It feels faster. But is it?
Experienced developers believed AI made them faster. When measured, they were slower. What that says about your own sense of productivity, and how to make the difference visible in your team.
- reading time
- 2 min
- sources
- 4 · 2 peer-reviewed
- published
- language
- Lees dit in het Nederlands
- In a controlled experiment, experienced developers were 19 percent slower with AI, while believing they were 20 percent faster.
- Other studies show big time savings. The difference comes down to the type of task and how experienced someone already is.
- Don't rely on gut feeling alone. Time a few recurring tasks, with and without AI.
Ask people whether AI makes them faster and most will say yes. It certainly feels that way: you type a question and a few seconds later there’s a complete answer. But does feeling faster mean being faster?
An experiment with a surprising result
In 2025, the research organisation METR put it to the test in a controlled experiment (Becker et al., 2025). Sixteen experienced open-source developers worked on 246 real tasks in projects they knew well. For each task, a coin flip decided whether they were allowed to use AI.
Beforehand, the developers expected AI to save them about a quarter of their time. Afterwards, they believed they had been around 20 percent faster. The measurements told a different story: with AI, tasks took them 19 percent longer on average.
Where did the time go? Among other things, into writing prompts, waiting for responses, and reviewing and fixing what the AI suggested. None of that feels like work. But it is.
Doesn’t AI make people faster, though?
It does, and other research shows it. On realistic writing tasks, people using ChatGPT finished about 40 percent faster and their work was rated higher (Noy & Zhang, 2023). Among customer support agents, issues resolved per hour went up by 15 percent on average, mostly among the less experienced (Brynjolfsson et al., 2025).
These findings don’t contradict each other. They show that the gain depends on two things:
- The type of task. AI is strong at some tasks and actually holds you back on others (Dell’Acqua et al., 2023).
- How skilled you already are. If you’re still learning a task, AI helps a lot. If you’re already very good at it, you can lose more time checking than you gain.
Why our gut feeling misleads us
Working with AI feels light. You don’t start from a blank page, and something comes back quickly. That creates a sense of progress, even when the end result takes longer. We easily forget to count the reviewing and correcting afterwards.
The METR study is a preprint with a small, specific group: experienced developers in their own complex projects, using early-2025 AI tools. So you can’t simply apply the 19 percent to your own team. The lesson about the gap between feeling and measuring applies much more widely, though.
How to make it visible in your team
- Pick three recurring tasks, say a quote, a report and a tricky client email.
- Track how long they take for a few weeks, with and without AI. Include the checking.
- Discuss the results together. Where does AI genuinely save time, and where does it only feel that way?
That takes you from “AI makes us faster” to “AI makes us faster at these tasks”, which is a far more useful insight.