this happens:

When I write code myself:

  • I understand every line
  • I know the trade-offs I made
  • I can explain why I chose this approach

When I use AI:

  • I don’t always understand the generated code
  • I don’t know what trade-offs were made
  • I can’t explain the approach as well

Over time, if I rely too much on AI, I lose the practice of solving problems myself. My skills atrophy.

When AI actually helps

After this experience, I found AI tools are useful for specific situations:

Good use cases:

  • Boilerplate code that follows clear patterns
  • Writing tests for existing code
  • Generating documentation
  • Explaining code I’m reviewing
  • Suggesting refactoring options

Poor use cases:

  • Complex business logic
  • Performance-critical code
  • Security-sensitive operations
  • Learning new concepts
  • Architecture decisions

The key is knowing when to use AI and when to code manually.

Summary

In this post, I explained why AI coding tools don’t deliver the promised 10x productivity gains. The key point is that prompt engineering and debugging AI-generated code takes as much time as writing code yourself, and relying too much on AI can actually hurt your development skills.

AI tools can help, but they’re not magic. Use them for the right tasks, and keep practicing your core coding skills.

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!

Comments

Tags: