Senior devs are getting slower with AI tools and don't even realize it. That's not my hot take. That's what a recent study found.
Experienced developers were 19% slower when using AI coding assistants. The kicker? Those same devs self-reported feeling about 20% faster. That's nearly a 40-percentage-point gap between perception and reality.
Take a moment to process that.
The Productivity Placebo
This is crazy because it completely changes the perspective on AI-tools. We were led to believe that these tools benefit experienced developers the most. Senior developers are aware of what quality code is, so they can better direct the AI. That was the idea.
But the data tells a different story. The developers who should benefit most are actually losing time. And they're so confident they're winning that they don't notice.
Why Seniors Might Be Losing Time
I have a hypothesis. Senior developers come equipped with solid mental models. They are able to maintain a complex system in their mind and write down solutions nearly at the same speed as their thoughts.
Now imagine adding an AI assistant to this mix. Suddenly you're reading suggestions, evaluating generated code, fixing subtle bugs the AI introduced, and context-switching between your plan and the AI's plan. That’s not speeding up. That’s slowing down in the guise of assistance.
→ Reading AI output takes time — time you wouldn't spend if you just wrote it yourself.
→ Reviewing generated code is harder than writing it — you have to verify someone else's logic against your mental model.
→ Subtle AI errors are the worst kind — they pass the vibe check but fail in production.
→ The "accept and fix" loop is deceptive — it feels fast but adds up.
Junior developers may actually find value in it as they are exposed to new patterns they might not know. For seniors, the AI is repeating what they already understand, and doing so at a pace that costs them time.
The Perception Gap Is the Real Problem
Not only is the 19% slowdown alarming, but also their level of confidence is even more frightening.
If you perceive 20% faster, you'll never doubt the tool. You'll never conduct the experiment. You'll advocate for AI assistants in every team meeting because your intuition tells you they are effective. That is a classic perception trap 🧠
This is important for teams to consider when making tooling decisions. If your most experienced engineers are unknowingly losing productivity and still championing a tool based on ‘vibes’ - that’s a problem.
Where AI Tools Actually Shine
I don't feel threatened by AI tools. In fact, I use them every day. But I've started being honest about where they help versus where they just make me feel productive.
→ Boilerplate and scaffolding — great. I don't need to type another Express route handler from memory.
→ Unfamiliar languages or APIs — solid. It's like having docs that talk back.
→ Complex logic in a codebase I know well — questionable. I'm usually faster on my own.
→ Debugging subtle issues — actively harmful sometimes. The AI confidently suggests wrong fixes and I waste time proving it wrong.
The problem is, the real benefits of it are not spread equally among all kinds of work. Where many teams get it wrong is treating it like an "always on" productivity tool.
The Takeaway
Measure, don't assume. If you're a senior dev who swears by Copilot or similar tools, try a week without them on a real task. Time it. Compare honestly. You might surprise yourself.
The best tools are the ones that make you faster and you can prove it. Otherwise, they simply provide a false sense of security.
**Have you ever measured your actual productivity with vs. without AI. For me, it doesn’t matter much, as I’m just having fun and learning new things.
Top comments (0)