👋 Welcome to CTO Teachings’s blog. While we are a recruiting company, we happen to provide a free blog for people just like you. It’s packed with hard-won lessons from CTO—who’s helped managers rise to Engineering Director and CTO roles.
There's some fascinating new data coming out about AI's impact on developer productivity, and it's not what you'd expect from the headlines.
On one hand, we're seeing impressive numbers: GitHub's research shows developers are 55% faster, generating up to 46% of code through AI. Microsoft's studies found a 26% increase in productivity. With the introduction of GitHub Copilot for businesses, developers reported coding up to 55% faster, with 85% feeling more confident in their code quality.
What's Actually Happening
A groundbreaking study from METR (Model Evaluation & Threat Research) reveals that experienced developers are actually 19% slower when using AI tools. Even more striking? These developers believed AI had made them 20% faster—that's a perception gap of nearly 40%.
This productivity paradox shows some critical truths about AI's impact on software development that we need to address.
The Great Divide: Good Developers vs. Everyone Else
What we're seeing is a clear split in the developer community. AI is taking really good developers and making them faster. But it's taking slower or less experienced developers and making them take longer to complete tasks.
This isn't just about speed—it's about understanding when AI is helping versus when it's leading you down the wrong path.
Why Hiring Just Got More Complicated
This shift makes hiring incredibly tricky. You can't just look for traditional coding skills anymore. You need developers who:
- Communicate exceptionally well through prompt engineering
- Know how to break down problems systematically
- Can quickly identify when AI is generating bad or flawed solutions
- Have the experience to catch subtle errors that newer developers would miss
Getting these developers is going to be more critical than ever for companies trying to stay competitive.
The Hidden Challenge: AI as a Junior Developer Who Never Learns
With 41% of all code now being generated by AI, developers have to become expert reviewers. They need to quickly identify when AI is making mistakes that could cause problems later—something I've experienced myself recently with Claude Code, particularly when it starts violating extensibility principles in the design.
The METR study highlighted that developers often find AI suggestions "frequently subtly wrong in ways that a newer dev wouldn't catch". This emphasizes why experienced developers are so crucial in the AI era—they're the ones who can spot these issues before they become technical debt.
Why Developers Think They're Faster When They're Not
Perhaps most concerning is the disconnect between perception and reality. While 75% of developers reported feeling more productive with AI tools, the data tells a different story: every 25% increase in AI adoption showed a 1.5% dip in delivery speed and a 7.2% drop in system stability.
This misperception isn't limited to individual developers. Economics experts predicted AI would improve productivity by 39%, and machine learning experts forecasted 38% gains—all dramatically overestimating the actual impact in complex, real-world scenarios.
What This Means for Your Team
The research points to several factors contributing to the slowdown:
- Time spent prompting and waiting: Developers spend significant time crafting prompts and waiting for AI responses
- Complex codebases: AI has trouble with big, established projects that have unwritten rules and team-specific ways of doing things
- Quality review overhead: Developers accepted less than 44% of AI generations, resulting in wasted time reviewing, testing, and modifying code that ultimately gets rejected
- Context limitations: AI lacks the undocumented, tacit knowledge that experienced developers rely on
The Path Forward: AI in Technical Interviews
This emphasizes the importance of actually using AI in code interviews—not to test if candidates can code without it, but to find developers who are really good at prompt engineering and coding. We need to identify those who can amplify their skills with these new tools while maintaining quality and architectural integrity.
The best developers of the future won't be those who can code the fastest or those who rely entirely on AI. They'll be the ones who know when to use AI, when to code themselves, and most importantly, when AI's suggestions will lead to problems down the road.
Conclusion: Embracing Reality Over Hype
The AI revolution in software development is real, but it's not the universal productivity boost we've been promised. It's creating a new class of highly productive developers while potentially slowing down others.
Companies that recognize this reality—and adapt their hiring, training, and development processes accordingly—will be the ones that truly benefit from AI-augmented development. Those that blindly chase the AI hype without understanding these nuances may find themselves with slower delivery, more bugs, and less maintainable code.
It's not about whether to use AI tools—it's about understanding who should use them, when they should be used, and how to build teams that can leverage them effectively while avoiding their pitfalls.
References
- GitHub Blog - Research: Quantifying GitHub Copilot's impact in the enterprise with Accenture - Study showing 55% faster coding and 85% confidence boost among developers
- METR - Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity - Randomized controlled trial revealing 19% slowdown despite perceived speedup among experienced developers
If you are ever looking for a job or expanding your team, connect with Shelly, Yuliia, or Harold on LinkedIn.
Thanks for reading,
CTO Teachings