Every AI coding tool on the market promises you'll ship faster. A recent study says experienced developers actually got 19% slower when using them.
Read my lips: NOT beginners. NOT interns. We need EXPERIENCED developers.
Why This Hurts
This isn't some edge case buried in a footnote. The slowdown was attributed in part to time spent reviewing and debugging AI-generated output, though the study also found a significant perception-vs-reality gap where developers believed AI was helping even when it wasn't.
Senior developers didn't simply take what the computer gave them. They read it, doubted it, experimented with it, found the little errors, and corrected them. That process consumed a lot more time than it would have taken to write the software in the first place.
And let's be honest about something that nobody talks about. This directly contradicts the core marketing pitch of every single AI coding assistant on the market.
The Productivity "Feeling"
I believe that’s relatable for many of us. You prompt Copilot, it spits out 40 lines, and you think wow, that was fast. It feels productive. It’s a dopamine rush. The code is right there. Just ship it.
But feeling fast and being fast are two very different things. The study suggests that for experienced devs, the gap between those two is not just real — it's negative.
→ AI generates code quickly, so you feel like you're moving
→ You spend the next 20 minutes verifying and fixing what it wrote
→ Net result: slower than if you'd just typed it yourself
The junior dev might accept the output and move on. The senior dev knows better. And that knowing better is exactly what costs them time. 🎯
The Uncomfortable Implication
If senior engineers — the ones companies pay the most — get slower with these tools, what does that mean for the "AI will 10x your team" narrative?
This indicates that the story is not fully accurate. At worst, it's a convenient story told by companies selling seats.
I'm not saying AI coding tools are completely of no use. I actually use them myself. They are helpful for boilerplate codes, trying out new libraries, or even explaining your code when you're alone. However, saying that they save time for everyone? That's just a marketing gimmick with no real evidence to back it up.
What Senior Devs Actually Do
Here's what I think people overlook. Skilled developers do more than just put words on a screen. They choose a direction.
They are the final safeguard against incorrect requirements. They catch architectural mistakes before they become expensive. They hold context about the system that no AI has access to.
When you give that to someone and then add an AI that generates with confidence code that looks right but often is subtly wrong, which means that have a new job – an AI babysitter 😄– not a productivity gain, a tax.
→ The value of a senior dev is judgment, not typing speed
→ AI tools optimize for typing speed
→ The mismatch is predictable
So What Do We Do With This?
I don't think the answer is "throw away Copilot." The answer is to stop pretending the productivity gain is universal and automatic.
Let’s be realistic about where AI is useful and where it’s not. Judge by outcomes, not by what seems cool. And maybe stop using "we gave everyone Copilot" as a substitute for actually investing in developer experience.
The 19% number is a gut check. It's not the final word on AI-assisted development. But it's a data point that deserves more attention than it's getting, especially when billions of dollars are being spent on the opposite assumption.
If you're a senior dev, has AI tooling actually made you faster — or does it just feel that way? I'd genuinely love to know. 🤔
Top comments (0)