We conduct a randomized controlled trial to understand how early-2025 AI tools affect the productivity of experienced open-source developers working on their own repositories. Surprisingly, we find that when developers use AI tools, they take 19% longer than without—AI makes them slower.
One area that it really helps for me is executive function. There’s a lot of things that I’m capable of doing, even quickly, but I’ve got this mental block that just makes me not want to do it. But if I can just write out some instructions and have a system do the rest, it’s much easier to get going.
I’ll still think through the problem and approach using AI to help coding at a “I need a function that will take this data in this format and do x, y, z to it and return that data in that format” level rather than something like a higher level description of my final goal.
Is it faster than what I could do if I focused on programming and get on a good roll? I dunno, it might still be. Is it faster than me actually trying to program something in the reality that I often exist in? Fuck yeah.
And even debugging and testing go way smoother. For one thing, I don’t have to deal with stupid typo bugs anymore. And for the bugs that still make it in, AI has been great at taking an idea of how to examine the data that would be a pain to implement and just doing it for me, especially if it involves some obscure API or language features since I don’t have to spend time finding its existence and then learning it (if it’s a one off, I won’t likely retain anything other than the existence part, which I can still get from looking at the AI generated code).
So it’s pretty great for programmers who know their shit but ADHD gets in the way of using it in a timely manner. It’s better than an intern, which is good for me but sucks for those who need to learn. There’s a good chance AI’s semi-competence if hand held by an expert is going to lead to a big lack of talent as those experts age out. Though with how quickly it’s improving, it might not even need the hand holding by then. I’m not sure which possibility is scarier.
One area that it really helps for me is executive function. There’s a lot of things that I’m capable of doing, even quickly, but I’ve got this mental block that just makes me not want to do it. But if I can just write out some instructions and have a system do the rest, it’s much easier to get going.
I’ll still think through the problem and approach using AI to help coding at a “I need a function that will take this data in this format and do x, y, z to it and return that data in that format” level rather than something like a higher level description of my final goal.
Is it faster than what I could do if I focused on programming and get on a good roll? I dunno, it might still be. Is it faster than me actually trying to program something in the reality that I often exist in? Fuck yeah.
And even debugging and testing go way smoother. For one thing, I don’t have to deal with stupid typo bugs anymore. And for the bugs that still make it in, AI has been great at taking an idea of how to examine the data that would be a pain to implement and just doing it for me, especially if it involves some obscure API or language features since I don’t have to spend time finding its existence and then learning it (if it’s a one off, I won’t likely retain anything other than the existence part, which I can still get from looking at the AI generated code).
So it’s pretty great for programmers who know their shit but ADHD gets in the way of using it in a timely manner. It’s better than an intern, which is good for me but sucks for those who need to learn. There’s a good chance AI’s semi-competence if hand held by an expert is going to lead to a big lack of talent as those experts age out. Though with how quickly it’s improving, it might not even need the hand holding by then. I’m not sure which possibility is scarier.